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 .
Continued Examination Under 37 CFR 1.114
A request for continued examination under 37 CFR 1.114, including the fee set forth in 37 CFR 1.17(e), was filed in this application after final rejection. Since this application is eligible for continued examination under 37 CFR 1.114, and the fee set forth in 37 CFR 1.17(e) has been timely paid, the finality of the previous Office action has been withdrawn pursuant to 37 CFR 1.114. Applicant's submission filed on February 26th, 2026 has been entered.
Information Disclosure Statement
The information disclosure statement filed February 26th, 2026 fails to comply with 37 CFR 1.98(a)(2), which requires a legible copy of each cited foreign patent document; each non-patent literature publication or that portion which caused it to be listed; and all other information or that portion which caused it to be listed. It has been placed in the application file, but the information referred to therein has not been considered.
Response to Amendment
The amendment filed January 27th, 2026 has been entered. Claims 1, 11, 12, and 17 have been amended. Claims 1-20 are pending and have been examined.
Response to Arguments
Applicant’s arguments with respect to claim(s) 1-20 have been considered but are moot because the new ground of rejection does not rely on any reference applied in the prior rejection of record for any teaching or matter specifically challenged in the argument.
Claim Rejections - 35 USC § 103
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 (i.e., changing from AIA to pre-AIA ) 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 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.
Claim(s) 1-7 and 9-14 is/are rejected under 35 U.S.C. 103 as being unpatentable over Sun et al. (US Pat. Pub. No. 2024/0249080 A1 hereinafter Sun), in view of Souche Christian et al. (EP Pat. No. 2881898 A1 hereinafter Souche), Nagaraju et al. (US Pat. Pub. No. 2024/0185001 A1 hereinafter Nagaraju), and Madaan et al. (Madaan, Aman, et al. "Self-refine: Iterative refinement with self-feedback, 2023." URL https://arxiv. org/abs/2303.17651v1 (2023). hereinafter Madaan).
Regarding claim 1, Sun discloses a method, comprising: generating, in a first phase by a large language model (LLM), an output for each question in a dataset in response to an input prompt provided to the LLM (Sun, Fig. 4; [0044]: "The system 400 can include a plurality of inputs 402, an LLM 404, a plurality of outputs 406, a score compute module 408, and a plurality of scores 410."); aggregating, in the first phase, the output for each question into a first output (Sun, Fig. 4; [0044]: "The outputs 406 can include reasoning paths and answers."; Here, the outputs are seen as being aggregated due to each selected output being an input to the system's second stage.); generating, in a second phase by the LLM, a plurality of outputs for each question in the dataset in response to a plurality of one-shot prompts provided to the LLM, wherein the plurality of one-shot prompts are generated by the LLM using the first output (Sun, Fig. 5; [0064]: "FIG. 5 depicts a block diagram of an example system 500 for the second stage of COSP. The second stage can query with the selected demonstrations. The system 500 can include generated outputs as in-context demonstrations 502, an LLM 504, which can correspond to the LLM 404 of FIG. 4, and output 506."; [0067]: "COSP can also be adapted to a few-shot setup, where a small number of labeled demonstrations Q can be available. The labeled demonstrations can be augmented with more demonstrations. Instead of querying LLMs with Zero-shot CoT in the first stage, Few-shot CoT can be utilized with Q."); aggregating, in a second phase, the plurality of outputs for each question into a second output, wherein aggregating the plurality of outputs uses a filtering that keeps outputs with the same answers for each question (Sun, Fig. 5; [0065]: "The final prediction for each question can then be output as the majority vote 508 across the predictions from both stages."). However, Sun fails to expressly recite wherein aggregating the plurality of outputs uses a follow instruction filtering that removes outputs that were unable to follow instructions provided to the LLM for solving the question; storing the second output with each question in the dataset; automatically learning by the LLM from a previous phase output generated by the LLM by providing the previous phase output as input to a next phase; and improving an accuracy of outputs provided by the LLM at each phase by using the previous phase output learned by the LLM in generating the plurality of outputs in the next phase by the LLM.
PNG
media_image1.png
572
805
media_image1.png
Greyscale
Sun, Fig. 4, included for reference.
PNG
media_image2.png
564
791
media_image2.png
Greyscale
Sun, Fig. 5, included for reference.
Souche teaches wherein aggregating the plurality of outputs uses a follow instruction filtering that removes outputs that were unable to follow instructions provided to the LLM for solving the question (Souche, Page 6, Paragraph 1 ([0038] of printed publication)): “In a subsequent operation 308, responses are received from the virtual assistants, and one or more best answers to the query are identified and transmitted to the user device 104 from which the query initiated. In particular, the virtual assistant interactivity platform 102 is for example capable of identifying cases in which a virtual assistant was unable to provide a useful response, for example by detecting standard wording such as "your query has not been understood". Such responses may be discarded.”).
Sun and Souche are analogous arts because they both belong to the same field of data processing. It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the self-improving LLMs of Sun to incorporate the teachings of Souche to remove model outputs that were unable to follow the provided instructions. This allows the system to remove any outputs that are not useful (Souche, Page 6, Paragraph 1). As such, the efficiency of the system can be improved by not processing outputs that are not useful. However, Sun, in view of Souche, fails to expressly recite storing the second output with each question in the dataset; automatically learning by the LLM from a previous phase output generated by the LLM by providing the previous phase output as input to a next phase; and improving an accuracy of outputs provided by the LLM at each phase by using the previous phase output learned by the LLM in generating the plurality of outputs in the next phase by the LLM.
Nagaraju teaches storing the second output with each question in the dataset (Nagaraju, [0035]: "Output generated using conversational MLM 192 (e.g., one or more question-answer pairs) may be stored in a dialogue data store 340.").
Sun, Souche, and Nagaraju are analogous arts because they both belong to the same field of data processing. It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the self-improving LLMs of Sun, as modified by the virtual assistant interactivity platform of Souche, to incorporate the teachings of Nagaraju to store the output question-answer pairs in a dataset. This can help generate additional examples for a training dataset (Nagaraju, [0002]). This helps create more varied datasets with minimal extra effort from the user. However, Sun, in view of Souche and Nagaraju, fails to expressly recite automatically learning by the LLM from a previous phase output generated by the LLM by providing the previous phase output as input to a next phase; and improving an accuracy of outputs provided by the LLM at each phase by using the previous phase output learned by the LLM in generating the plurality of outputs in the next phase by the LLM.
Madaan teaches automatically learning by the LLM from a previous phase output generated by the LLM by providing the previous phase output as input to a next phase (Madaan, Page 1, Paragraph 2: “In this paper, we demonstrate that large language models (LLMs) can effectively replicate this human cognitive process by employing iterative feedback and refinement.”; Page 2, Paragraph 2: “SELF-REFINE consists of an iterative loop between two components: FEEDBACK, and REFINE, which work in tandem to generate high-quality outputs. Given an initial draft output generated by a model M (0 ), we pass it back to the same model M (1 ) to get feedback (1 ). Feedback on the initial output is passed back to the same model (3 ), to iteratively refine (0 ) the previously generated output. This process is repeated iteratively for a specified number of iterations, or until the model itself determines that no further refinement is necessary.”); and improving an accuracy of outputs provided by the LLM at each phase by using the previous phase output learned by the LLM in generating the plurality of outputs in the next phase by the LLM (Madaan, Fig. 8 shows each iteration of the LLM improving the accuracy of the output.).
PNG
media_image3.png
484
848
media_image3.png
Greyscale
Madaan, Fig. 8, included for reference.
Sun, Souche, Nagaraju, and Madaan are analogous arts because they each belong to the same field of data processing. It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the self-improving LLMs of Sun, as modified by the virtual assistant interactivity platform of Souche and the dataset generation method of Nagaraju, to incorporate the teachings of Madaan to repeatedly learn and improve the model by feeding the output back into the model. This allows the system to mimic the human creative generation process without an expensive human feedback loop (Madaan, Page 2, Paragraph 2). As such, the system can provide better outputs without extra human input.
Regarding claim 2, the rejection of claim 1 is incorporated. Sun, in view of Souche, Nagaraju, and Madaan, discloses all of the current invention as stated above. Sun further discloses wherein the input prompt is a chain of thought (COT) prompt that breaks a question into a series of intermediate steps that the LLM uses to lead to an answer for the question provided in the output (Sun, [0004]: "For instance, starting with a test dataset, where part of the test dataset is not yet known to the LLM, an output prediction from a LLM can be computed by prompting and using zero-shot chain of thought (CoT).").
Regarding claim 3, the rejection of claim 2 is incorporated. Sun, in view of Souche, Nagaraju, and Madaan, discloses all of the current invention as stated above. Sun further discloses wherein the COT prompt is thinking step by step (Sun, Fig. 4; [0026]: "Modern large language models (LLMs) have demonstrated impressive capabilities at sophisticated tasks, often through step-by-step reasoning.").
Regarding claim 4, the rejection of claim 1 is incorporated. Sun, in view of Souche, Nagaraju, and Madaan, discloses all of the current invention as stated above. Sun further discloses wherein in the input prompt is different settings of the LLM and the output for a question is based on the different settings (Sun, [0045]: "For each test question x.sup.(i), the LLM 404 can be queried m times with a non-zero temperature to extract multiple reasoning paths {r.sub.j.sup.(i)}.sub.j=1.sup.m and potentially different answers {ŷ.sub.j.sup.(i)}.sub.j=1.sup.m, such as according to Eq. 1."; "The non-zero temperature can be greater than 0 but less than or equal to 1, such as 0.7.").
Regarding claim 5, the rejection of claim 1 is incorporated. Sun, in view of Souche, Nagaraju, and Madaan, discloses all of the current invention as stated above. Sun further discloses wherein the input prompt is different contexts and the output for a question is based on the different contexts (Sun, [0045]: "For each test question x.sup.(i), the LLM 404 can be queried m times with a non-zero temperature to extract multiple reasoning paths {r.sub.j.sup.(i)}.sub.j=1.sup.m and potentially different answers {ŷ.sub.j.sup.(i)}.sub.j=1.sup.m, such as according to Eq. 1."; Here, the different reasoning paths are seen as different contexts.).
Regarding claim 6, the rejection of claim 1 is incorporated. Sun, in view of Souche, Nagaraju, and Madaan, discloses all of the current invention as stated above. Sun further discloses wherein aggregating the output includes performing an identity matching of the output to a question (Sun, Fig. 5; [0064]: " The system 500 can include generated outputs as in-context demonstrations 502"; Here, the in-context demonstrations are outputs matched with their respective questions.).
Regarding claim 7, the rejection of claim 1 is incorporated. Sun, in view of Souche, Nagaraju, and Madaan, discloses all of the current invention as stated above. Sun further discloses wherein the first output includes a question and answer pair for each question in the dataset (Sun, Fig. 5; [0064]: " The system 500 can include generated outputs as in-context demonstrations 502"; Here, the in-context demonstrations are outputs matched with their respective questions.).
Regarding claim 9, the rejection of claim 7 is incorporated. Sun, in view of Souche, Nagaraju, and Madaan, discloses all of the current invention as stated above. Sun further discloses wherein the plurality of one-shot prompts provide different examples of questions with answers from within the output pair generated in the first phase for the LLM to use in generating the plurality of outputs for each question (Sun, Fig. 5; [0064]: " The system 500 can include generated outputs as in-context demonstrations 502"; Here, the in-context demonstrations are outputs matched with their respective questions.; [0067]: "COSP can also be adapted to a few-shot setup, where a small number of labeled demonstrations Q can be available. The labeled demonstrations can be augmented with more demonstrations. Instead of querying LLMs with Zero-shot CoT in the first stage, Few-shot CoT can be utilized with Q.").
Regarding claim 10, the rejection of claim 1 is incorporated. Sun, in view of Souche, Nagaraju, and Madaan, discloses all of the current invention as stated above. Sun further discloses wherein the plurality of outputs provide diverse answers to each question (Sun, [0045]: "the LLM 404 can be queried m times with a non-zero temperature to extract multiple reasoning paths {r.sub.j.sup.(i)}.sub.j=1.sup.m and potentially different answers {ŷ.sub.j.sup.(i)}.sub.j=1.sup.m, such as according to Eq. 1.").
Regarding claim 11, the rejection of claim 1 is incorporated. Sun, in view of Souche, Nagaraju, and Madaan, discloses all of the current invention as stated above. Sun further discloses wherein the filtering uses a majority of votes to filter the plurality of outputs and the second output is selected from an answer with a majority of votes (Sun, Fig. 5; [0065]: "The final prediction for each question can then be output as the majority vote 508 across the predictions from both stages.").
Regarding claim 12, the rejection of claim 1 is incorporated. Sun, in view of Souche, Nagaraju, and Madaan, discloses all of the current invention as stated above. Souche further teaches wherein the second output is selected from outputs that followed the instructions (Souche, Page 6, Paragraph 1: “In a subsequent operation 308, responses are received from the virtual assistants, and one or more best answers to the query are identified and transmitted to the user device 104 from which the query initiated. In particular, the virtual assistant interactivity platform 102 is for example capable of identifying cases in which a virtual assistant was unable to provide a useful response, for example by detecting standard wording such as "your query has not been understood". Such responses may be discarded.”). The same motivation for claim 1 applies equally to claim 12.
Regarding claim 13, the rejection of claim 1 is incorporated. Sun, in view of Souche, Nagaraju, and Madaan, discloses all of the current invention as stated above. Sun further discloses wherein aggregating the plurality of outputs is performed by a LLM to generate the second output (Sun, Fig. 5; [0065]: "The final prediction for each question can then be output as the majority vote 508 across the predictions from both stages.").
Regarding claim 14, the rejection of claim 1 is incorporated. Sun, in view of Souche, Nagaraju, and Madaan, discloses all of the current invention as stated above. Sun further discloses wherein the first phase and the second phase are part of a self-learning framework that improves an accuracy of the LLM (Sun, [0037]: "With three different LLMs on a range of tasks, COSP can demonstrate a 10-15% improvement in average accuracy for 6 arithmetic and logical reasoning tasks over Zero-shot CoT with self-consistency baseline in PaLM-62B and GPT-3, and >3% improvement in PaLM-540B.").
Claim(s) 8 and 15-16 is/are rejected under 35 U.S.C. 103 as being unpatentable over Sun, in view of Souche, Nagaraju, and Madaan, as applied to claims 1-7 and 9-14 above, and further in view of Wei et al. (US Pat. Pub. No. 2023/0394328 A1 hereinafter Wei).
Regarding claim 8, the rejection of claim 7 is incorporated. Sun, in view of Souche, Nagaraju, and Madaan, discloses all of the current invention as stated above. However, Sun, in view of Souche, Nagaraju, and Madaan, fails to expressly recite wherein the plurality of one-shot prompts are different question and answer pairs randomly selected from the first output.
Wei teaches wherein the plurality of one-shot prompts are different question and answer pairs randomly selected from the first output (Wei, [0054]: "In some embodiments, sampled outputs 420 can include a number of outputs sampled from an output layer of a machine-learned model 400."; "In some embodiments, outputs are randomly sampled.").
Sun, Souche, Nagaraju, Madaan, and Wei are analogous arts because they all belong to the same field of data processing. It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the self-improving LLMs of Sun, as modified by the virtual assistant interactivity platform of Souche, the dataset generation method of Nagaraju, and the LLM self-refining method of Madaan, to incorporate the teachings of Wei to randomly sample from the model outputs. This can help check the diversity of the outputs (Wei, [0054]-[0055]). This in turn can indicate a confidence level for the final outputs.
Regarding claim 15, the rejection of claim 1 is incorporated. Sun, in view of Souche, Nagaraju, and Madaan, discloses all of the current invention as stated above. However, Sun, in view of Souche, Nagaraju, and Madaan, fails to expressly recite further comprising: using the second output as input in a next phase for use by the LLM.
Wei teaches further comprising: using the second output as input in a next phase for use by the LLM (Wei, Fig. 5; [0061]: "To perform another pass of query recursion 520, a new instructive sequence can be composed from the body of prior knowledge about the problem at hand, which can include new information generated by the model 504. For instance, query component(s) 528 can incorporate query component(s) 524 as well as the response component(s) 526."). The same motivation for claim 12 applies equally to claim 15.
Sun, Souche, Nagaraju, Madaan, and Wei are analogous arts because they all belong to the same field of data processing. It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the self-improving LLMs of Sun, as modified by the virtual assistant interactivity platform of Souche, the dataset generation method of Nagaraju, and the LLM self-refining method of Madaan, to incorporate the teachings of Wei to use query recursion along with instructive sequences to generate a final output. This allows the system to leverage its own prior work to build toward a final output (Wei, [0031]). This allows an input to be broken down into steps that each effect the output of the next step.
PNG
media_image4.png
792
551
media_image4.png
Greyscale
Wei, Fig. 5, included for reference.
Regarding claim 16, the rejection of claim 15 is incorporated. Sun, in view of Souche, Nagaraju, Madaan, and Wei, discloses all of the current invention as stated above. Wei further teaches generating, in the next phase by the LLM, a plurality of outputs for each question in the dataset in response to a plurality of input prompts provided to the LLM, wherein the plurality of input prompts are based on the second output; aggregating, in the next phase by the LLM, the plurality of outputs into a phase output; and storing the phase output with each question in the dataset (Wei, Fig. 5; [0061]: "In this manner, for instance, the model 504 can process additional query component(s) (e.g., the original target query, in bold) by leveraging its prior outputs to generate response component(s) 530."). The same motivation for claim 15 applies equally to claim 16.
Claim(s) 17 and 19-20 is/are rejected under 35 U.S.C. 103 as being unpatentable over Sun, in view of Souche and Madaan.
Regarding claim 17, Sun discloses a method, comprising: generating, in a phase of a self-learning framework by a large language model (LLM), a plurality of outputs for a question in response to different input prompts provided with the question to the LLM (Sun, Fig. 4; [0044]: "The system 400 can include a plurality of inputs 402, an LLM 404, a plurality of outputs 406, a score compute module 408, and a plurality of scores 410."); aggregating, in the phase of the self-learning framework by the LLM, the plurality of outputs for the question into a phase output (Sun, Fig. 4; [0044]: "The outputs 406 can include reasoning paths and answers."; Here, the outputs are seen as being aggregated due to each selected output being an input to the system's second stage.), wherein aggregating the plurality of outputs uses a filtering that keeps outputs with the same answers for each question (Sun, Fig. 5; [0065]: "The final prediction for each question can then be output as the majority vote 508 across the predictions from both stages."); and providing the phase output as an input prompt generated by the LLM to a next phase of the self-learning framework for use by the LLM (Sun, Fig. 4 and Fig. 5; [0064]: “The system 500 can include generated outputs as in-context demonstrations 502”). However, Sun fails to expressly recite wherein aggregating the plurality of outputs uses a follow instruction filtering that removes outputs that were unable to follow instructions provided to the LLM for solving the question; automatically learning by the LLM from a previous phase output generated by the LLM by providing the previous phase output as input to a next phase; and improving an accuracy of outputs provided by the LLM at each phase by using the previous phase output learned by the LLM in generating the plurality of outputs in the next phase by the LLM.
Souche teaches wherein aggregating the plurality of outputs uses a follow instruction filtering that removes outputs that were unable to follow instructions provided to the LLM for solving the question (Souche, Page 6, Paragraph 1 ([0038] of publication)): “In a subsequent operation 308, responses are received from the virtual assistants, and one or more best answers to the query are identified and transmitted to the user device 104 from which the query initiated. In particular, the virtual assistant interactivity platform 102 is for example capable of identifying cases in which a virtual assistant was unable to provide a useful response, for example by detecting standard wording such as "your query has not been understood". Such responses may be discarded.”).
Sun and Souche are analogous arts because they both belong to the same field of data processing. It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the self-improving LLMs of Sun to incorporate the teachings of Souche to remove model outputs that were unable to follow the provided instructions. This allows the system to remove any outputs that are not useful (Souche, Page 6, Paragraph 1). As such, the efficiency of the system can be improved by not processing outputs that are not useful. However, Sun, in view of Souche, fails to expressly recite automatically learning by the LLM from a previous phase output generated by the LLM by providing the previous phase output as input to a next phase; and improving an accuracy of outputs provided by the LLM at each phase by using the previous phase output learned by the LLM in generating the plurality of outputs in the next phase by the LLM.
Madaan teaches automatically learning by the LLM from a previous phase output generated by the LLM by providing the previous phase output as input to a next phase (Madaan, Page 1, Paragraph 2: “In this paper, we demonstrate that large language models (LLMs) can effectively replicate this human cognitive process by employing iterative feedback and refinement.”; Page 2, Paragraph 2: “SELF-REFINE consists of an iterative loop between two components: FEEDBACK, and REFINE, which work in tandem to generate high-quality outputs. Given an initial draft output generated by a model M (0 ), we pass it back to the same model M (1 ) to get feedback (1 ). Feedback on the initial output is passed back to the same model (3 ), to iteratively refine (0 ) the previously generated output. This process is repeated iteratively for a specified number of iterations, or until the model itself determines that no further refinement is necessary.”); and improving an accuracy of outputs provided by the LLM at each phase by using the previous phase output learned by the LLM in generating the plurality of outputs in the next phase by the LLM (Madaan, Fig. 8 shows each iteration of the LLM improving the accuracy of the output.).
Sun, Souche, and Madaan are analogous arts because they each belong to the same field of data processing. It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the self-improving LLMs of Sun, as modified by the virtual assistant interactivity platform of Souche, to incorporate the teachings of Madaan to repeatedly learn and improve the model by feeding the output back into the model. This allows the system to mimic the human creative generation process without an expensive human feedback loop (Madaan, Page 2, Paragraph 2). As such, the system can provide better outputs without extra human input.
Regarding claim 19, the rejection of claim 17 is incorporated. Sun, in view of Souche and Madaan, discloses all of the current invention as stated above. Sun further discloses wherein each phase of the self-learning framework improves an accuracy of outputs provided by the LLM (Sun, [0037]: "With three different LLMs on a range of tasks, COSP can demonstrate a 10-15% improvement in average accuracy for 6 arithmetic and logical reasoning tasks over Zero-shot CoT with self-consistency baseline in PaLM-62B and GPT-3, and >3% improvement in PaLM-540B.").
Regarding claim 20, the rejection of claim 17 is incorporated. Sun, in view of Souche and Madaan, discloses all of the current invention as stated above. Sun further discloses wherein the plurality of outputs provide diverse answers to the question and aggregating the plurality of outputs includes identifying the phase output from the diverse answers (Sun, [0045]: "the LLM 404 can be queried m times with a non-zero temperature to extract multiple reasoning paths {r.sub.j.sup.(i)}.sub.j=1.sup.m and potentially different answers {ŷ.sub.j.sup.(i)}.sub.j=1.sup.m, such as according to Eq. 1."; [0046]: “With the candidate pool 406 generated, S demonstrations can be selected from it.”).
Claim(s) 18 is/are rejected under 35 U.S.C. 103 as being unpatentable over Sun, in view of Souche and Madaan, as applied to claims 17 and 19-20 above, and further in view of Wei.
Regarding claim 18, the rejection of claim 17 is incorporated. Sun, in view of Souche and Madaan, discloses all of the current invention as stated above. Sun further discloses generating, in the next phase by the LLM, the plurality of outputs for the question in response to using information in the phase output for the different input prompts provided with the question to the LLM (Sun, Fig. 5; [0064]: "FIG. 5 depicts a block diagram of an example system 500 for the second stage of COSP. The second stage can query with the selected demonstrations. The system 500 can include generated outputs as in-context demonstrations 502, an LLM 504, which can correspond to the LLM 404 of FIG. 4, and output 506."); aggregating, in the next phase of the self-learning framework, the plurality of outputs into a next phase output (Sun, Fig. 5; [0065]: "The final prediction for each question can then be output as the majority vote 508 across the predictions from both stages."). However, Sun, in view of Souche and Madaan, fails to expressly recite providing the next phase output to another phase of the self-learning framework for use by the LLM.
Wei teaches providing the next phase output to another phase of the self-learning framework for use by the LLM (Wei, Fig. 5; [0061]: "In this manner, for instance, the model 504 can process additional query component(s) (e.g., the original target query, in bold) by leveraging its prior outputs to generate response component(s) 530.").
Sun, Madaan, and Wei are analogous arts because they all belong to the same field of data processing. It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the self-improving LLMs of Sun, as modified by the virtual assistant interactivity platform of Souche and the LLM self-refining method of Madaan, to incorporate the teachings of Wei to use query recursion along with instructive sequences to generate a final output. This allows the system to leverage its own prior work to build toward a final output (Wei, [0031]). This allows an input to be broken down into steps that each effect the output of the next step.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to TYLER J BECKER whose telephone number is (703)756-1271. The examiner can normally be reached M-Th, 7:15am-5:45pm PT.
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) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Daniel Washburn can be reached at (571) 272-5551. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/TYLER BECKER/ Examiner, Art Unit 2657
/DANIEL C WASHBURN/ Supervisory Patent Examiner, Art Unit 2657
Prosecution Insights