Prosecution Insights
Last updated: November 20, 2026
Application No. 18/643,548

TECHNIQUES FOR IMPROVED LENGTH CONSTRAINT COMPLIANCE

Non-Final OA §101§102§103
Filed
Apr 23, 2024
Examiner
STANLEY, JEREMY L
Art Unit
2127
Tech Center
2100 — Computer Architecture & Software
Assignee
Google LLC
OA Round
1 (Non-Final)
50%
Grant Probability
Moderate
1-2
OA Rounds
8m
Est. Remaining
88%
With Interview

Examiner Intelligence

Grants 50% of resolved cases
50%
Career Allowance Rate
150 granted / 299 resolved
-4.8% vs TC avg
Strong +38% interview lift
Without
With
+38.1%
Interview Lift
resolved cases with interview
Typical timeline
3y 3m
Avg Prosecution
20 currently pending
Career history
329
Total Applications
across all art units

Statute-Specific Performance

§101
11.0%
-29.0% vs TC avg
§103
53.9%
+13.9% vs TC avg
§102
13.9%
-26.1% vs TC avg
§112
16.7%
-23.3% vs TC avg
Black line = Tech Center average estimate • Based on career data from 299 resolved cases

Office Action

§101 §102 §103
DETAILED ACTION Notice of Pre-AIA or AIA Status The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA . This action is responsive to the Application filed on April 23, 2024. Claims 1-20 are pending in the case. Claims 1 and 14 are the independent claims. This action is non-final. Claim Rejections – 35 USC § 101 35 U.S.C. 101 reads as follows: Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title. Claims 1-20 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea (mental steps) without significantly more. This judicial exception is not integrated into a practical application because any additional elements amount to implementing the abstract idea on a generic computer. The claim(s) does/do not include additional elements that are sufficient to amount to significantly more than the judicial exception. Regarding independent claims 1, 14, and 20, and relying on the evaluation flowchart in MPEP 2106: Step 1 (Is the claim to a process, machine, manufacture, or composition of matter?): Yes. Claim 1 is a method (process). Claim 14 is a method (process). Claim 20 is a non-transitory computer-readable medium (manufacture). Step 2a Prong One (Does the claim recite an abstract idea?): Yes. Claims 1 and 20 recite: analyzing the input prompt to identify one or more length constraints intended to be imposed on the generative model response (a mental process of determination, such as a human mentally analyzing a prompt to identify length constraints); evaluating the generative model response for compliance with one or more of the length constraints (a mental process of determination, such as a human mentally determining whether the response complies with the length constraints); based on a determination that the generative model response fails to comply with one or more of the length constraints, modifying the generative model training example to generate a synthetic generative model training example for which the one or more length constraints are satisfied (a mental process of determination, including one performed with a physical aid such as pen and paper, such as a human mentally determining that the response fails to comply with the length constraint and further determining how to modify the training example such that the length constraint is satisfied).. Under the broadest reasonable interpretation, these steps may be performed mentally, using mental observation and mental determination, including by a human using a physical aid such as pen and paper, including a human mentally performing observations and mentally performing mathematical calculations, and therefore correspond to the Mental Processes grouping. Step 2a Prong Two (Does the claim recite additional elements that integrate the judicial exception into a practical application?): No. Claims 1 and 20 additionally recite: implemented using one or more processors (in claim 1) (mere instructions to apply the exception using generic computer components as discussed in MPEP 2106.05(f)); non-transitory computer-readable medium comprising instructions that, in response to execution by one or more processors, cause the one or more processors to (in claim 20) (mere instructions to apply the exception using generic computer components as discussed in MPEP 2106.05(f)); retrieving a candidate generative model training example, wherein the candidate generative model training example includes an input prompt and a generative model response that was generated by processing the input prompt using one or more generative models (insignificant extra-solution activity as discussed in MPEP 2106.05(g), with respect to the “retrieving” and “generated by processing” limitations, and a field of use and technological environment as discussed in MPEP 2106.05(h), with respect to the limitations regarding the contents of the training example); training one or more of the generative models using the synthetic generative model training example (mere instructions to apply the exception using generic computer components as discussed in MPEP 2106.05(f)). To the extent that a human cannot mentally perform the steps of modifying the generative model training example to generate a synthetic generative model training example entirely within the human mind or using a physical aid such as pen and paper, it is noted that these limitations appear to recite insignificant extra-solution activity as discussed in MPEP 2106.05(g). Therefore, in view of the considerations set forth in MPEP 2106.04(d), 2106.05(a)-(c) and (e)-(h), the additional elements as disclosed above alone or in combination do not integrate the judicial exception into a practical application as they are mere insignificant extra solution activity, combined with implementing the abstract idea using generic computer components. Step 2b (Does the claim recite additional elements that amount to significantly more than the judicial exception): No. Relying on the same analysis as Step 2a Prong Two (see MPEP 2106.05.I.A: Limitations that the courts have found not to be enough to qualify as “significantly more” when recited in a claim with a judicial exception include:…Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, e.g., a limitation indicating that a particular function such as creating and maintaining electronic records is performed by a computer, as discussed in Alice Corp., 573 U.S. at 225-26, 110 USPQ2d at 1984 (see MPEP 2106.05(f));…Simply appending well-understood, routine, conventional activities previously known to the industry, specified at a high level of generality, to the judicial exception...; Adding insignificant extra-solution activity to the judicial exception, as discussed in MPEP 2106.05(g);…)), claims 1 and 20 do not recite any additional elements that amount to significantly more than the abstract idea. As discussed above, Claims 1 and 20 additionally recite: implemented using one or more processors (in claim 1) (mere instructions to apply the exception using generic computer components as discussed in MPEP 2106.05(f)); non-transitory computer-readable medium comprising instructions that, in response to execution by one or more processors, cause the one or more processors to (in claim 20) (mere instructions to apply the exception using generic computer components as discussed in MPEP 2106.05(f)); retrieving a candidate generative model training example, wherein the candidate generative model training example includes an input prompt and a generative model response that was generated by processing the input prompt using one or more generative models (insignificant extra-solution activity such as necessary data gathering and outputting as discussed in MPEP 2106.05(g), with respect to the “retrieving” and “generated by processing” limitations, which can further be re-evaluated in Step 2B as a well understood, routine, and conventional activity MPEP 2106.05(d) of receiving or transmitting data over a network and/or storing and retrieving information in memory, and a field of use and technological environment as discussed in MPEP 2106.05(h), with respect to the limitations regarding the contents of the training example); training one or more of the generative models using the synthetic generative model training example (mere instructions to apply the exception using generic computer components as discussed in MPEP 2106.05(f)). To the extent that a human cannot mentally perform the steps of modifying the generative model training example to generate a synthetic generative model training example entirely within the human mind or using a physical aid such as pen and paper, it is noted that these limitations appear to recite insignificant extra-solution activity as discussed in MPEP 2106.05(g), such as necessary data gathering and outputting, and which can further be re-evaluated in Step 2B as a well understood, routine, and conventional activity MPEP 2106.05(d) of receiving or transmitting data over a network and/or storing and retrieving information in memory. The additional elements as discussed above, in combination with the abstract idea, are not sufficient to amount to significantly more than the judicial exception as they are well, understood, routine and conventional activity as disclosed in combination with generic computer functions and components used to implement the abstract idea. Claim 14 recites: analyzing the input prompt to identify one or more length constraints intended to be imposed on the generative model responses (a mental process of determination, such as a human mentally analyzing a prompt to identify length constraints); evaluating the candidate generative model responses for compliance with one or more of the length constraints (a mental process of determination, such as a human mentally determining whether the response complies with the length constraints); based on the evaluating, selecting, for inclusion in a generative model training example, the input prompt and the candidate generative model response that most closely complies with one or more of the length constraints (a mental process of determination, such as a human mentally selecting an input prompt and response that most closely comply with the constraint). Under the broadest reasonable interpretation, these steps may be performed mentally, using mental observation and mental determination, including by a human using a physical aid such as pen and paper, including a human mentally performing observations and mentally performing mathematical calculations, and therefore correspond to the Mental Processes grouping. Step 2a Prong Two (Does the claim recite additional elements that integrate the judicial exception into a practical application?): No. Claim 14 additionally recites: implemented using one or more processors (mere instructions to apply the exception using generic computer components as discussed in MPEP 2106.05(f)); retrieving a generative model interaction set that includes an input prompt and two or more candidate generative model responses that were generated by processing the input prompt using one or more generative models (insignificant extra-solution activity as discussed in MPEP 2106.05(g), with respect to the “retrieving” and “generated by processing” limitations, and a field of use and technological environment as discussed in MPEP 2106.05(h), with respect to the limitations regarding the contents of the interaction set); training one or more of the generative models using the generative model training example (mere instructions to apply the exception using generic computer components as discussed in MPEP 2106.05(f)). Therefore, in view of the considerations set forth in MPEP 2106.04(d), 2106.05(a)-(c) and (e)-(h), the additional elements as disclosed above alone or in combination do not integrate the judicial exception into a practical application as they are mere insignificant extra solution activity, combined with implementing the abstract idea using generic computer components. Step 2b (Does the claim recite additional elements that amount to significantly more than the judicial exception): No. Relying on the same analysis as Step 2a Prong Two (see MPEP 2106.05.I.A: Limitations that the courts have found not to be enough to qualify as “significantly more” when recited in a claim with a judicial exception include:…Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, e.g., a limitation indicating that a particular function such as creating and maintaining electronic records is performed by a computer, as discussed in Alice Corp., 573 U.S. at 225-26, 110 USPQ2d at 1984 (see MPEP 2106.05(f));…Simply appending well-understood, routine, conventional activities previously known to the industry, specified at a high level of generality, to the judicial exception...; Adding insignificant extra-solution activity to the judicial exception, as discussed in MPEP 2106.05(g);…)), claim 14 does not recite any additional elements that amount to significantly more than the abstract idea. As discussed above, Claim 14 additionally recites: implemented using one or more processors (mere instructions to apply the exception using generic computer components as discussed in MPEP 2106.05(f)); retrieving a generative model interaction set that includes an input prompt and two or more candidate generative model responses that were generated by processing the input prompt using one or more generative models (insignificant extra-solution activity such as necessary data gathering and outputting as discussed in MPEP 2106.05(g), with respect to the “retrieving” and “generated by processing” limitations, which can further be re-evaluated in Step 2B as a well understood, routine, and conventional activity MPEP 2106.05(d) of receiving or transmitting data over a network and/or storing and retrieving information in memory, and a field of use and technological environment as discussed in MPEP 2106.05(h), with respect to the limitations regarding the contents of the interaction set); training one or more of the generative models using the generative model training example (mere instructions to apply the exception using generic computer components as discussed in MPEP 2106.05(f)). The additional elements as discussed above, in combination with the abstract idea, are not sufficient to amount to significantly more than the judicial exception as they are well, understood, routine and conventional activity as disclosed in combination with generic computer functions and components used to implement the abstract idea. Regarding dependent claim 2: Step 2a Prong One: incorporates the rejection of claim 1. The claim further recites wherein the modifying comprises altering one or more of the length constraints of the input prompt to match one or more length features of the generative model response (a mental process of determination, including one performed using a physical aid such as pen and paper, such as a human mentally determining that the modification should include altering the length constraint in the input prompt to match the response). Step 2a Prong Two: the claim does not recite any other limitations in addition to the abstract idea discussed above. Step 2b: the claim does not recite any other limitations in addition to the abstract idea discussed above. Regarding dependent claim 3: Step 2a Prong One: incorporates the rejection of claim 2. The claim further recites wherein the match comprises a fuzzy match (further describing the mental process in which the mentally determined match between the length constraint and the response is an inexact, or fuzzy, match). Step 2a Prong Two: the claim does not recite any other limitations in addition to the abstract idea discussed above. Step 2b: the claim does not recite any other limitations in addition to the abstract idea discussed above. Regarding dependent claim 4: Step 2a Prong One: incorporates the rejection of claim 1. The claim further recites wherein the modifying comprises altering the generative model response to match one or more of the length constraints (a mental process of determination, including one performed using a physical aid such as pen and paper, such as a human mentally determining that the modification should include altering the response to match the length constraint). Step 2a Prong Two: the claim does not recite any other limitations in addition to the abstract idea discussed above. Step 2b: the claim does not recite any other limitations in addition to the abstract idea discussed above. Regarding dependent claim 5: Step 2a Prong One: incorporates the rejection of claim 4. The claim further recites wherein the match comprises a fuzzy match (further describing the mental process in which the mentally determined match between the length constraint and the response is an inexact, or fuzzy, match). Step 2a Prong Two: the claim does not recite any other limitations in addition to the abstract idea discussed above. Step 2b: the claim does not recite any other limitations in addition to the abstract idea discussed above. Regarding dependent claim 6: Step 2a Prong One: incorporates the rejection of claim 4. Step 2a Prong Two: the claim additionally recites wherein the altering comprises processing the input prompt using one or more of the generative models to generate a new generative model response (mere instructions to apply the exception using generic computer components as discussed in MPEP 2106.05(f) and a field of use and technological environment as discussed in MPEP 2106.05(h)). Step 2b: the claim additionally recites wherein the altering comprises processing the input prompt using one or more of the generative models to generate a new generative model response (mere instructions to apply the exception using generic computer components as discussed in MPEP 2106.05(f) and a field of use and technological environment as discussed in MPEP 2106.05(h)). Regarding dependent claim 7: Step 2a Prong One: incorporates the rejection of claim 1. The claim further recites wherein analyzing the input prompt comprises parsing the input prompt to detect one or more linguistic concepts and numeric modifiers of the one or more linguistic concepts (a mental process of evaluation, such as a human mentally determining linguistic concepts and numeric modifiers in the prompt). Step 2a Prong Two: the claim does not recite any other limitations in addition to the abstract idea discussed above. Step 2b: the claim does not recite any other limitations in addition to the abstract idea discussed above. Regarding dependent claim 8: Step 2a Prong One: incorporates the rejection of claim 7. Step 2a Prong Two: the claim additionally recites wherein the one or more linguistic concepts include a sentence, and the one or more numeric modifiers includes a number of requested sentences (a field of use and technological environment as discussed in MPEP 2106.05(h)). Step 2b: the claim additionally recites wherein the one or more linguistic concepts include a sentence, and the one or more numeric modifiers includes a number of requested sentences (a field of use and technological environment as discussed in MPEP 2106.05(h)). Regarding dependent claim 9: Step 2a Prong One: incorporates the rejection of claim 7. Step 2a Prong Two: the claim additionally recites wherein the one or more linguistic concepts include a word, and the one or more numeric modifiers includes a number of requested words (a field of use and technological environment as discussed in MPEP 2106.05(h)). Step 2b: the claim additionally recites wherein the one or more linguistic concepts include a word, and the one or more numeric modifiers includes a number of requested wrods (a field of use and technological environment as discussed in MPEP 2106.05(h)). Regarding dependent claim 10: Step 2a Prong One: incorporates the rejection of claim 7. Step 2a Prong Two: the claim additionally recites wherein the one or more linguistic concepts include a paragraph, and the one or more numeric modifiers includes a number of requested paragraphs (a field of use and technological environment as discussed in MPEP 2106.05(h)). Step 2b: the claim additionally recites wherein the one or more linguistic concepts include a paragraph, and the one or more numeric modifiers includes a number of requested paragraphs (a field of use and technological environment as discussed in MPEP 2106.05(h)). Regarding dependent claim 11: Step 2a Prong One: incorporates the rejection of claim 1. The claim further recites wherein analyzing the input prompt comprises performing natural language processing (NLP) on the input prompt to identify an intent behind the prompt and one or more parameters of the intent, wherein one or more parameters of the intent include one or more of the length constraints (a mental process of determination, such as a human mentally analyzing/processing the natural language of the input prompt to determine the intent of the prompt, including an associated length constraint parameter). Step 2a Prong Two: the claim does not recite any other limitations in addition to the abstract idea discussed above. Step 2b: the claim does not recite any other limitations in addition to the abstract idea discussed above. Regarding dependent claim 12: Step 2a Prong One: incorporates the rejection of claim 1. The claim further recites wherein analyzing the input prompt comprises: assembling data indicative of the input prompt into an auxiliary input prompt; assembling, into the auxiliary input prompt, data indicative of a natural language request to identify the one or more length constraints in the input prompt (a mental process of determination, such as a human mentally analyzing/processing the input prompt to determine content/data of the input prompt to be used for forming an auxiliary input prompt, including the natural language request to identify length constraints in the prompt). Step 2a Prong Two: the claim further recites processing the auxiliary input prompt using one or more of the generative models to generate auxiliary generative model output indicative of one or more of the length constraints (mere instructions to apply the exception using generic computer components as discussed in MPEP 2106.05(f)). Step 2b: the claim further recites processing the auxiliary input prompt using one or more of the generative models to generate auxiliary generative model output indicative of one or more of the length constraints (mere instructions to apply the exception using generic computer components as discussed in MPEP 2106.05(f)). To the extent that a human cannot mentally perform the steps of assembling data…into an auxiliary input prompt; assembling, into the auxiliary input prompt, data… entirely within the human mind or using a physical aid such as pen and paper, it is noted that these limitations appear to recite insignificant extra-solution activity as discussed in MPEP 2106.05(g), such as necessary data gathering and outputting, and also as well-understood, routine, conventional activity as discussed in MPEP 2106.05(d), such as receiving or transmitting data over a network, storing and retrieving information in memory, etc. Regarding dependent claim 13: Step 2a Prong One: incorporates the rejection of claim 1. Step 2a Prong Two: the claim further recites wherein one or more of the generative models comprises a large language model (LLM) (mere instructions to apply the exception using generic computer components as discussed in MPEP 2106.05(f)). Step 2b: the claim further recites wherein one or more of the generative models comprises a large language model (LLM) (mere instructions to apply the exception using generic computer components as discussed in MPEP 2106.05(f)). Regarding dependent claim 15: Step 2a Prong One: incorporates the rejection of claim 14. Step 2a Prong Two: the claim further recites wherein the two or more candidate generative model responses comprise first and second candidate generative model responses, generated using the same generative model, which are different from each other (a field of use and technological environment as discussed in MPEP 2106.05(h)). Step 2b: the claim further recites wherein the two or more candidate generative model responses comprise first and second candidate generative model responses, generated using the same generative model, which are different from each other (a field of use and technological environment as discussed in MPEP 2106.05(h)). Regarding dependent claim 16: Step 2a Prong One: incorporates the rejection of claim 15. Step 2a Prong Two: the claim further recites wherein the first and second candidate generative models responses differ from each other due to a temperature parameter used in association with the generative model (a field of use and technological environment as discussed in MPEP 2106.05(h)). Step 2b: the claim further recites wherein the first and second candidate generative models responses differ from each other due to a temperature parameter used in association with the generative model (a field of use and technological environment as discussed in MPEP 2106.05(h)). Regarding dependent claim 17: Step 2a Prong One: incorporates the rejection of claim 14. Step 2a Prong Two: the claim further recites wherein the two or more candidate generative model responses comprise a first candidate generative model response generated using a first generative model and a second candidate generative model response generated using a second generative model that is different from the first generative model (a field of use and technological environment as discussed in MPEP 2106.05(h)). Step 2b: the claim further recites wherein the two or more candidate generative model responses comprise a first candidate generative model response generated using a first generative model and a second candidate generative model response generated using a second generative model that is different from the first generative model (a field of use and technological environment as discussed in MPEP 2106.05(h)). Regarding dependent claim 18: Step 2a Prong One: incorporates the rejection of claim 17. Step 2a Prong Two: the claim further recites wherein the second generative model comprises fewer parameters than the first generative model (a field of use and technological environment as discussed in MPEP 2106.05(h)). Step 2b: the claim further recites wherein the second generative model comprises fewer parameters than the first generative model (a field of use and technological environment as discussed in MPEP 2106.05(h)). Regarding dependent claim 19: Step 2a Prong One: incorporates the rejection of claim 14. Step 2a Prong Two: the claim further recites wherein the selecting is performed using a trained reward function (mere instructions to apply the exception using generic computer components as discussed in MPEP 2106.05(f) and a field of use and technological environment as discussed in MPEP 2106.05(h)). Step 2b: the claim further recites wherein the selecting is performed using a trained reward function (mere instructions to apply the exception using generic computer components as discussed in MPEP 2106.05(f) and a field of use and technological environment as discussed in MPEP 2106.05(h)). Therefore, in view of the considerations set forth in MPEP 2106.04(d), 2106.05(a)-(c) and (e)-(h), the additional elements as recited in the dependent claims discussed above alone or in combination do not integrate the judicial exception into a practical application as they are mere insignificant extra solution activity, combined with implementing the abstract idea using generic computer components, and limitations describing a field of use or technological environment. The additional elements as discussed above, in combination with the abstract idea, are not sufficient to amount to significantly more than the judicial exception as they are well, understood, routine and conventional activity as disclosed in combination with generic computer functions and components used to implement the abstract idea, and limitations describing a field of use or technological environment. Claim Rejections – 35 USC § 102 In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status. The following is a quotation of the appropriate paragraphs of 35 U.S.C. 102 that form the basis for the rejections under this section made in this Office action: A person shall be entitled to a patent unless – (a)(1) the claimed invention was patented, described in a printed publication, or in public use, on sale or otherwise available to the public before the effective filing date of the claimed invention. (a)(2) the claimed invention was described in a patent issued under section 151, or in an application for patent published or deemed published under section 122(b), in which the patent or application, as the case may be, names another inventor and was effectively filed before the effective filing date of the claimed invention. Claims 14, 15, 17, and 19 are rejected under 35 U.S.C. 102(a)(1) as being anticipate by R. Jie, X. Meng, L. Shang, X. Jiang, Q. Liu. “Prompt-Based Length Controlled Generation with Reinforcement Learning.” ArXiv.org. September 30, 2023. Retrieved from the Internet: https://arxiv.org/pdf/2308.12030. (hereinafter referred to as Jie). With respect to claim 14, Jie teaches a method implemented using one or more processors and comprising: retrieving a generative model interaction set that includes an input prompt and two or more candidate generative model responses that were generated by processing the input prompt using one or more generative models (e.g. page 4, Fig. 1, showing receiving/retrieving a user utterance with a control instruction, extracting a standard control prompt from the user utterance, and a corresponding set of LLM output sequences which are generative model responses; page 5, final paragraph, generated sequences; page 7, section 3.5, N output sequences; page 10, final paragraph, setting number of output sequences; i.e. multiple LLM outputs/responses may be generated in response to a user utterance); analyzing the input prompt to identify one or more length constraints intended to be imposed on the generative model responses (e.g. page 4, first paragraph, original user utterances may involve a control instruction on length constraint; understanding user intention on length constraint; introducing standard prompt extractor to parse the information of length constrain from user utterance and thus generate a standard length control prompt; page 5, Fig. 2, showing the user utterance including the length constraint and the resulting standard control prompt including the identified length constraint; page 6, first paragraph, SPE taking user utterance as input and outputting SCP consisting of basic description of what length constraint should be satisfied); evaluating the candidate generative model responses for compliance with one or more of the length constraints (e.g. page 4, second paragraph, reward model taking both standard control prompt and generated sequence (i.e. of the LLM in response to the original user utterance) and outputs a score to evaluate how well the generated sequence meets the requirement of this control prompt; page 5, first paragraph, evaluating whether the generated text meets the requirement of length control instruction using reward model; using the actual length of the output sequence, we calculate the rewards based on the right column of Table 1); based on the evaluating, selecting, for inclusion in a generative model training example, the input prompt and the candidate generative model response that most closely complies with one or more of the length constraints (e.g. page 4, final paragraph, the score can be applied as a reward signal in reinforcement learning method to fine-tune LLMs or applied to rank and select the generated sequences in inference; Fig. 1 caption, in the training stage, the scores given by the reward model are used for the reinforcement learning method; in the inference stage, the scores are applied for ranking and selecting the output sequences generated by LLMs; page 5, first paragraph, score used as reward for fine-tuning existing LLMs by leveraging reinforcement learning, or to rank and select candidates generated by LLMs; page 7, section 3.5, applying reward model to score, rank, and select from set of generated samples; selecting one sequence out of N output sequences as final output sequence, and using it for the RL fine-tuning phase; page 12, section 4.5, fine-tune the pre-trained GPT-medium or BERT-large as the trainable reward models to predict the reward labels by minimizing MSE losses; applying the modified PPO algorithm with the reward models with the GPT-small models to perform length-control summarization on the augmented data based on CNNDM; i.e., a plurality of generative model responses corresponding to the input prompt may be output, and these may be evaluated, scored, and ranked according to how well they meet the actual requested length, and then the output of the generative model having the highest ranking (i.e. which meets the actual length requirement) is selected for use in further training/fine-tuning, analogous to selecting an input prompt and response that that most closely complies with one or more of the length constraints for use/inclusion in a training example); and training one or more of the generative models using the generative model training example (e.g. page 7, section 3.5, selecting one sequence out of N output sequences as final output sequence, and using it for the RL fine-tuning phase; page 9, section 4.1.3, describing supervised finetuning of the GPT models using randomly sampled augmented control prompts and reference texts; performing SFT on labelled data to enable pre-trained GPTs to summarize texts with a length control ability; page 9, section 4.1.4, further improving the accuracy of length control via reinforcement learning; processing augmented inputs with SCP and using rule-based reward model based on the extracted standard control information and/or applying a trainable reward model). With respect to claim 15, Jie teaches all of the limitations of claim 14 as previously discussed, and further teaches wherein the two or more candidate generative model responses comprise first and second candidate generative model responses, generated using the same generative model, which are different from each other (e.g. page 4, Fig. 1, showing receiving/retrieving a user utterance with a control instruction, extracting a standard control prompt from the user utterance, and a corresponding set of LLM output sequences which are generative model responses; page 5, final paragraph, generated sequences; page 7, section 3.5, N output sequences; page 10, final paragraph, setting number of output sequences; i.e. Jie teaches at least one LLM and multiple LLM outputs/responses may be generated in response to a user utterance, which can include two or more responses generated by a same generative model/LLM). With respect to claim 17, Jie teaches all of the limitations of claim 14 as previously discussed, and further teaches wherein the two or more candidate generative model responses comprise a first candidate generative model response generated using a first generative model and a second candidate generative model response generated using a second generative model that is different from the first generative model (e.g. page 4, Fig. 1, showing receiving/retrieving a user utterance with a control instruction, extracting a standard control prompt from the user utterance, and a corresponding set of LLM output sequences which are generative model responses; multiple LLMs including GPT, LLaMA, PALM, etc., as shown in Fig. 1; page 5, final paragraph, generated sequences; page 7, section 3.5, N output sequences; page 10, final paragraph, setting number of output sequences; i.e. Jie teaches multiple LLMs (see Fig. 1) and multiple LLM outputs/responses generated in response to a user utterance, which can include at least a first response generated by a first LLM and a second response generated by a second LLM which is different). With respect to claim 19, Jie teaches all of the limitations of claim 14 as previously discussed, and further teaches wherein the selecting is performed using a trained reward function (e.g. page 4, final paragraph, the score can be applied as a reward signal in reinforcement learning method to fine-tune LLMs or applied to rank and select the generated sequences in inference; Fig. 1 caption, in the training stage, the scores given by the reward model are used for the reinforcement learning method; in the inference stage, the scores are applied for ranking and selecting the output sequences generated by LLMs; page 5, first paragraph, score used as reward for fine-tuning existing LLMs by leveraging reinforcement learning, or to rank and select candidates generated by LLMs; page 7, section 3.5, applying reward model to score, rank, and select from set of generated samples; selecting one sequence out of N output sequences as final output sequence, and using it for the RL fine-tuning phase; page 12, section 4.5, fine-tune the pre-trained GPT-medium or BERT-large as the trainable reward models to predict the reward labels by minimizing MSE losses; applying the modified PPO algorithm with the reward models with the GPT-small models to perform length-control summarization on the augmented data based on CNNDM). Claim Rejections – 35 USC § 103 The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action: A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made. The factual inquiries set forth in Graham v. John Deere Co., 383 U.S. 1, 148 USPQ 459 (1966), that are applied for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows: 1. Determining the scope and contents of the prior art. 2. Ascertaining the differences between the prior art and the claims at issue. 3. Resolving the level of ordinary skill in the pertinent art. 4. Considering objective evidence present in the application indicating obviousness or nonobviousness. This application currently names joint inventors. In considering patentability of the claims under pre-AIA 35 U.S.C. 103(a), the examiner presumes that the subject matter of the various claims was commonly owned at the time any inventions covered therein were made absent any evidence to the contrary. Applicant is advised of the obligation under 37 CFR 1.56 to point out the inventor and invention dates of each claim that was not commonly owned at the time a later invention was made in order for the examiner to consider the applicability of pre-AIA 35 U.S.C. 103(c) and potential pre-AIA 35 U.S.C. 102€, (f) or (g) prior art under pre-AIA 35 U.S.C. 103(a). Claims 1-7, 11-13, and 20 are rejected under 35 U.S.C. 103 as being unpatentable over Jie in view of Murphy et al. (US 20240291853 A1). With respect to claims 1 and 20, Jie teaches at least on non-transitory computer-readable medium comprising instructions that, in response to execution by one or more processors, cause the one or more processors to perform a method; and the method implemented using one or more processors and comprising: retrieving a candidate generative model training example, wherein the candidate generative model training example includes an input prompt and a generative model response that was generated by processing the input prompt using one or more generative models (e.g. page 4, Fig. 1, showing receiving/retrieving a user utterance with a control instruction, extracting a standard control prompt from the user utterance, and a corresponding LLM output sequence which is a generative model response); analyzing the input prompt to identify one or more length constraints intended to be imposed on the generative model response (e.g. page 4, first paragraph, original user utterances may involve a control instruction on length constraint; understanding user intention on length constraint; introducing standard prompt extractor to parse the information of length constrain from user utterance and thus generate a standard length control prompt; page 5, Fig. 2, showing the user utterance including the length constraint and the resulting standard control prompt including the identified length constraint; page 6, first paragraph, SPE taking user utterance as input and outputting SCP consisting of basic description of what length constraint should be satisfied); evaluating the generative model response for compliance with one or more of the length constraints (e.g. page 4, second paragraph, reward model taking both standard control prompt and generated sequence (i.e. of the LLM in response to the original user utterance) and outputs a score to evaluate how well the generated sequence meets the requirement of this control prompt; page 5, first paragraph, evaluating whether the generated text meets the requirement of length control instruction using reward model; using the actual length of the output sequence, we calculate the rewards based on the right column of Table 1); based on a determination that the generative model response fails to comply with one or more of the length constraints, modifying the candidate generative model training example to generate a synthetic generative model training example for which one or more length constraints are satisfied (e.g. page 4, final paragraph, the score can be applied as a reward signal in reinforcement learning method to fine-tune LLMs or applied to rank and select the generated sequences in inference; Fig. 1 caption, in the training stage, the scores given by the reward model are used for the reinforcement learning method; in the inference stage, the scores are applied for ranking and selecting the output sequences generated by LLMs; page 5, first paragraph, score used as reward for fine-tuning existing LLMs by leveraging reinforcement learning, or to rank and select candidates generated by LLMs; page 7, section 3.5, applying reward model to score, rank, and select from set of generated samples; selecting one sequence out of N output sequences as final output sequence, and using it for the RL fine-tuning phase; pages 7-8, section 4.1.1, designing set of SCPs with five types of control instructions with randomly sampled target summary length from 50 tokens to 150 tokens and filling these lengths into field of randomly sampled SCP; examples of templates shown in Fig. 2 and Appendix A.3; creating augmented input data by replacing placeholders in augmented templates with target numbers and original articles; page 12, section 4.5, building a simulated dataset with 100,000 examples using the original CNNDM and NYT datasets by concatenating sampled standard control prompts with randomly sampled target lengths from 50 to 150 and the summaries from the labels; the score labels are then calculated by comparing the real lengths of labeled summaries and the sampled control prompts using table 1; then we fine-tune the pre-trained GPT-medium or BERT-large as the trainable reward models to predict the reward labels by minimizing MSE losses; applying the modified PPO algorithm with the reward models with the GPT-small models to perform length-control summarization on the augmented data based on CNNDM; compare with claims 4 and 6, indicating that this recited modifying step can include altering the response to match the length constraint by processing the input prompt using the generative model to generate a new generative model response; i.e., a plurality of generative model responses corresponding to the input prompt may be output, and these may be evaluated, scored, and ranked according to how well they meet the actual requested length, and then the output of the generative model having the highest ranking (i.e. which meets the actual length requirement) is selected for use in further training/fine-tuning, where this selection of this output meeting the actual length requirement as the sample for training/fine-tuning is analogous to modifying/altering the generative model response to match the length constraint by using the generative model to generate a new response as recited in claims 4 and 6); and training one or more of the generative models using the synthetic generative model training example (e.g. page 7, section 3.5, selecting one sequence out of N output sequences as final output sequence, and using it for the RL fine-tuning phase; page 9, section 4.1.3, describing supervised finetuning of the GPT models using randomly sampled augmented control prompts and reference texts; performing SFT on labelled data to enable pre-trained GPTs to summarize texts with a length control ability; page 9, section 4.1.4, further improving the accuracy of length control via reinforcement learning; processing augmented inputs with SCP and using rule-based reward model based on the extracted standard control information and/or applying a trainable reward model). Assuming arguendo that Jie does not explicitly disclose based on a determination that the generative model response fails to comply with one or more of the length constraints, modifying the candidate generative model training example to generate a synthetic generative model training example for which one or more length constraints are satisfied, Murphy teaches based on a determination that the generative model response fails to comply with one or more of the length constraints, modifying the candidate generative model training example to generate a synthetic generative model training example for which one or more length constraints are satisfied (e.g. paragraph 0390 and 0393, describing use us generative models and LLMs, such as for generating new data samples similar to the training data; generating synthetic data for training more robust discriminative models; paragraph 0435-0436, generative AI model 302 designed to generate new, synthetic data that resembles its training data; formatting script including instructions/code to structure, preprocess, or format data in a way that’s optimal for interaction with or processing by an LLM including structuring prompts and formatting the model’s outputs for specific applications; paragraph 0437 and 0439-0440, application using LLM to generate content based on user inputs, where formatting script is used to tailor prompts to fit specific use cases or elicit more accurate responses from the model, including adding specific instructions or context to the prompt that guides the model in generating the desired output and post-processing model outputs, including by truncating model responses to fit length constraints; paragraph 0443, using generative model and formatting script to produce a summarized human-readable report using LLM; paragraph 0452, using generative mode3l to create synthetic datasets; i.e. the model output, to be used as synthetic training data, may be modified, such as truncating the model output to meet a length constraint where the original output fails to comply with this length constraint). Accordingly, it would have been obvious to one of ordinary skill in the art before the effective filing date of the invention having the teachings of Jie and Murphy in front of him to have modified the teachings of Jie (directed to prompt-based length controlled generation with reinforcement learning), to incorporate the teachings of Murphy (directed to using generative models and LLMs n a threat mitigation platform, including for generating synthetic training data and formatting it to meet various requirements including length constraints) to include the capability to modify the generative model response/output to satisfy the length constrain when it fails to meet the length constraint in its original format. One of ordinary skill would have been motivated to perform such a modification in order to help integrate large language models into broader applications or workflows, ensuring that interaction between human users and the AI is as seamless and effective as possible as described in Murphy (paragraph 0442). With respect to claim 2, Jie in view of Murphy teaches all of the limitations of claim 1 as previously discussed, and Jie further teaches wherein the modifying comprises altering one or more of the length constraints of the input prompt to match one or more length features of the generative model response (e.g. page 7, section 3.5, applying reward model to score, rank, and select from set of generated samples; selecting one sequence out of N output sequences as final output sequence, and using it for the RL fine-tuning phase; pages 7-8, section 4.1.1, designing set of SCPs with five types of control instructions with randomly sampled target summary length from 50 tokens to 150 tokens and filling these lengths into field of randomly sampled SCP; examples of templates shown in Fig. 2 and Appendix A.3; creating augmented input data by replacing placeholders in augmented templates with target numbers and original articles; page 9, section 4.1.3, replacing length placeholder of augmented control prompt with real text length of reference summary and performing SFT on the labelled data to enable GPTs to summarize texts with length control ability; page 9, section 4.1.4, further improving accuracy of length control in LLMs by processing augmented inputs with SCP and using reward model; page 12, section 4.5, building a simulated dataset with 100,000 examples using the original CNNDM and NYT datasets by concatenating sampled standard control prompts with randomly sampled target lengths from 50 to 150 and the summaries from the labels; the score labels are then calculated by comparing the real lengths of labeled summaries and the sampled control prompts using table 1; then we fine-tune the pre-trained GPT-medium or BERT-large as the trainable reward models to predict the reward labels by minimizing MSE losses; applying the modified PPO algorithm with the reward models with the GPT-small models to perform length-control summarization on the augmented data based on CNNDM; compare with claims 4 and 6, indicating that this recited modifying step can include altering the response to match the length constraint by processing the input prompt using the generative model to generate a new generative model response; i.e. the modification of the training example can include modifying the target length of the training sample to be a real/actual length in the input prompt, or by selecting a new training example (i.e. generated by the model, for the purposes of performing fine tuning) which has the highest reward/score based on being the closest match between the desired length in the input prompt and the actual length of the generated response). With respect to claim 3, Jie in view of Murphy teaches all of the limitations of claim 2 as previously discussed, and Jie further teaches wherein the match comprises a fuzzy match (e.g. page 4, second paragraph and Fig. 1 caption, using reward score for ranking and selecting output sequences generated by LLMs, score applied as reward signal In reinforcement learning method to fine-tune LLMs or rank and select generated sequences; page 5, first paragraph and Table 1, scoring generated sequences based on length constraint and using the score as a reward for fine-tuning/ranking and selecting candidates; rewards calculated based on Table 1; as shown in Table 1, the reward/score is calculated based on how closely the target length (i.e. in the input prompt) and actual generated length (i.e. of the output) match one another, but does not require an exact match; page 7, section 3.5, applying reward model to score, rank, and select from set of generated samples; selecting one sequence out of N output sequences as final output sequence, and using it for the RL fine-tuning phase; i.e. the selected sequences used in fine-tuning are selected using a reward score which indicates how closely the target and actual lengths match one another, and a highest ranking/scoring of these is selected for use as the training example for fine-tuning, where the selected sequence may have a closest match while not being an exact match, such that the reward score may indicate a fuzzy match instead of being limited to only indicating exact matches). With respect to claim 4, Jie in view of Murphy teaches all of the limitations of claim 1 as previously discussed, and Murphy further teaches wherein the modifying comprises altering the generative model response to match one or more of the length constraints (e.g. paragraph 0390 and 0393, describing use us generative models and LLMs, such as for generating new data samples similar to the training data; generating synthetic data for training more robust discriminative models; paragraph 0435-0436, generative AI model 302 designed to generate new, synthetic data that resembles its training data; formatting script including instructions/code to structure, preprocess, or format data in a way that’s optimal for interaction with or processing by an LLM including structuring prompts and formatting the model’s outputs for specific applications; paragraph 0437 and 0439-0440, application using LLM to generate content based on user inputs, where formatting script is used to tailor prompts to fit specific use cases or elicit more accurate responses from the model, including adding specific instructions or context to the prompt that guides the model in generating the desired output and post-processing model outputs, including by truncating model responses to fit length constraints; paragraph 0443, using generative model and formatting script to produce a summarized human-readable report using LLM; paragraph 0452, using generative mode3l to create synthetic datasets; i.e. the model output, to be used as synthetic training data, may be modified, such as truncating the model output to meet a length constraint where the original output fails to comply with this length constraint). Accordingly, it would have been obvious to one of ordinary skill in the art before the effective filing date of the invention having the teachings of Jie and Murphy in front of him to have modified the teachings of Jie (directed to prompt-based length controlled generation with reinforcement learning), to incorporate the teachings of Murphy (directed to using generative models and LLMs n a threat mitigation platform, including for generating synthetic training data and formatting it to meet various requirements including length constraints) to include the capability to modify the generative model response/output to satisfy the length constrain when it fails to meet the length constraint in its original format. One of ordinary skill would have been motivated to perform such a modification in order to help integrate large language models into broader applications or workflows, ensuring that interaction between human users and the AI is as seamless and effective as possible as described in Murphy (paragraph 0442). With respect to claim 5, Jie in view of Murphy teaches all of the limitations of claim 4 as previously discussed, and Jie further teaches wherein the match comprises a fuzzy match (e.g. page 4, second paragraph and Fig. 1 caption, using reward score for ranking and selecting output sequences generated by LLMs, score applied as reward signal In reinforcement learning method to fine-tune LLMs or rank and select generated sequences; page 5, first paragraph and Table 1, scoring generated sequences based on length constraint and using the score as a reward for fine-tuning/ranking and selecting candidates; rewards calculated based on Table 1; as shown in Table 1, the reward/score is calculated based on how closely the target length (i.e. in the input prompt) and actual generated length (i.e. of the output) match one another, but does not require an exact match; page 7, section 3.5, applying reward model to score, rank, and select from set of generated samples; selecting one sequence out of N output sequences as final output sequence, and using it for the RL fine-tuning phase; i.e. the selected sequences used in fine-tuning are selected using a reward score which indicates how closely the target and actual lengths match one another, and a highest ranking/scoring of these is selected for use as the training example for fine-tuning, where the selected sequence may have a closest match while not being an exact match, such that the reward score may indicate a fuzzy match instead of being limited to only indicating exact matches). With respect to claim 6, Jie in view of Murphy teaches all of the limitations of claim 4 as previously discussed, and Jie further teaches wherein the altering comprises processing the input prompt using one or more of the generative models to generate a new generative model response (e.g. page 4, final paragraph, the score can be applied as a reward signal in reinforcement learning method to fine-tune LLMs or applied to rank and select the generated sequences in inference; Fig. 1 caption, in the training stage, the scores given by the reward model are used for the reinforcement learning method; in the inference stage, the scores are applied for ranking and selecting the output sequences generated by LLMs; page 5, first paragraph, score used as reward for fine-tuning existing LLMs by leveraging reinforcement learning, or to rank and select candidates generated by LLMs; page 7, section 3.5, applying reward model to score, rank, and select from set of generated samples; selecting one sequence out of N output sequences as final output sequence, and using it for the RL fine-tuning phase; page 10, setting number of output sequences to 8 and selecting the one with the highest reward; compare with claims 4 and 6, indicating that this recited modifying step can include altering the response to match the length constraint by processing the input prompt using the generative model to generate a new generative model response; i.e., a plurality of generative model responses corresponding to the input prompt may be output, and these may be evaluated, scored, and ranked according to how well they meet the actual requested length, and then the output of the generative model having the highest ranking (i.e. which meets the actual length requirement) is selected for use in further training/fine-tuning, where this selection of this output meeting the actual length requirement as the sample for training/fine-tuning is analogous to modifying/altering the generative model response to match the length constraint by using the generative model to generate a new response). With respect to claim 7, Jie in view of Murphy teaches all of the limitations of claim 1 as previously discussed, and Jie further teaches wherein analyzing the input prompt comprises parsing the input prompt to detect one or more linguistic concepts and numeric modifiers of the one or more linguistic concepts (e.g. page 4, first paragraph, original user utterances may involve a control instruction on length constraint; understanding user intention on length constraint; introducing standard prompt extractor to parse the information of length constraint from user utterance and thus generate a standard length control prompt; page 5, Fig. 2, showing the user utterance including the length constraint and the resulting standard control prompt including the identified length constraint; page 5, section 3.2, using SPE to parse each user utterance and get its type of length constraint and target values; page 6, first paragraph, SPE taking user utterance as input and outputting SCP consisting of basic description of what length constraint should be satisfied; predicting type of SCP and target numbers involved as shown in Fig. 2). With respect to claim 11, Jie in view of Murphy teaches all of the limitations of claim 1 as previously discussed, and Jie further teaches wherein analyzing the input prompt comprises performing natural language processing (NLP) on the input prompt to identify an intent behind the prompt and one or more parameters of the intent, wherein one or more of the parameters of the intent include one or more of the length constraints (e.g. page 4, first paragraph, original user utterances may involve a control instruction on length constraint; understanding user intention on length constraint; introducing standard prompt extractor to parse the information of length constrain from user utterance and thus generate a standard length control prompt; page 5, Fig. 2, showing the user utterance including the length constraint and the resulting standard control prompt including the identified length constraint; page 6, first paragraph, SPE taking user utterance as input and outputting SCP consisting of basic description of what length constraint should be satisfied). With respect to claim 12, Jie in view of Murphy teaches all of the limitations of claim 1 as previously discussed, and Jie further teaches wherein analyzing the input prompt comprises: assembling data indicative of the input prompt into an auxiliary input prompt; assembling, into the auxiliary prompt, data indicative of a natural language request to identify the one or more length constraints in the input prompt; and processing the auxiliary input prompt using one or more of the generative models to generate auxiliary generative model output indicative of one or more of the length constraints (e.g. page 4, first paragraph, using SPE to parse the information of length constraint from user utterance and thus generate a standard length control prompt which includes different types of length constraint and can be further applied for rule-based inference and evaluation; as shown in Fig. 1, the prompt extractor extracts the standard control prompt from the user utterance; pages 5-6, section 3.3 and Fig. 2, outputting SCPs using generative model and/or discriminative model, where the SPE takes user utterance as input and outputs SCP consisting of basic description of what length constraint should be satisfied; page 6, section 3.4, reward/critic model taking concatenation of SCP and generated text as input (and provide corresponding reward as output); page 12, section 4.5, indicating that the reward model can be a trainable reward model such as a GPT model; i.e. the actual user input prompt/utterance is analyzed to assemble a standard control prompt, analogous to the recited auxiliary prompt, where this is then provided as input to a model for further processing, including outputting of information indicative of the length constraints (including a reward value/score indicating closeness of the length constraint to an actual length)). With respect to claim 13, Jie in view of Murphy teaches all of the limitations of claim 4 as previously discussed, and Jie further teaches wherein one or more of the generative models comprises a large language model (LLM) (e.g. page 4, second paragraph and Fig. 1, user utterance passed through LLMs like GPT-family, PALM, LLaMA, Pangu, Ernie, etc.; LLMs generating output sequence according to user utterance; reward model; page 12, section 4.5, indicating that the reward model can be a trainable reward model such as a GPT model). Claims 8-10 are rejected under 35 U.S.C. 103 as being unpatentable over Jie in view of Murphy, further in view of A. Lu, H. Zhang, Y, Zhang, X. Wang, D. Yang. “Bounding the Capabilities of Large Language Models in Open Text Generation with Prompt Constraints.” ArXiv.org. February 17, 2023. Retrieved from the Internet: https://arxiv.org/pdf/2302.09185. (hereinafter referred to as “Lu”). With respect to claim 8, Jie in view of Murphy teaches all of the limitations of claim 7 as previously discussed. Jie and Murphy do not explicitly disclose wherein the one or more linguistic concepts include a sentence, and the one or more numeric modifiers include a number of requested sentences. However, Lu teaches wherein the one or more linguistic concepts include a sentence, and the one or more numeric modifiers include a number of requested sentences (e.g. page 7, section 5.1, discussing structural constraints on LLM outputs, including numerical constraints constraining text to a set or bounded number of words, sentences, or paragraphs; creating prompts with numerical requirements: five, ten, twenty, on three different language structure levels: word, sentence, and paragraph). Accordingly, it would have been obvious to one of ordinary skill in the art before the effective filing date of the invention having the teachings of Jie, Murphy, and Lu in front of him to have modified the teachings of Jie (directed to prompt-based length controlled generation with reinforcement learning) and Murphy (directed to using generative models and LLMs in a threat mitigation platform, including for generating synthetic training data and formatting it to meet various requirements including length constraints), to incorporate the teachings of Lu (directed to prompt constraints for LLMs) to include the capability to include, as the linguistic concepts, words, sentences, or paragraphs, and, as the numeric modifiers, a number of requested words, sentences, or paragraphs. One of ordinary skill would have been motivated to perform such a modification in order to facilitate finer-grained analyses of open text generation, analyze capabilities and limitations of LLMs on open-ended text generation, and analyze in-context mitigation strategies to improve model performance as described in Lu (page 2, bulleted list at end of first column, continuing on second column). With respect to claim 9, Jie in view of Murphy teaches all of the limitations of claim 7 as previously discussed. Jie and Murphy do not explicitly disclose wherein the one or more linguistic concepts include a word, and the one or more numeric modifiers include a number of requested words. However, Lu teaches wherein the one or more linguistic concepts include a word, and the one or more numeric modifiers include a number of requested words (e.g. page 7, section 5.1, discussing structural constraints on LLM outputs, including numerical constraints constraining text to a set or bounded number of words, sentences, or paragraphs; creating prompts with numerical requirements: five, ten, twenty, on three different language structure levels: word, sentence, and paragraph). Accordingly, it would have been obvious to one of ordinary skill in the art before the effective filing date of the invention having the teachings of Jie, Murphy, and Lu in front of him to have modified the teachings of Jie (directed to prompt-based length controlled generation with reinforcement learning) and Murphy (directed to using generative models and LLMs in a threat mitigation platform, including for generating synthetic training data and formatting it to meet various requirements including length constraints), to incorporate the teachings of Lu (directed to prompt constraints for LLMs) to include the capability to include, as the linguistic concepts, words, sentences, or paragraphs, and, as the numeric modifiers, a number of requested words, sentences, or paragraphs. One of ordinary skill would have been motivated to perform such a modification in order to facilitate finer-grained analyses of open text generation, analyze capabilities and limitations of LLMs on open-ended text generation, and analyze in-context mitigation strategies to improve model performance as described in Lu (page 2, bulleted list at end of first column, continuing on second column). With respect to claim 10, Jie in view of Murphy teaches all of the limitations of claim 7 as previously discussed. Jie and Murphy do not explicitly disclose wherein the one or more linguistic concepts include a paragraph, and the one or more numeric modifiers include a number of requested paragraphs. However, Lu teaches wherein the one or more linguistic concepts include a paragraph, and the one or more numeric modifiers include a number of requested paragraphs(e.g. page 7, section 5.1, discussing structural constraints on LLM outputs, including numerical constraints constraining text to a set or bounded number of words, sentences, or paragraphs; creating prompts with numerical requirements: five, ten, twenty, on three different language structure levels: word, sentence, and paragraph). Accordingly, it would have been obvious to one of ordinary skill in the art before the effective filing date of the invention having the teachings of Jie, Murphy, and Lu in front of him to have modified the teachings of Jie (directed to prompt-based length controlled generation with reinforcement learning) and Murphy (directed to using generative models and LLMs in a threat mitigation platform, including for generating synthetic training data and formatting it to meet various requirements including length constraints), to incorporate the teachings of Lu (directed to prompt constraints for LLMs) to include the capability to include, as the linguistic concepts, words, sentences, or paragraphs, and, as the numeric modifiers, a number of requested words, sentences, or paragraphs. One of ordinary skill would have been motivated to perform such a modification in order to facilitate finer-grained analyses of open text generation, analyze capabilities and limitations of LLMs on open-ended text generation, and analyze in-context mitigation strategies to improve model performance as described in Lu (page 2, bulleted list at end of first column, continuing on second column). Claims 16 and 18 are rejected under 35 U.S.C. 103 as being unpatentable over Jie in view of Lu. With respect to claim 16, Jie teaches all of the limitations of claim 15 as previously discussed. Jie does not explicitly disclose wherein the first and second candidate generative model responses differ from each other due to a temperature parameter used in association with the generative model. However, Lu teaches wherein the first and second candidate generative model responses differ from each other due to a temperature parameter used in association with the generative model (e.g. page 6, section 4.5, describing comparing model responses to prompts for at least one model according to multiple different temperature parameters; higher temperature creates better results, but a lower temperature repeats an output that happens to perform well). Accordingly, it would have been obvious to one of ordinary skill in the art before the effective filing date of the invention having the teachings of Jie, Murphy, and Lu in front of him to have modified the teachings of Jie (directed to prompt-based length controlled generation with reinforcement learning) and Murphy (directed to using generative models and LLMs in a threat mitigation platform, including for generating synthetic training data and formatting it to meet various requirements including length constraints), to incorporate the teachings of Lu (directed to prompt constraints for LLMs) to include the capability to utilize a temperature parameter to vary generative model responses. One of ordinary skill would have been motivated to perform such a modification in order to facilitate finer-grained analyses of open text generation, analyze capabilities and limitations of LLMs on open-ended text generation, and analyze in-context mitigation strategies to improve model performance as described in Lu (page 2, bulleted list at end of first column, continuing on second column). With respect to claim 18, Jie teaches all of the limitations of claim 17 as previously discussed. Jie does not explicitly disclose wherein the second generative model comprises fewer parameters than the first generative model. However, Lu teaches wherein the second generative model comprises fewer parameters than the first generative model (e.g. page 6, section 4.5, describing comparing model responses to prompts for multiple different models, and noting that the models correspond to Instruct GPT models of 350M, 1.3B, and 6.7B parameters; i.e. responses to prompts are generated using multiple models, where at least one model comprises fewer parameters and at least one other model). Accordingly, it would have been obvious to one of ordinary skill in the art before the effective filing date of the invention having the teachings of Jie, Murphy, and Lu in front of him to have modified the teachings of Jie (directed to prompt-based length controlled generation with reinforcement learning) and Murphy (directed to using generative models and LLMs in a threat mitigation platform, including for generating synthetic training data and formatting it to meet various requirements including length constraints), to incorporate the teachings of Lu (directed to prompt constraints for LLMs) to include the capability to utilize generative models having different amounts of parameters, such that the second generative model comprises fewer parameters than the first generative model. One of ordinary skill would have been motivated to perform such a modification in order to facilitate finer-grained analyses of open text generation, analyze capabilities and limitations of LLMs on open-ended text generation, and analyze in-context mitigation strategies to improve model performance as described in Lu (page 2, bulleted list at end of first column, continuing on second column). It is noted that any citation to specific pages, columns, lines, or figures in the prior art references and any interpretation of the references should not be considered to be limiting in any way. “The use of patents as references is not limited to what the patentees describe as their own inventions or to the problems with which they are concerned. They are part of the literature of the art, relevant for all they contain,” In re Heck, 699 F.2d 1331, 1332-33, 216 USPQ 1038, 1039 (Fed. Cir. 1983) (quoting in re Lemelson, 397 F.2d 1006, 1009, 158 USPQ 275, 277 (GCPA 1968)). Further, a reference may be relied upon for all that it would have reasonably suggested to one having ordinary skill the art, including nonpreferred embodiments. Merck & Co, v. Biocraft Laboratories, 874 F.2d 804, 10 USPQ2d 1843 (Fed. Cir.), cert, denied, 493 U.S. 975 (1989). See also Upsher-Smith Labs. v. Pamlab, LLC, 412 F,3d 1319, 1323, 75 USPQ2d 1213, 1215 (Fed. Cir, 2005): Celeritas Technologies Ltd. v. Rockwell International Corp., 150 F.3d 1354, 1361, 47 USPQ2d 1516, 1522-23 (Fed. Cir. 1998). Conclusion The prior art made of record and not relied upon is considered pertinent to applicant’s disclosure. Any inquiry concerning this communication or earlier communications from the examiner should be directed to JEREMY L STANLEY whose telephone number is (469)295-9105. The examiner can normally be reached on Monday-Friday from 9:00 AM to 5:00 PM CST. If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Abdullah Al Kawsar, can be reached at telephone number (571) 270-3169. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300. Information regarding the status of an application may be obtained from Patent Center and the Private Patent Application Information Retrieval (PAIR) system. Status information for published applications may be obtained from Patent Center or Private PAIR. Status information for unpublished applications is available through Patent Center and Private PAIR for authorized users only. Should you have questions about access to the Private PAIR system, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) Form at https://www.uspto.gov/patents/uspto-automated- interview-request-air-form. /JEREMY L STANLEY/ Primary Examiner, Art Unit 2127
Read full office action

Prosecution Timeline

Apr 23, 2024
Application Filed
Nov 13, 2026
Non-Final Rejection mailed — §101, §102, §103 (current)

Precedent Cases

Applications granted by this same examiner with similar technology

Patent 12801984
SYSTEM AND METHOD FOR MONITORING AND REMOVING DRIFT IN MACHINE LEARNING MODELS
3y 7m to grant Granted Nov 17, 2026
Patent 12792908
METHOD FOR SOLVING PROBLEM AND SYSTEM THEREOF
3y 6m to grant Granted Nov 10, 2026
Patent 12783866
DATA AUGMENTATION FOR TRAINING ARTIFICIAL INTELLIGENCE MODEL
5y 0m to grant Granted Nov 03, 2026
Patent 12783864
COMPUTER-IMPLEMENTED METHOD FOR CREATING ENCODED DATA
4y 11m to grant Granted Nov 03, 2026
Patent 12770030
METHOD AND SYSTEM FOR EXCEPTION MANAGEMENT
3y 3m to grant Granted Oct 20, 2026
Study what changed to get past this examiner. Based on 5 most recent grants.

Strategy Recommendation AI-generated — please review before filing

Get a prosecution strategy drawn from examiner precedents, rejection analysis, and claim mapping.
Typically takes 5-10 seconds — AI-generated, attorney review required before filing

Prosecution Projections

1-2
Expected OA Rounds
50%
Grant Probability
88%
With Interview (+38.1%)
3y 3m (~8m remaining)
Median Time to Grant
Low
PTA Risk
Based on 299 resolved cases by this examiner. Grant probability derived from career allowance rate.

Sign in with your work email

Enter your email to receive a magic link. No password needed.

Personal email addresses (Gmail, Yahoo, etc.) are not accepted.

Free tier: 3 strategy analyses per month