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 .
Status of Claims
Claims 1-20 are pending in this application.
Claim Objections
Claims 1, 8, 10, 17, and 19 are objected to because of the following informalities:
In claims 1, 8, 10, 17 and 19, “deploying the second neural network language model” should be corrected to read – deploying the second neural network based language model --.
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 therefore, 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 without significantly more.
Step 2A, Prong One: The independent claim 1 recites “receiving, via a data interface, a task request for a first neural network based language model to generate a response; generating, by the first neural network based language model, a plurality of candidate responses to the task request; iteratively allocating compute resources for the first neural network based language model to conduct one or more updated generations, guided by a second neural network based language model to iteratively evaluate at least the plurality of candidate responses, thereby resulting in a final response out of the plurality of candidate responses; computing a metric indicating helpfulness of the second neural network model based language model to scale test-time performance of the first neural network language model; and deploying the second neural network language model in conjunction with the first neural network language model for test-time resource scaling”.
Claims 1, 10, and 19 recite producing one or more candidate answers to a user task, judging the first language model then evaluating those candidates and helps choose, rank, or refine them using the second language model.
[Abstract idea indicators]
Receiving a task from a user and generating a plurality of candidate responses— a task that a human routinely performs mentally or with conventional tools.
Repeatedly evaluating and modifying the candidates, i.e., a decision-making process.
Computing a metric and combining the first and second model that are mental processes with pen and paper.
Accordingly, the claims are directed to the judicial exception of a mental process.
Step 2A, Prong Two: This judicial exception is not integrated into a practical application. The computer is recited at a high-level of generality (i.e., as performing a generic computer function, e.g. collecting data, calculating, and being used as an applying) such that it amounts no more than mere instructions to apply the exception using a generic computer. Accordingly, there additional element does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea. The claim is directed to an abstract idea.
Step 2B — Claims Do Not Recite an Inventive Concept That Transforms the Mental Process into Patent-Eligible Subject Matter
The claims add generic, well-understood computer components (memory and control circuitry) and do not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to the integration of the abstract idea into a practical application, the additional element of using a computer amounts to no more than mere instructions to apply an exception using a generic computer component. Mere instructions to apply an exception using a generic computer component cannot provide an inventive concept. The claim is not patent eligible.
Applying Alice step two and relevant Federal Circuit precedent:
The recitation of conventional computer components (memory, processor, machine-readable medium) performing routine functions does not supply an inventive concept.
The mere invocation of “neural network based language model” without particularity does not demonstrate an unconventional machine or technique or a specific improvement in computer technology.
The claims recite high-level, result-oriented steps (e.g., “receiving,” “generating,” “allocating”, “resulting”, “computing”, “deploying”) that describe mental processes rather than specific technical means for performing those processes.
Because the claims lack limitations that tie the mental-process steps to a particular way of achieving a technological improvement (for example, a novel model architecture, specialized data representation, unique training regimen that yields demonstrable technical performance gains, a specialized streaming/decoding pipeline that reduces latency by a quantifiable amount, or hardware/software co-design), the additional elements do not transform the mental processes into significantly more.
Therefore, claims 1, 10, and 19 fail to recite an inventive concept sufficient to transform the judicial exception into patent-eligible subject matter.
With respect to dependent claims 2 and 11, the claims do not include additional elements that are sufficient to amount to significantly more than the judicial exception.
With respect to dependent claims 3, 12, and 20, the claims do not include additional elements that are sufficient to amount to significantly more than the judicial exception.
With respect to dependent claims 4 and 13, the claims do not include additional elements that are sufficient to amount to significantly more than the judicial exception.
With respect to dependent claims 5 and 14, the claims do not include additional elements that are sufficient to amount to significantly more than the judicial exception.
With respect to dependent claims 6 and 15, the claims do not include additional elements that are sufficient to amount to significantly more than the judicial exception.
With respect to dependent claims 7 and 16, the claims do not include additional elements that are sufficient to amount to significantly more than the judicial exception.
With respect to dependent claims 8 and 17, the claims do not include additional elements that are sufficient to amount to significantly more than the judicial exception.
With respect to dependent claims 9 and 18, the claims do not include additional elements that are sufficient to amount to significantly more than the judicial exception.
Conclusion — Rejection
Claims 1-20 are rejected under 35 U.S.C. § 101 as being directed to a judicial exception (mental processes) and failing to recite additional elements that amount to significantly more than the judicial exception.
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 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.
The text of those sections of Title 35, U.S. Code not included in this action can be found in a prior Office action.
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 the examiner presumes that the subject matter of the various claims was commonly owned as of the effective filing date of the claimed invention(s) absent any evidence to the contrary. Applicant is advised of the obligation under 37 CFR 1.56 to point out the inventor and effective filing dates of each claim that was not commonly owned as of the effective filing date of the later invention in order for the examiner to consider the applicability of 35 U.S.C. 102(b)(2)(C) for any potential 35 U.S.C. 102(a)(2) prior art against the later invention.
Claims 1-20 are rejected under pre-AIA 35 U.S.C. 103(a) as being unpatentable over Chen et al., (US 2026/0268124 A1, filed on 2025-Mar-10) in view of Cui et al., (US 2025/0139374 A1).
Regarding claim 1, Chen discloses a method of test-time resource scaling for neural network based language models, the method comprising:
receiving, via a data interface, a task request for a first neural network based language model to generate a response (Fig. 2, step 202, [0055] receiving a request for an output sequence);
generating, by the first neural network based language model, a plurality of candidate responses to the task request (Figs. 1 and 2, steps 204 and 206, [0017][0056][0059][0060] generating a plurality of candidate output sequences by a generative neural network);
iteratively allocating compute resources for the first neural network based language model to conduct one or more updated generations, [guided by a second neural network based language model] to iteratively evaluate at least the plurality of candidate responses, thereby resulting in a final response out of the plurality of candidate responses (Fig. 2, steps 208-214, [0061]-[0076] repeatedly performing one or more iterations for generating/updating the adjusted candidate output sequence by a generative neural network to outputting a selected candidate output sequence);
computing a metric indicating helpfulness of the second neural network model based language model to scale test-time performance of the first neural network language model (Figs. 2 and 3, [0075] “The reward model can be any machine learning model, e.g., any neural network, that has been configured through training to process an output sequence to generate a reward score for the output sequence that measures one or more properties of the output sequence”); and
deploying the second neural network language model in conjunction with the first neural network language model for test-time resource scaling ([0004]-[0007][0040] outperforming repeated sampling with reward model scoring for self-enhanced test-time/inference scaling of generative neural networks).
Chen does not explicitly teach however Cui does explicitly teach:
a/the second neural network based language model (Figs. 1A, 3, 4A, [0044] performing semantic consistency checks between the reference source (102) and the output (104) of the primary large language model by a criteria model (136, 308) is a large language model).
Therefore, it would have been obvious to one of ordinary skill before the effective filing date of the claimed invention to incorporate the method of self-enhanced test-time scaling of generative neural networks as taught by Chen with the method of adapting a criteria model, which may be referred to as a divide-conquer evaluator as taught by Cui to provide an architecture for an improved automated evaluation of large language models (Cui, [0010]).
Regarding claim 2, Chen in view of Cui discloses the method of claim 1, and Chen further discloses:
wherein the iteratively allocating compute resources for the first neural network based language model to conduct one or more updated generations comprises: generating, by the second neural network based language model, a ranking of the plurality of candidate responses; and selecting the final response from the ranking of the plurality of candidate responses (Chen, [0051][0072][0075] generating a reward score for the of candidate output sequence using a reward model and selecting a candidate output sequence from the plurality of candidate output sequences that has the highest reward score).
Regarding claim 3, Chen in view of Cui discloses the method of claim 1, and Chen further discloses:
wherein the plurality of candidate responses are partial responses to the task request, and the iteratively allocating compute resources for the first neural network based language model to conduct one or more updated generations comprises: selecting, [by the second neural network based language model], multiple candidate responses from the plurality of candidate responses; conducting a beam search by resampling an output distribution of the first neural network based language model to generate a new line continuing from the multiple candidate responses to form updated candidate responses, respectively; and iteratively generating updated candidate responses until a last full responses is generated and selected ([0042] generating multiple different initial candidate output sequences based on the same request, e.g., by using beam search decoding from score distributions generated by the generative neural network, using a Sample-and-Rank decoding strategy).
selecting, [by the second neural network based language model], multiple candidate responses from the plurality of candidate responses (Cui, Figs. 1A, 3, 4A, [0044] performing semantic consistency checks by a criteria model which is indicative of the “second …model”).
The previous motivation statement as in claim 1 is still applied.
Regarding claim 4, Chen in view of Cui discloses the method of claim 1, and Cui further discloses:
wherein the iteratively allocating compute resources for the first neural network based language model to conduct one or more updated generations comprises:
Cui does explicitly teach:
generating, by the second neural network based language model, corresponding critiques for each of the plurality of candidate responses (Cui, Fig. 2B, step 252, identifying an inconsistent sentence that is inconsistent with the reference source by a criteria model);
refining, by the first neural network based language model, a plurality of revised candidate responses based on the corresponding critiques (Fig. 2B, steps 252 and 254, [0101][0102] Repeating to return until the output satisfies a predetermined criteria output by the criteria model); and
selecting, by the second neural network based language model, the final response by ranking the plurality of candidate responses and the plurality of revised candidate responses together (Figs. 2A and 2B, steps 208, 256 and 258, [0037] generating the modified output (126) which may be a rewritten version of an inconsistent sentence (110), which is rewritten to be consistent with the reference source).
The previous motivation statement as in claim 1 is still applied.
Regarding claim 5, Chen in view of Cui discloses the method of claim 1, and Cui further discloses:
wherein the metric is computed based on an average performance of a greedily decoded response by the first neural network based language model, and an average performance of the final response selected by the second neural network based language model (Cui, [0019][0046][0070] generating a vector that represents quantitative assessments of the reasons (for consistency or inconsistency of the sentences in the primary model output) in a numeric score system and generating new candidate sentences (i.e., an inconsistent sentence (110)) based on the reason or reasons why the inconsistent sentence (110) was inconsistent with the reference source (102) and based on the original output sentence of the primary large language model (134)).
The previous motivation statement as in claim 1 is still applied.
Regarding claim 6, Chen in view of Cui discloses the method of claim 5, and Cui further discloses:
wherein the average performance is quantified by one or more of: numerical equality of parsed final answer; equivalence of parsed final answer; grading on final answer using another neural network based language model; whether a generated code passes a test; prompt-level accuracy (Cui, [0070] generates a first data structure including a first vector which stores, for each of the sentences, a corresponding evaluation of a given sentence as being consistent or inconsistent with the reference source, and a corresponding reason for the corresponding evaluation of the given sentence).
The previous motivation statement as in claim 1 is still applied.
Regarding claim 7, Chen in view of Cui discloses the method of claim 1, and Chen further discloses:
wherein the task request comprises one or more of math reasoning, code generation and instruction following ([0033] input sequence can comprise a natural language description of one or more tasks which may be the output sequence can comprise a sequence of instructions).
Regarding claim 8, Chen in view of Cui discloses the method of claim 1, and Cui further discloses:
wherein the deploying the second neural network language model in conjunction with the first neural network language model for test-time resource scaling is in response to the metric being greater than a threshold (Cui,
[0095][0107] the primary large language model output improvement process may continue until a threshold consistency score is reached and rewriting may be performed responsive to the metric satisfying a threshold value).
The previous motivation statement as in claim 1 is still applied.
Regarding claim 9, Chen in view of Cui discloses the method of claim 8, and Cui further discloses:
evaluating a third neural network based language model for test-time resource scaling with the first neural network based language model, in response to the metric being less than the threshold (Cui, [0091] retraining, responsive to the metric failing to satisfy a threshold, the primary large language model).
The previous motivation statement as in claim 1 is still applied.
Regarding claims 10-18, Claims 10-18 are the corresponding system claims to method claims 1-9. Therefore, claims 10-18 are rejected using the same rationale as applied to claims 1-9 above.
Regarding claims 19 and 20, Claims 19 and 20 are the corresponding medium claims to method claims 1 and 3. Therefore, claims 19 and 20 are rejected using the same rationale as applied to claims 1 and 3 above.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure. Please see attached form PTO-892.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to SEONG-AH A. SHIN whose telephone number is (571)272-5933. The examiner can normally be reached 9 AM-3PM.
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, Pierre-Louis Desir can be reached at 571-272-7799. 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.
Seong-ah A. Shin
Primary Examiner
Art Unit 2659
/SEONG-AH A SHIN/Primary Examiner, Art Unit 2659
Prosecution Insights