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 office action is responsive to the above application filed February 21, 2024.
Claims 1-20 are pending, all examined and rejected.
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 without significantly more.
With regard to claim 1,
Step 1, MPEP 2106.03:
These limitations have been determined, under Step 1, to be statutory categories of invention:
A computer-implemented method (CIM), the CIM comprising:
tuning a plurality of Large Language Models (LLMs) to debate one another to determine solutions for incidents;
causing data associated with a first incident to be input into the LLMs;
incorporating solutions output by the LLMs into a recommendation for resolving the first incident, wherein the recommendation weighs trade-offs of different possible resolutions for solving the first incident;
and outputting the recommendation to a user interface of a user device.
Step 2A, Prong 1:
This part of the eligibility analysis evaluates whether the claim recites a judicial exception. As
explained in MPEP 2106.04, subsection II, a claim “recites” a judicial exception when the judicial
exception is “set forth” or “described” in the claim.
Claim 1 recites:
A computer-implemented method (CIM), the CIM comprising:
tuning a plurality of Large Language Models (LLMs) to debate one another to determine solutions for incidents;
causing data associated with a first incident to be input into the LLMs;
incorporating solutions output by the LLMs into a recommendation for resolving the first incident, wherein the recommendation weighs trade-offs of different possible resolutions for solving the first incident;
and outputting the recommendation to a user interface of a user device.
The broadest reasonable interpretation of the bolded limitations above are directed to a mental process able to be performed in the human mind or by a human using pen and paper and organizing human activity, limited to the subgrouping: fundamental economic principles or practices, commercial or legal interactions, or managing personal behavior or relationships or interactions between people. Training people to debate with one another to determine solutions for the incidents is considered managing personal behavior. A human mind can incorporate solutions into a recommendation for resolving the incident mentally or with pen and paper. As such, the claim recites at least one abstract idea.
Step 2A , Prong 1 (Yes)
Step 2A, Prong 2:
Claim 1 recites:
A computer-implemented method (CIM), the CIM comprising:
tuning a plurality of Large Language Models (LLMs) to debate one another to determine solutions for incidents;
causing data associated with a first incident to be input into the LLMs;
incorporating solutions output by the LLMs into a recommendation for resolving the first incident, wherein the recommendation weighs trade-offs of different possible resolutions for solving the first incident; and
outputting the recommendation to a user interface of a user device.
This part of the eligibility analysis evaluates whether the claim as a whole integrates the recited judicial exception into a practical application of the exception or whether the claim is "directed to" the judicial exception. This evaluation is performed by (1) identifying whether there are any additional elements recited in the claim beyond the judicial exception, and (2) evaluating those additional elements individually and in combination to determine whether the claim as a whole integrates the exception into a practical application. See MPEP 2106.04(d).
The additional element of “causing data associated with a first incident to be input into the LLMs” amount no more than mere data gathering. See MPEP 2106.05(g).
The additional element of “tuning a plurality of large language models (LLMs)”, “output by the LLMs”, “the recommendation weights trade-offs of different possible for solving the first incident”, “outputting the recommendation to a user interface of a user device” amount no more than mere instructions to implement the abstract idea. See MPEP 2106.05(g).
Step 2A, Prong 2 (No).
Step 2B:
Claim 1 recites:
A computer-implemented method (CIM), the CIM comprising:
tuning a plurality of Large Language Models (LLMs) to debate one another to determine solutions for incidents;
causing data associated with a first incident to be input into the LLMs;
incorporating solutions output by the LLMs into a recommendation for resolving the first incident, wherein the recommendation weighs trade-offs of different possible resolutions for solving the first incident; and
outputting the recommendation to a user interface of a user device.
This part of the eligibility analysis evaluates whether the claim as a whole amount to significantly more than the recited exception i.e., whether any additional element, or combination of additional elements, adds an inventive concept to the claim. See M PEP 2106.05.
As explained with respect to Step 2A, the additional element is “causing data associated with a first incident to be input into the LLMs” which at best is insignificant mere data gathering. See MPEP 2106.05(g).
Additionally, the additional elements of “tuning a plurality of large language models (LLMs)”, “output by the LLMs”, “the recommendation weights trade-offs of different possible for solving the first incident”, “outputting the recommendation to a user interface of a user device” which at best are insignificant instructions to implement the abstract idea. See MPEP 2106.05(g).
Step 2B (No)
Claim 1 is ineligible.
With respect to claims 10 and 19,
These claims are similar in scope to claim 1 and are rejected under a similar rationale.
Dependent Claims:
Claims 3, 5, 12, 14 recite further mere data gathering (“training data is used to tune”, “some of the data associated with a first incident is received in a question from a user device”) and as explained above these do not provide a practical application or inventive concept and thus are ineligible.
Claims 4, 6-9, 13, 15-18 recite further mere instructions to implement the abstract idea () and as explained above these do not provide a practical application or inventive concept and thus are ineligible.
Claims 2, 11, 20: The claim recites “encoder models”, “encoder-decoder models”, “decoder-only models”.
With respect to Step 2A Prong 2, these elements recited at a high level of generality and thus are generic computer components performing computer functions. See MPEP 2106.05(f).
With respect to Step 2B, these elements of “encoder models”, “encoder-decoder models”, “decoder-only models” which at best are insignificant extra-solution activity as recited at a high level of generality. See MPEP 2106.05(d).
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 (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 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 1, 5, 10, 14, 19 are rejected under 35 U.S.C 102(a)(1) as being anticipated by Xiong, Kai, et al. "Examining inter-consistency of large language models collaboration: An in-depth analysis via debate." Findings of the association for computational linguistics: EMNLP 2023. 2023. [Xiong]
Regarding claim 1,
Xiong teaches A computer-implemented method (CIM), the CIM comprising:
tuning a plurality of Large Language Models (LLMs) to debate one another to determine solutions for incidents; (See Page 1, column 2, Introduction “In specific, we take multi-choice commonsense reasoning as the example task as it can accurately quantify the inter-inconsistency of LLMs collaboration. Then we formulate a three-stage debate to align with real-world scenarios: (1) Fair debate between two LLMs with comparable capabilities. (2) Mismatched debate between two LLMs who exhibit vastly different levels of abilities. (3) Roundtable debate including more than two LLMs for debates. See Page 1, Column 2, Fig. 1, See Page 18, Prompt 13 and Prompt 14 – tuning the LLMs using prompt-tuning and choosing models to debate each other.)
causing data associated with a first incident to be input into the LLMs; (See Page 2, column 2, Table 1, Commonsense reasoning dataset statistics – The datasets associated with the tasks are being used as input to LLMs.)
incorporating solutions output by the LLMs into a recommendation for resolving the first incident, wherein the recommendation weighs trade-offs of different possible resolutions for solving the first incident; (See Page 3, Figure 2. See Page 5, column 1, FORD defeats Col-S, Col-H, and corresponding single LLMs on almost all datasets (except for LLaMA & Vicuna on Social IQa). It is because FORD can make LLMs obtain more comprehensive and precise perspectives of the question. This signifies that LLMs with comparable abilities possess a spirit of collaboration to effectively and performantly achieve a shared goal. See Page 13, Prompt 3 – The LLMs collaborate with each other to incorporate the resolution/achieve a shard goal. Fig. 2 step 2 shows how different solutions are weighed.)
and outputting the recommendation to a user interface of a user device.(See Page 11, column 2, Prompt 1, Prompt 2 – the answer/resolution is displayed to user through a user device.)
With respect to claim 5,
As discussed with regard to claim 1, Xiong teaches all the limitations.
Xiong further teaches the incident is a security incident, where at least some of the data associated with a first incident is received in a question from a user device. (See Page 18, Prompt 13, 14. – The user gives information/data in the question when asking the LLMs from user device.)
Regarding claims 10, 19,
These claims are similar in scope to claim 1 and are rejected under a similar rationale.
Regarding claim 14,
The claim is similar in scope to claim 5 and is rejected under a similar rationale.
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 2, 11, 20 are rejected under 35 U.S.C 103 as being unpatentable over Xiong in view of Ferguson, U.S PG Pub 20260221231, filed January 09, 2024.
With respect to claim 2,
As discussed with regard to claim 1, Xiong teaches all of the limitations.
Xiong does not explicitly teach the plurality of LLMs each have different model architectures selected from the group consisting of: encoder models, encoder-decoder models, and decoder-only models.
Ferguson teaches the plurality of LLMs each have different model architectures selected from the group consisting of: encoder models, encoder-decoder models, and decoder-only models. (See Fig. 1A. See [0040] – The LLM has a family-specific encoder and decoder, a layer of transformer encoder and decoder. The LLM also has a layer of encoder only, and a layer of decoder only.)
It would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention to combine the layers of decoders and encoders in the architecture of the LLM with the LLMs as taught by Xiong. One would be motivated to do so to accurately expand and reproduce unaligned protein sequences. [Ferguson, 0041].
With respect to claim 11, 20,
These claims are similar in scope to claim 2 and are rejected under a similar rationale.
Claim 3-4, 12-13 are rejected under 35 U.S.C as being unpatentable over Xiong in view of Gupta, Maanak, et al. "From chatgpt to threatgpt: Impact of generative ai in cybersecurity and privacy." IEEE access 11 (2023): 80218-80245.
With respect to claim 3,
As discussed with regard to claim 1, Xiong teaches all of the limitations.
Xiong further teaches training data is used to tune the LLMs to debate one another. (See Page 2, column 2, table 1. – The datasets are being used to train the LLMs so they can debate each other.)
Xiong does not explicitly teach the training data is selected from the group consisting of: security data including past incident data, vulnerability, and mitigation data, and cyberthreat intelligence data.
Gupta teaches the training data is selected from the group consisting of: security data including past incident data, vulnerability, and mitigation data, and cyberthreat intelligence data. (See page 3, column 1, B. Impact of GenAI in cybersecurity and privacy, “These tools leverage the information from LLMs trained on the massive amount of cyber threat intelligence data that includes vulnerabilities, attack patterns, and indications of attack. Cyber defenders can use this large sum of information to enhance their threat intelligence capability by extracting insights and identifying emerging threats” – The LLM is trained on vulnerabilities, attack patterns, and cyber threat intelligence data.)
It would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention to combine security, cyberthreat intelligence, mitigation data as taught by Gupta with the ability to train the LLMs using data as taught by Xiong. One would be motivated to do so to enhance the threat intelligence capability of the LLMs. [Gupta]
With respect to claim 4,
As discussed with regard to claim 3, Xiong and Gupta teach all the limitations.
Xiong further teaches the tuning utilizes prompt-tuning techniques. (See at least Page 18, Prompt 12-14 – User input information to the LLMs for training/tuning)
With respect to claims 12-13,
These claims are similar in scope to claims 3-4 and are rejected under a similar rationale.
Claim 6-7, 15-16 are rejected under 35 U.S.C as being unpatentable over Xiong in view of Kratzwald, Bernhard, and Stefan Feuerriegel. "Learning from on-line user feedback in neural question answering on the web." The World Wide Web Conference. 2019.
With respect to claim 6,
As discussed with regard to claim 1, Xiong teaches all the limitations.
Xiong does not explicitly teach receiving, from the user device, feedback about the recommendation for resolving the first incident; analyzing the feedback to determine whether the feedback is positive feedback or negative feedback; in response to a determination that the feedback is positive feedback, providing a reward to at least some of the LLMs; and in response to a determination that the feedback is negative feedback, providing the negative feedback to at least some of the LLMs.
Kratzwald teaches receiving, from the user device, feedback about the recommendation for resolving the first incident; analyzing the feedback to determine whether the feedback is positive feedback or negative feedback; in response to a determination that the feedback is positive feedback, providing a reward to at least some of the LLMs; and in response to a determination that the feedback is negative feedback, providing the negative feedback to at least some of the LLMs. (See Page 4, column 1, 3.1.2 Feedback Collection “The QA system takes a question q as input and then returns a triplet ⟨q, a,p⟩, where a refers to the answer and p to the surrounding paragraph that can be used for assessing the fit (see the example presentation in Fig. 1). Let ψ denote the (optional) feedback, i. e., an up-vote or a down-vote”. See Page 5, column 1, Algorithm 1. – The up-down vote system determines/analyze whether the feedback is negative or positive for the neural network/LLM neural network. The negative feedback is provided to the neural network/LLM neural network and the neural network/ LLM neural network have to refine the displayed answer. If positive feedback is given, an improved paragraph is generated and the dataset is updated with the new paragraph, the new dataset is then being used for training/providing reward for the LLM neural network)
It would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention to combine feedback system, providing reward and negative feedback for the neural network as taught by Kratzwald with the ability to let LLM neural networks debating each other as taught by Xiong. One would be motivated to do so to achieve dynamic learning based on user interactions. [Kratzwald, Page 3, column 2, 3.1.1 Motivation behind shallow user feedback.]
With respect to claim 7,
As discussed with regard to claim 6, Xiong and Kratzwald teach all the limitations.
Xiong further teaches assigning weights to the LLMs, wherein each of the weights establish an extent of influence that an associated LLM has while debating the other LLMs to determine solutions for incidents. (See page 4, column 1, 3.3. Debate summarization “For samples that reached a consensus, the conclusion is the consensus stance. For samples without a consensus, we assign equal weights to all arguments for the conclusion. Refer to Appendix E for more details.”. See page 7, column 2, “GPT-4 as the judge can further boost the performance of FORD. It is mainly because GPT- 4 can assign higher weights to more convincing arguments, then draw more precise conclusions.” – weighs are assigned to the LLMs. More weights mean the model can have more convincing arguments/more influence.)
With respect to claim 15,
The claim is similar in scope to claim 6 and is rejected under a similar rationale.
With respect to claim 16,
The claim is similar in scope to claim 7 and is rejected under a similar rationale.
Claim 8, 17 are rejected under 35 U.S.C 103 as being unpatentable over Xiong and Kratzwald in view of Medalion, U.S PG Pub 20230394226, filed June 01, 2022.
With respect to claim 8,
As discussed with regard to claim 7, Xiong and Kratzwald teach all the limitations.
Xiong further teaches determining the LLMs that debated for including the solutions output by the LLMs. (See at least Figure 1, 2. See page 17. Prompt 10-13. – The LLMs that debate for including the solutions are shown for display.)
Xiong and Kratzwald do not explicitly teach decreasing the weights assigned to the determined LLMs a predetermined amount.
Medalion teaches decreasing the weights assigned to the determined LLMs a predetermined amount. (See [0087] and [0101] – decreasing the weights an amount of score according to a scoring or weighting algorithm/predetermined amount when the model performs poorly.)
It would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention to combine decreasing the weights of a LLM neural network for a predetermined amount when performs poorly as taught by Medalion with the feedback system that can gives negative feedback to LLM neural network as taught by Kratzwald. One would be motivated to do so to train the mode further. [Medalion, 0087]
With respect to claim 17,
The claim is similar in scope to claim 8 and is rejected under a similar rationale.
Claim 9, 18 are rejected under 35 U.S.C 103 as being unpatentable over Xiong in view of Zhang, Libo, et al. "A recommendation model based on deep neural network." IEEE Access 6 (2018): 9454-9463.
With respect to claim 9,
As discussed with regard to claim 9, Xiong teaches all the limitations.
Xiong does not explicitly teach determining a preferred one of the resolutions, wherein the preferred resolution is highlighted within the recommendation.
Zhang teaches determining a preferred one of the resolutions, wherein the preferred resolution is highlighted within the recommendation. (See Page 6, column 1, D. Making recommendations “When the training of the proposed model is completed, we can use it to predict a user's rating score on the items that have not been rated by the user. When making recommendations for a specific user, we can recommend the items with the highest predicted score for the user.” – The recommendation system recommends the items to user based on the predicted score, and the items/resolutions that have highest scores/preferred will be recommended to the user.)
It would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention to recommend the solution with highest recommending score as taught by Zhang with the recommending resolutions getting from the LLMs debating as taught by Xiong. One would be motivated to do so the user can see the best resolution from the LLMs.
With respect to claim 18,
The claim is similar in scope to claim 9 and is rejected under a similar rationale.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant’s disclosure.
Smit, Andries, et al. "Should we be going mad? a look at multi-agent debate strategies for llms." arXiv preprint arXiv:2311.17371 (2023).
Kaheh, Mehrdad, Danial Khosh Kholgh, and Panos Kostakos. "Cyber sentinel: Exploring conversational agents in streamlining security tasks with gpt-4." arXiv preprint arXiv:2309.16422 (2023).
Any inquiry concerning this communication or earlier communications from the examiner should be directed to MICHAEL HOANG LE whose telephone number is (571)270-7292. The examiner can normally be reached Monday-Friday 8:00 am - 5 pm.
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, Matthew Ell can be reached at 5712703264. 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.
/M.H.L./Examiner, Art Unit 2141
/MATTHEW ELL/Supervisory Patent Examiner, Art Unit 2141
Prosecution Insights