DETAILED ACTION
1. This office action is in response to the Application No. 18643797 filed on 04/23/2024. Claims 1-21 are presented for examination and are currently pending.
Notice of Pre-AIA or AIA Status
2. The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
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.
3. Claims 1-21 are rejected under 35 U.S.C 101 because the claimed invention is directed towards an abstract idea without significantly more.
Step 1
Independent claim 1 is directed to a method, and falls into one of the four statutory categories.
Step 2A, Prong 1
Claim 1 recites the following abstract ideas:
decomposing the compound prompt into a plan having a plurality of steps (Mental process directed to decomposing the compound prompt into a plan having a plurality of steps. This can be done with the use of a pen and paper)
mapping a domain to each of a plurality of distinct machine learning models (Mental process directed to mapping a domain to each of a plurality of distinct machine learning models. This can be done by observing the plurality of distinct machine learning models and making a judgement on mapping a domain to the distinct machine learning models);
for each of the plurality of steps: selecting one of the plurality of distinct machine learning models by matching the step to the corresponding mapped domain (Mental process directed to selecting one of the plurality of distinct machine learning models by matching the step to the corresponding mapped domain. This can be done by observing the plurality of distinct machine learning models and making a judgement on the matching to select the appropriate machine learning model);
Step 2A, Prong 2
Claim 1 recites the following additional elements:
via a planner module instantiated in machine-readable memory and operable by a processor (This is directed to mere instructions to apply the judicial exception in a computer component. This limitation does not integrate the abstract idea into a practical application. See MPEP 2106.05(f)));
and generating a language output using the selected one of the distinct machine learning models (This is directed to mere instructions to apply a judicial exception. This limitation does not integrate the abstract idea into a practical application. See MPEP 2106.05(f));
and integrating the language outputs of each of the plurality of steps into a syntactically and semantically coherent final output via an integration module utilizing a large language model (This is amount to generally linking the use of a judicial exception to a particular technological environment or field of use. This limitation does not integrate the abstract idea into a practical application. See MPEP 2106.05(h))).
Step 2B
Claim 1 recites the following additional elements:
via a planner module instantiated in machine-readable memory and operable by a processor (This is directed to mere instructions to apply the judicial exception in a computer component. This limitation does not amount to significantly more than the judicial exception, See MPEP 2106.05 (f));
and generating a language output using the selected one of the distinct machine learning models (This is directed to mere instructions to apply a judicial exception. This limitation does not amount to significantly more than the judicial exception, See MPEP 2106.05 (f));
and integrating the language outputs of each of the plurality of steps into a syntactically and semantically coherent final output via an integration module utilizing a large language model (This is amount to generally linking the use of a judicial exception to a particular technological environment or field of use. This limitation does not amount to significantly more than the judicial exception. See MPEP 2106.05 (h)).
4. Dependent claim 2 is directed to a method, and falls into one of the four statutory categories.
Claim 2 do not recite any abstract ideas.
Claim 2 recites the following additional elements:
further comprising training each of the plurality of distinct machine learning models using different training data specific to its respective domain (This limitation is directed to mere instructions to apply an exception. This limitation does not integrate the abstract idea into a practical application. See MPEP 2106.05(f))).
Claim 2 recites the following additional elements:
further comprising training each of the plurality of distinct machine learning models using different training data specific to its respective domain (This limitation is directed to mere instructions to apply an exception. This limitation does not amount to significantly more than the judicial exception. See MPEP 2106.05(f))).
5. Dependent claim 3 is directed to a method, and falls into one of the four statutory categories.
Claim 3 do not recite any abstract ideas.
Claim 3 recites the following additional elements:
wherein at least a subset of the plurality of distinct machine learning models are trained entirely separately from others of the plurality of distinct machine learning models, without overlapping training data (This limitation is directed to generally linking the use of a judicial exception to a particular technological environment or field of use. This limitation does not integrate the abstract idea into a practical application. See MPEP 2106.05(h)).
Claim 3 recites the following additional elements:
wherein at least a subset of the plurality of distinct machine learning models are trained entirely separately from others of the plurality of distinct machine learning models, without overlapping training data (This limitation is directed to generally linking the use of a judicial exception to a particular technological environment or field of use. This limitation does not amount to significantly more than the judicial exception. See MPEP 2106.05(h)).
6. Dependent claim 4 is directed to a method, and falls into one of the four statutory categories.
Claim 4 do not recite any abstract ideas.
Claim 4 recites the following additional elements:
wherein at least a subset of the plurality of distinct machine learning models are specialized in a respective domain via fine-tuning or transfer learning (This limitation is directed to generally linking the use of a judicial exception to a particular technological environment or field of use. This limitation does not integrate the abstract idea into a practical application. See MPEP 2106.05(h)).
Claim 4 recites the following additional elements:
wherein at least a subset of the plurality of distinct machine learning models are specialized in a respective domain via fine-tuning or transfer learning (This limitation is directed to generally linking the use of a judicial exception to a particular technological environment or field of use. This limitation does not amount to significantly more than the judicial exception. See MPEP 2106.05(h))
7. Dependent claim 5 is directed to a method, and falls into one of the four statutory categories.
Claim 5 recite the following abstract ideas:
wherein mapping the domain to each of the plurality of distinct machine learning models comprises mapping a subject-matter specialization to each of the plurality of distinct machine learning models (Mental process directed to mapping each domain to each of the plurality of distinct machine learning models. This can be done by observing the domain and making a judgement on mapping the domain the distinct machine learning models).
Claim 5 do not recite any additional elements.
8. Dependent claim 6 is directed to a method, and falls into one of the four statutory categories.
Claim 6 do not recite any abstract ideas.
Claim 6 recite the following additional elements:
wherein mapping the domain to each of the plurality of distinct machine learning models comprises training a selection model via machine learning to map compound prompts to one or more of the plurality of distinct machine learning models (This limitation is directed to mere instructions to implement a judicial exception. This limitation does not integrate the abstract idea into a practical application. See MPEP 2106.05(f)).
Claim 6 recite the following additional elements:
wherein mapping the domain to each of the plurality of distinct machine learning models comprises training a selection model via machine learning to map compound prompts to one or more of the plurality of distinct machine learning models (This limitation is directed to mere instructions to implement a judicial exception. This limitation does not amount to significantly more than the judicial exception. See MPEP 2106.05(f)).
9. Dependent claim 7 is directed to a method, and falls into one of the four statutory categories.
Claim 7 do not recite any abstract ideas.
Claim 7 recites the following additional elements:
wherein generating a language output using the selected one of the distinct machine learning models comprises providing the selected one of the plurality of distinct machine learning models with at least a portion of the complex prompt, one of the language outputs from another of the plurality of steps, or both (This limitation is directed to generally linking the use of a judicial exception to a particular technological environment or field of use. This limitation does not integrate the abstract idea into a practical application. See MPEP 2106.05(h)).
Claim 7 recites the following additional elements:
wherein generating a language output using the selected one of the distinct machine learning models comprises providing the selected one of the plurality of distinct machine learning models with at least a portion of the complex prompt, one of the language outputs from another of the plurality of steps, or both (This limitation is directed to generally linking the use of a judicial exception to a particular technological environment or field of use. This limitation does not amount to significantly more than the judicial exception. See MPEP 2106.05(h)).
10. Dependent claim 8 is directed to a method, and falls into one of the four statutory categories.
Claim 8 do not recite any abstract ideas.
Claim 8 recites the following additional elements:
wherein generating each language output for each of the plurality of steps after the first comprises providing the selected one of the plurality of machine learning models with a prompt including an output of one or more preceding steps (This is amount to generally linking the use of a judicial exception to a particular technological environment or field of use. This limitation does not integrate the abstract idea into a practical application. See MPEP 2106.05(h)).
Claim 8 recites the following additional elements:
wherein generating each language output for each of the plurality of steps after the first comprises providing the selected one of the plurality of machine learning models with a prompt including an output of one or more preceding steps (This is amount to generally linking the use of a judicial exception to a particular technological environment or field of use. This limitation does not amount to significantly more than the judicial exception. See MPEP 2106.05(h)).
11. Dependent claim 9 is directed to a method, and falls into one of the four statutory categories.
Claim 9 do not recite any abstract ideas.
Claim 9 recite the following additional elements:
wherein at least a subset of the plurality of machine learning models are large language models (This limitation is directed to generally linking the use of a judicial exception to a particular technological environment or field of use. This limitation does not integrate the abstract idea into a practical application. See MPEP 2106.05(h)).
Claim 9 recite the following additional elements:
wherein at least a subset of the plurality of machine learning models are large language models (This limitation is directed to generally linking the use of a judicial exception to a particular technological environment or field of use. This limitation does not amount to significantly more than the judicial exception. See MPEP 2106.05(h)).
12. Dependent claim 10 is directed to a method, and falls into one of the four statutory categories.
Claim 10 do not recite any abstract ideas.
Claim 10 recite the following additional elements:
wherein the large language model is one of the plurality of distinct machine learning models having a corresponding generalist domain (This limitation is directed to generally linking the use of a judicial exception to a particular technological environment or field of use. This limitation does not integrate the abstract idea into a practical application. See MPEP 2106.05(h))).
Claim 10 recite the following additional elements:
wherein the large language model is one of the plurality of distinct machine learning models having a corresponding generalist domain ((This limitation is directed to generally linking the use of a judicial exception to a particular technological environment or field of use. This limitation does not amount to significantly more than the judicial exception. See MPEP 2106.05(h))).
13. Dependent claim 11 is directed to a method, and falls into one of the four statutory categories.
Claim 11 do not recites any abstract ideas.
Claim 11 recite the following additional elements:
wherein the selection of one of the plurality of distinct machine learning models for each step is performed by the large language model (This limitation is directed to instructions to implement the judicial exception. This limitation does not integrate the abstract idea into a practical application. See MPEP 2106.05(f)).
Claim 11 recite the following additional elements:
wherein the selection of one of the plurality of distinct machine learning models for each step is performed by the large language model (This limitation is directed to instructions to implement the judicial exception. This limitation does not amount to significantly more than the judicial exception. See MPEP 2106.05(f)).
14. Dependent claim 12 is directed to a method, and falls into one of the four statutory categories.
Claim 12 recites the following abstract ideas:
wherein decomposing the compound prompt into a plan having a plurality of steps comprises identifying the plurality of steps, ordering the plurality of steps, and identifying outputs from at least one of the plurality of steps to be received as inputs by another of the plurality of steps (Mental process directed to identifying, ordering of steps and identifying outputs. This can be done by observing the compound prompt and making a judgement on the identification and ordering of plurality of steps).
Claim 12 do not recite any additional elements.
15. Dependent claim 13 is directed to a method, and falls into one of the four statutory categories.
Claim 13 do not recite any abstract ideas.
Claim 13 recites the following additional elements:
wherein at least some of the plurality of steps are executed sequentially (This limitation is directed to generally linking the use of a judicial exception to a particular technological environment or field of use. This limitation does not integrate the abstract idea into a practical application. See MPEP 2106.05(h)).
Claim 13 recites the following additional elements:
wherein at least some of the plurality of steps are executed sequentially (This limitation is directed to generally linking the use of a judicial exception to a particular technological environment or field of use. This limitation does not amount to significantly more than the judicial exception. See MPEP 2106.05(h)).
16. Independent claim 14 is directed to a system, and falls into one of the four statutory categories.
Claim 14 recites the following abstract ideas:
to decompose the complex prompt into a plan having a plurality of steps (Mental process directed to decomposing the compound prompt into a plan having a plurality of steps. This can be done by observing the complex prompt and making a judgement on the decomposition of the complex prompt);
to select one of the plurality of specialized LLMs to execute each of the plurality of steps (Mental process directed to selecting one of the plurality of specialized LLMs. This can be done by observing the plurality of specialized LLMs and making a judgement on the selection of specialized LLMs);
Claim 14 recites the following additional elements:
an input device (This limitation is directed to a computer component. This is directed to high level recitation of generic computer component and it does not integrate the abstract idea into a practical application. See MPEP 2106.05(f)) configured
to receive the complex prompt (This limitation is directed to insignificant extra solution activity of data transmission. This limitation does not integrate the abstract idea into a practical application. See MPEP 2106.05(g));
a logic processor (This limitation is directed to a computer component. This is directed to high level recitation of generic computer component and it does not integrate the abstract idea into a practical application. See MPEP 2106.05(f));
machine-readable memory (This limitation is directed to computer software. This is directed to high level recitation of generic computer component. This does not integrate the abstract idea into a practical application. See MPEP 2106.05(f));
a plurality of specialized large language models (LLMs) instantiated in the machine-readable memory (This limitation is directed to mere instructions to apply a judicial exception. This does not integrate the abstract idea into a practical application. See MPEP 2106.05(f)),
each of the specialized LLMs having a corresponding domain of specialization (This limitation is directed to generally linking the use of a judicial exception to a particular technological environment or field of use. This does not integrate the abstract idea into a practical application. See MPEP 2106.05(h)); and
a manager comprising: a planner instantiated in the machine-readable memory and operable (This limitation is directed to mere instructions to apply a judicial exception. This does not integrate the abstract idea into a practical application. See MPEP 2106.05(f))
via the logic processor (This limitation is directed to a computer component. This is directed to high level recitation of generic computer component and it does not integrate the abstract idea into a practical application. See MPEP 2106.05(f))
a selection module instantiated in the machine-readable memory and operable (This limitation is directed to mere instructions to apply a judicial exception. This does not integrate the abstract idea into a practical application. See MPEP 2106.05(f))
and an integration module instantiated in the machine-readable memory and operable via the logic processor to generate a language output responsive to the complex prompt from outputs of each of the selected ones of the plurality of specialized LLMs (This limitation is directed to mere instructions to apply a judicial exception. This does not integrate the abstract idea into a practical application. See MPEP 2106.05(f)).
Claim 14 recites the following additional elements:
an input device (This limitation is directed to a computer component. This is directed to high level recitation of generic computer component. This limitation does not amount to significantly more than the judicial exception. See MPEP 2106.05(f)) configured
to receive the complex prompt (This limitation is directed to insignificant extra solution activity of data transmission and it is well understood routine and conventional. This does not amount to significantly more than judicial exception. See MPEP 2106.05(d)(II), example i);
a logic processor (This limitation is directed to a computer component. This is directed to high level recitation of generic computer component .This does not amount to significantly more than judicial exception. See MPEP 2106.05(f));
machine-readable memory (This limitation is directed to computer software. This is directed to high level recitation of generic computer component. This does not amount to significantly more than judicial exception. See MPEP 2106.05(f));
a plurality of specialized large language models (LLMs) instantiated in the machine-readable memory (This limitation is directed to mere instructions to apply a judicial exception. This does not amount to significantly more than judicial exception. See MPEP 2106.05(f)),
each of the specialized LLMs having a corresponding domain of specialization (This limitation is directed to generally linking the use of a judicial exception to a particular technological environment or field of use. This does not amount to significantly more than judicial exception. See MPEP 2106.05(h)); and
a manager comprising: a planner instantiated in the machine-readable memory and operable (This limitation is directed to mere instructions to apply a judicial exception. This does not amount to significantly more than judicial exception. See MPEP 2106.05(f))
via the logic processor (This limitation is directed to a computer component. This is directed to high level recitation of generic computer component. This does not amount to significantly more than judicial exception. See MPEP 2106.05(f))
a selection module instantiated in the machine-readable memory and operable (This limitation is directed to mere instructions to apply a judicial exception. This does not amount to significantly more than judicial exception. See MPEP 2106.05(f))
and an integration module instantiated in the machine-readable memory and operable via the logic processor to generate a language output responsive to the complex prompt from outputs of each of the selected ones of the plurality of specialized LLMs (This limitation is directed to mere instructions to apply a judicial exception. This does not amount to significantly more than judicial exception. See MPEP 2106.05(f)).
17. Dependent claim 15 is directed to a system, and falls into one of the four statutory categories.
Claim 15 do not recite any abstract ideas.
Claim 15 recites the additional elements:
wherein the manager further comprises a generalist LLM (This limitation is directed to generally linking the use of a judicial exception to a particular technological environment or field of use. See MPEP 2106.05(h)), and
wherein the integration module generates the language output from outputs of at least a subset of the selected ones of the plurality of specialized large language models using the generalist large language model (This limitation is directed to mere instructions to apply a judicial exception. This does not integrate the abstract idea into a practical application. See MPEP 2106.05(f)).
Claim 15 recites the additional elements:
wherein the manager further comprises a generalist LLM (This limitation is directed to generally linking the use of a judicial exception to a particular technological environment or field of use. This does not amount to significantly more than judicial exception. See MPEP 2106.05(h)), and
wherein the integration module generates the language output from outputs of at least a subset of the selected ones of the plurality of specialized large language models using the generalist large language model (This limitation is directed to mere instructions to apply a judicial exception. This does not amount to significantly more than judicial exception. See MPEP 2106.05(f)).
16. Dependent claim 16 is directed to a system, and falls into one of the four statutory categories.
Claim 16 do not recite any abstract ideas.
Claim 16 recites the following additional elements:
wherein the generalist LLM is a Meta-Language Model (This limitation is directed to generally linking the use of a judicial exception to a particular technological environment or field of use. This does not integrate the abstract idea into a practical application. See MPEP 2106.05(h)).
Claim 16 recites the following additional elements:
wherein the generalist LLM is a Meta-Language Model (This limitation is directed to generally linking the use of a judicial exception to a particular technological environment or field of use. This does not amount to significantly more than judicial exception. See MPEP 2106.05(h)).
17. Dependent claim 17 is directed to a system, and falls into one of the four statutory categories.
Claim 17 recites the following abstract ideas:
maps each of the plurality of steps to one of the plurality of specialized large language models (Mental process directed to mapping each of the plurality of steps to one of the plurality of specialized large language models. This can be done by observing the plurality of steps and making a judgement on the mapping of the steps to the specialized large language models)
Claim 17 recites the following additional elements:
wherein the selection module (This limitation is directed to mere instructions to apply a judicial exception. This does not integrate the abstract idea into a practical application. See MPEP 2106.05(f))
using the generalist large language model (This limitation is directed to mere instructions to apply a judicial exception. This does not integrate the abstract idea into a practical application. See MPEP 2106.05(f)).
Claim 17 recites the following additional elements:
wherein the selection module (This limitation is directed to mere instructions to apply a judicial exception. This does not amount to significantly more than judicial exception. See MPEP 2106.05(f))
using the generalist large language model (This limitation is directed to mere instructions to apply a judicial exception. This does not amount to significantly more than judicial exception. See MPEP 2106.05(f)).
18. Dependent claim 18 is directed to a system, and falls into one of the four statutory categories.
Claim 18 recites the following abstract ideas:
identifying the corresponding domain of specialization and at least one of an input format and an output format for each of the plurality of specialized large language models (Mental process directed to identifying the corresponding domain of specialization, input and output format of the LLMs. This can be done by observing and making a judgement on the identification of the domain of specialization, input format and an output format of the specialized large language models).
Claim 18 recites the following additional elements:
wherein the selection module comprises a model record (This limitation is directed to a particular type or source of data, which is field of use. This does not integrate the abstract idea into a practical application. See MPEP 2106.05(h))
Claim 18 recites the following additional elements:
wherein the selection module comprises a model record (This limitation is directed to a particular type or source of data, which is field of use. This does not amount to significantly more than judicial exception. See MPEP 2106.05(h))
19. Dependent claim 19 is directed to a system, and falls into one of the four statutory categories.
Claim 19 do not recite any abstract ideas.
Claim 19 recites the following additional elements:
further comprising a database communicatively coupled with at least one of plurality of specialized large language models to provide context injection for that respective specialized large language model (This limitation is directed to generally linking the use of a judicial exception to a particular technological environment or field of use. This does not integrate the abstract idea into a practical application. See MPEP 2106.05(h)).
Claim 19 recites the following additional elements:
further comprising a database communicatively coupled with at least one of plurality of specialized large language models to provide context injection for that respective specialized large language model (This limitation is directed to generally linking the use of a judicial exception to a particular technological environment or field of use. This does not amount to significantly more than judicial exception. See MPEP 2106.05(h)).
20. Dependent claim 20 is directed to a system, and falls into one of the four statutory categories.
Claim 20 do not recite any abstract ideas.
Claim 20 recites the following additional elements:
wherein each of the plurality of specialized large language models is trained for its respective domain using different training data and/or parameters than all others of the plurality of specialized large language models (This limitation is directed to generally linking the use of a judicial exception to a particular technological environment or field of use. This does not integrate the abstract idea into a practical application. See MPEP 2106.05(h)).
Claim 20 recites the following additional elements:
wherein each of the plurality of specialized large language models is trained for its respective domain using different training data and/or parameters than all others of the plurality of specialized large language models (This limitation is directed to generally linking the use of a judicial exception to a particular technological environment or field of use. This does not amount to significantly more than judicial exception. See MPEP 2106.05(h)).
21. Dependent claim 21 is directed to a system, and falls into one of the four statutory categories.
With regards to claim 21, it is substantially similar to claim 4, and is rejected in the same manner and reasoning applying.
Claim Rejections - 35 USC § 112
The following is a quotation of 35 U.S.C. 112(b):
(b) CONCLUSION.—The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the inventor or a joint inventor regards as the invention.
The following is a quotation of 35 U.S.C. 112 (pre-AIA ), second paragraph:
The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the applicant regards as his invention.
4. Claim 9 is rejected under 35 U.S.C. 112(b) or 35 U.S.C. 112 (pre-AIA ), second paragraph, as being indefinite for failing to particularly point out and distinctly claim the subject matter which the inventor or a joint inventor (or for applications subject to pre-AIA 35 U.S.C. 112, the applicant), regards as the invention.
Claim 9 recites “wherein at least a subset of the plurality of machine learning models are large language models” which lacks antecedent basis. It is not clear which plurality of machine learning models “the plurality of machine learning models” is referring to.
Claim 9 depends on claim 1 and claim 1 recites “a plurality of distinct machine learning models” and not “a plurality of machine learning models”.
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.
5. Claims 1-21 are rejected under 35 U.S.C. 103 as being unpatentable over
Madisetti et al. (US12001462 filed 07/07/2023) in view of Balasubramaniam et al. (US12626695 filed 09/19/2023)
Regarding claim 1, Madisetti teaches a method of processing a compound prompt (A method of answering queries from user prompts, abstract; The prompt is sent to AI Input Broker 810 which generates multiple derived prompts for different categories 924, col. 7, lines 58-60, Fig. 10;The Examiner notes the prompt is a compound prompt because it generates multiple derived prompts and the instant specification discloses: “The term “compound prompt” can refer to a prompt explicitly including multiple separate tasks” [0014]),
the method comprising: decomposing the compound prompt into a plan having a plurality of steps via a planner module (The prompt is sent to AI Input Broker 810 which generates multiple derived prompts for different categories 924, col. 7, lines 58-60, Fig. 10. The Examiner notes AI Input Broker 810 is the planner module that decomposes the prompt into a plan of derived prompts for different categories 924, Fig. 10) instantiated in machine-readable memory and operable by a processor (an illustration of an in-memory processing architecture for h-LLMs, col. 9, lines 38-39.);(the processor and operable to store software and other digital information thereupon in one or both of transitory and non-transitory status (such as ..., memory, and the like), col. 10, lines 44-48. The Examiner notes that the planner module is instantiated in memory);
mapping a domain to each of a plurality of distinct machine learning models (The base h-LLM model 104 is fine-tuned to generate multiple h-LLM models which are specialized to perform specific tasks such as Question Answering, Information Extraction, Sentiment Analysis, Image Captioning, Object Recognition, Instruction Following, Classification, Inferencing, and Sentence Similarity, col. 5, lines 47-52. The Examiner notes that specific tasks like Question Answering is a domain);
for each of the plurality of steps: selecting one of the plurality of distinct machine learning models by matching the step to the corresponding mapped domain (Choosing h-LLMs ... and then assigns the request (or its derived form) to one...of the AI h-LLM models within the selected AI h-LLM model categories, col. 3, lines 35-37);
and generating a language output using the selected one of the distinct machine learning models (The ... prompts 822 are sent multiple h-LLMs 824 which produce the results, col. 7, lines 36-37); and
integrating the language outputs of each of the plurality of steps into a syntactically and semantically coherent final output via an integration module utilizing a large language model (The merger block 1516 combines and merges the h-LLMs from the batch layer and real-time layer to produce a combined h-LLM. The merged h-LLM is used with the query layer 1518 to respond to prompts, col. 9, lines 32-36); the model is able to generate coherent and contextually relevant responses based on the query in the prompt, col. 2, lines 34-35).
Madisetti is silent about the term instantiated.
Balasubramaniam teaches a method of processing a compound prompt (An input of a LLM is called a prompt (col. 4, line 3); a prompt including a user input (col. 4, lines 11-12); Step 604, Input: (What is the fastest bird and how many are there in the wild?), Fig. 6A. The Examiner notes the prompt “What is the fastest bird and how many are there in the wild?” is a compound prompt. The Examiner notes instant specification discloses: “The term “compound prompt” can refer to a prompt explicitly including multiple separate tasks or steps, e.g., “(1) identify the three highest-selling jazz musicians of the 1970s, and then (2) generate a report comparing the musical styles of these three musicians.”, instant specification: ([0014])),
the method comprising: decomposing the compound prompt into a plan having a plurality of steps via a planner module (In instances where the plan generation component 735 generates more than one task to be completed in order to perform the action responsive to the user input, the plan generation component 735 may further maintain and prioritize the list of tasks as the processing of the system 100 with respect to the user input is performed (col. 18, lines 9-15))
instantiated in machine-readable memory and operable by a processor (An LLM agent component 752 may correspond to a custom instantiation of an LLM (and other components) that is configured to handle user inputs relating to a particular domain/functionality (col. 28, lines 39-42); The action plan execution component 745 may send the action data 747a-n to the API provider component 750, the LLM agent component 752 (col. 28, lines 21-23); The LLM container 150 may load (414) the encoded prompt into the GPU memory (col. 11, lines 39-40); Each of the LLM containers 150a-150n may be in communication with its own corresponding cache 170 that may be stored across multiple GPUs 165, col. 6, lines 53-55);
It would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention to have modified the method of Madisetti to incorporate the teachings of Balasubramaniam for the benefit of an advanced artificial intelligence system designed to process, understand, and generate human-like text based on massive amounts of data (Balasubramaniam, col. 15, lines 40-42)
Regarding claim 2, Madisetti and Balasubramaniam teaches the method of claim 1, Madisetti teaches further comprising training each of the plurality of distinct machine learning models using different training data specific to its respective domain (Referring now to FIG. 2 is an illustration of h-LLMs trained with different training sets (col. 5, lines 53-54); For example, as shown in FIG. 2, h-LLM-1 152 is trained with training set-1 150, h-LLM-2 156 is trained with training set-2 154, h-LLM-3 160 is trained with training set-3 158, and h-LLM-3_4 164 is trained with training set-3 158 and training set-4 162, col. 5, lines 64-67).
Regarding claim 3, Madisetti and Balasubramaniam teaches the method of claim 2, Madisetti teaches wherein at least a subset of the plurality of distinct machine learning models are trained entirely separately from others of the plurality of distinct machine learning models, without overlapping training data (Referring now to FIG. 2 is an illustration of h-LLMs trained with different training sets (col. 5, lines 53-54). The Examiner notes that h-LLM-1 and h-LLM-2 are trained separately without overlapping training data).
Regarding claim 4, Madisetti and Balasubramaniam teaches the method of claim 2, Madisetti teaches wherein at least a subset of the plurality of distinct machine learning models are specialized in a respective domain via fine-tuning or transfer learning (The base h-LLM model 104 is fine-tuned to generate multiple h-LLM models which are specialized to perform specific tasks such as Question Answering, Information Extraction, Sentiment Analysis, Image Captioning, Object Recognition, Instruction Following, Classification, Inferencing, and Sentence Similarity, col. 5, lines 47-52).
Regarding claim 5, Madisetti and Balasubramaniam teaches the method of claim 1, Madisetti teaches wherein mapping the domain to each of the plurality of distinct machine learning models comprises mapping a subject-matter specialization to each of the plurality of distinct machine learning models (In Fig. 3, Multiple h-LLMs Specialized for Specific Tasks/categories 208. The Examiner notes the first h-LLMs is mapped to Question Answering, second h-LLMs is mapped to Information Extraction, third h-LLMs is mapped to Sentiment Analysis and so on).
Regarding claim 6, Madisetti and Balasubramaniam teaches the method of claim 1, Balasubramaniam teaches wherein mapping the domain to each of the plurality of distinct machine learning models comprises training a selection model via machine learning (train one or more components of the system 100 (col. 45, line 30); the system 100 may include ... multiple LLM containers 150a-150n (e.g., a group of LLM containers) and a load balancer 160, col. 5, lines 12-14) to map compound prompts to one or more of the plurality of distinct machine learning models (The load balancer 160 may select a LLM container from a group of LLM containers 150a-150n to process a prompt, col. 5, lines 59-60).
The same motivation to combine independent claim 1 applies here.
Regarding claim 7, Madisetti and Balasubramaniam teaches the method of claim 1, Balasubramaniam teaches generating a language output using the selected one of the distinct machine learning models (The output generated by the LLM may be a natural language output responsive to the prompt, col. 16, lines 28-30) comprises providing the selected one of the plurality of distinct machine learning models with at least a portion of the complex prompt, one of the language outputs from another of the plurality of steps (The response language model 1020 may send the model output data 1025 to the compliance component 1030 (col. 36, lines 66-67); In some embodiments, the compliance component 1030 may include/implement an ML model, col. 37, lines 11-13. The Examiner notes the instant specification discloses: “this disclosure will treat complex and compound prompts as equivalent” [0014]), or both.
The same motivation to combine independent claim 1 applies here.
Regarding claim 8, Madisetti and Balasubramaniam teaches the method of claim 7, Balasubramaniam teaches wherein generating each language output for each of the plurality of steps after the first comprises providing the selected one of the plurality of machine learning models with a prompt including an output of one or more preceding steps (For example, based on processing the first example prompt data provided above, the response language model 1020 may determine an ambiguity exists with respect to the size of the pizza to be ordered and may generate model output data 1025c: {“What size pizza should I order?”,} {“What size pizza does the user usually order?”,}, col. 36, lines 41-47).
The same motivation to combine dependent claim 7 applies here.
Regarding claim 9, Madisetti and Balasubramaniam teaches the method of claim 1, Madisetti teaches wherein at least a subset of the plurality of machine learning models are large language models (FIG. 2 is an illustration of h-LLMs trained with different training sets, according to an embodiment of the invention, col. 3, lines 61-62).
Regarding claim 10, Madisetti and Balasubramaniamteaches the method of claim 1, Balasubramaniam teaches wherein the large language model is one of the plurality of distinct machine learning models (In some embodiments, the language models 820, 840, 940, 1020 may be fine-tuned to perform a particular task(s), col. 43, lines 29-30) having a corresponding generalist domain (This allows the model to learn more nuanced and complex language patterns across different tasks, leading to better generalization and performance. Col. 43, lines 40-42).
The same motivation to combine independent claim 1 applies here.
Regarding claim 11, Madisetti and Balasubramaniam teaches the method of claim 10, Balasubramaniam teaches wherein the selection of one of the plurality of distinct machine learning models for each step (The load balancer 160 may select a LLM container from a group of LLM containers 150a-150n to process a prompt (col. 5, lines 59-60); The load balancer 160 may receive (step 2) the prompt 140. The load balancer 160 may determine which LLM container 150 the user input/request is to be routed, col. 5, lines 38-41) is performed by the large language model (In some embodiments, the LLM orchestrator 130 may be in communication with multiple LLM containers 150 via the load balancer 160 (col. 5, lines 36-38); the LLM orchestrator 130 may be in communication with (or may include) the LLM container(s) 150, which may implement a LLM and corresponding components to enable processing by the LLM, col. 5, lines 32-35).
The same motivation to combine dependent claim 10 applies here.
Regarding claim 12, Madisetti and Balasubramaniamteaches the method of claim 1, Balasubramaniam teaches wherein decomposing the compound prompt into a plan having a plurality of steps comprises identifying the plurality of steps (The LLM container 150, in some examples, may perform more than one processing iteration with respect to an individual natural language input and may be configured to follow a processing/reasoning format of: “Think”, “Action”, “Observe” and “Response.”, col. 6, lines 27-31),
ordering the plurality of steps (The Think step is the LLM's reasoning process to identify the intermediate task to perform; the Action is given the intermediate task, in this step select actions (APIs, components, etc.) in order to complete the task and generate the command to perform the action, col. 6, lines 31-35), and
identifying outputs from at least one of the plurality of steps to be received as inputs by another of the plurality of steps (and the Observe step contains the results from performance of the action(s). The Think, Action and Observe steps may be performed for multiple iterations. The Response step is a response to the natural language input generated based on the iterative processing performed via the Think, Action and Observe steps, col. 6, lines 36-41).
The same motivation to combine independent claim 1 applies here.
Regarding claim 13, Madisetti and Balasubramaniam teaches the method of claim 1, Madisetti teaches wherein at least some of the plurality of steps are executed sequentially (FIG. 5 is an illustration ... where multiple h-LLMs of increasing precision and accuracy are created in a sequential manner, col. 4, lines 4-6).
Regarding claim 14, Madisetti teaches a system (Referring now to FIG. 1 is an illustration of the training process for creating multiple specialized large language models for specific tasks/categories, col. 5, lines 37-39) for generating a response (an input is given to the model in the form of a prompt and the model is able to generate coherent and contextually relevant responses based on the query in the prompt, col. 2, lines 32-35) to a complex prompt (A method of answering queries from user prompts, abstract; The prompt is sent to AI Input Broker 810 which generates multiple derived prompts for different categories 924, col. 7, lines 58-60;The Examiner notes the prompt is both compound prompt and complex because it generates multiple derived prompts and the instant specification discloses: “this disclosure will treat complex and compound prompts as equivalent” [0014]),
the system comprising: an input device configured to receive the complex prompt (User 900 enters a prompt in user interface 902, col. 7, lines col. 7, lines 57-58);
a logic processor (a processor configured to execute commands received from software (such as microprocessors, field-programmable gate arrays, integrated circuits, and the like), col. 10, lines 40-42. The Examiner notes that the instant specification discloses: “Processor 102 is a logic-capable device configured to execute software, applications, and/or programs stored on memory 104. Examples of processor 102 can include one or more of a processor, a microprocessor,” [0019]);
machine-readable memory (the processor and operable to store software and other digital information thereupon in one or both of transitory and non-transitory status (such as ..., memory, and the like), col. 10, lines 44-48);
a plurality of specialized large language models (LLMs) instantiated in the machine-readable memory (multiple h-LLMs specialized for specific tasks or categories 208 (col. 6, lines 17-18); Referring now to FIG. 17 , an illustration of an in-memory processing architecture for h-LLMs, col. 9, lines 38-39. The Examiner notes that large language models (LLMs) are instantiated in memory)
each of the specialized LLMs having a corresponding domain of specialization (multiple h-LLM models which are specialized to perform specific tasks such as Question Answering, Information Extraction, Sentiment Analysis, Image Captioning, Object Recognition, Instruction Following, Classification, Inferencing, and Sentence Similarity, col. 5, lines 47-52); and
a manager (Fig. 10) comprising:
a planner (Referring now to FIG. 10 ... The prompt is sent to AI Input Broker 810 which generates multiple derived prompts for different categories 924, col. 7, lines 55-60) instantiated in the machine-readable memory and operable via the logic processor (an in-memory processing architecture for h-LLMs, (col. 9, lines 38-39); the processor and operable to store software and other digital information thereupon in one or both of transitory and non-transitory status (such as ..., memory, and the like), col. 10, lines 44-48. The Examiner notes that the planner is instantiated in memory) to decompose the complex prompt into a plan having a plurality of steps (The prompt is sent to AI Input Broker 810 which generates multiple derived prompts for different categories 924, col. 7, lines 58-60, Fig. 10. The Examiner notes AI Input Broker 810 is the planner that decomposes the prompt into a plan of derived prompts for different categories 924, Fig. 10);
a module (The derived prompt is sent to multiple h-LLMs 1108 connected in series. The derived prompt goes to the first h-LLM in the sequence which generates results, col. 8, lines 25-28, Fig. 10. The Examiner notes that multiple h-LLMs as selection module) instantiated in the machine-readable memory and operable via the logic processor (an in-memory processing architecture for h-LLMs, (col. 9, lines 38-39); the processor and operable to store software and other digital information thereupon in one or both of transitory and non-transitory status (such as ..., memory, and the like), col. 10, lines 44-48. The Examiner notes that the selection module is instantiated in memory) to select one of the plurality of specialized LLMs to execute each of the plurality of steps (Choosing h-LLMs ... and then assigns the request (or its derived form) to one...of the AI h-LLM models within the selected AI h-LLM model categories, col. 3, lines 35-37);
an integration module (The merger block 1516 combines and merges the h-LLMs from the batch layer and real-time layer to produce a combined h-LLM, col. 9, lines 32-34); the model is able to generate coherent and contextually relevant responses based on the query in the prompt, col. 2, lines 34-35) instantiated in the machine-readable memory and operable via the logic processor (an in-memory processing architecture for h-LLMs, (col. 9, lines 38-39); the processor and operable to store software and other digital information thereupon in one or both of transitory and non-transitory status (such as..., memory, and the like), col. 10, lines 44-48. The Examiner notes that the integration module is instantiated in memory)
to generate a language output responsive to the complex prompt from outputs of each of the selected ones of the plurality of specialized LLMs (The merged h-LLM is used with the query layer 1518 to respond to prompts, col. 9, lines 35-36); the model is able to generate coherent and contextually relevant responses based on the query in the prompt, col. 2, lines 34-35).
Madisetti is silent about the term instantiated and selection module
Balasubramaniam teaches a system for generating a response (The LLM container 150 may process natural language inputs and generate a corresponding response, col. 5, lines 63-64) to a complex prompt (a prompt including a user input (col. 4, lines 11-12); Step 604, Input: (What is the fastest bird and how many are there in the wild?), Fig. 6A. The Examiner notes the prompt “What is the fastest bird and how many are there in the wild?” is a complex prompt. The Examiner notes instant specification discloses: “this disclosure will treat complex and compound prompts as equivalent” [0014])
an input device configured to receive the complex prompt (a user device 110 from which the user input is received, col. 6, lines 3-4);
a logic processor; machine-readable memory (The orchestrator component 1130 may include memory and logic that enables the orchestrator component 1130 to transmit various pieces and forms of data to various components of the system (col. 47, lines 62-66); Each of the LLM containers 150a-150n may be in communication with its own corresponding cache 170 that may be stored across multiple GPUs 165, col. 6, lines 53-55);
a plurality of specialized large language models (LLMs) instantiated in the machine-readable memory (An LLM agent component 752 may correspond to a custom instantiation of an LLM (and other components) that is configured to handle user inputs relating to a particular domain/functionality (col. 28, lines 39-42),
a manager (system 100, Fig. 1 and 7) comprising: a planner (In instances where the plan generation component 735 generates more than one task to be completed in order to perform the action responsive to the user input, the plan generation component 735 may further maintain and prioritize the list of tasks as the processing of the system 100 with respect to the user input is performed (col. 18, lines 9-15)) instantiated in the machine-readable memory and operable via the logic processor (An LLM agent component 752 may correspond to a custom instantiation of an LLM (and other components) that is configured to handle user inputs relating to a particular domain/functionality (col. 28, lines 39-42); The orchestrator component 1130 may include memory and logic that enables the orchestrator component 1130 to transmit various pieces and forms of data to various components of the system (col. 47, lines 62-66); Each of the LLM containers 150a-150n may be in communication with its own corresponding cache 170 that may be stored across multiple GPUs 165, col. 6, lines 53-55),
a selection module (The load balancer 160 may select a LLM container from a group of LLM containers 150a-150n to process a prompt, col. 5, lines 59-60, Fig. 1) instantiated in the machine-readable memory and operable via the logic processor (An LLM agent component 752 may correspond to a custom instantiation of an LLM (and other components) that is configured to handle user inputs relating to a particular domain/functionality (col. 28, lines 39-42); The orchestrator component 1130 may include memory and logic that enables the orchestrator component 1130 to transmit various pieces and forms of data to various components of the system (col. 47, lines 62-66); Each of the LLM containers 150a-150n may be in communication with its own corresponding cache 170 that may be stored across multiple GPUs 165, col. 6, lines 53-55), and
It would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention to have modified the method of Madisetti to incorporate the teachings of Balasubramaniam for the benefit of an advanced artificial intelligence system designed to process, understand, and generate human-like text based on massive amounts of data (Balasubramaniam, col. 15, lines 40-42)
Regarding claim 15, Madisetti and Balasubramaniam teaches the system of claim 14, Madisetti teaches wherein the manager further comprises a generalist LLM (Data 200 is used to train a base h-LLM model 204, col. 6, lines 14-15. The Examiner notes base h-LLM model is the generalist model), and
wherein the integration module generates the language output from outputs of at least a subset of the selected ones of the plurality of specialized large language models using the generalist large language model (a base h-LLM 1502 which is fine tuned 1504 in batch to generate fine-tuned h-LLM 1508... The merger block 1516 combines and merges the h-LLMs from the batch layer and real-time layer to produce a combined h-LLM. The merged h-LLM is used with the query layer 1518 to respond to prompts 1520, col. 9, lines 29-36).
Regarding claim 16, Madisetti and Balasubramaniam teaches the system of claim 15, Madisetti teaches wherein the generalist LLM is a Meta-Language Model (Data 200 is used to train a base h-LLM model 204 ... to generate multiple h-LLMs specialized for specific tasks, col. 6, lines 14-18. The Examiner notes base h-LLM model is a Meta-Language Model because it generates specialized LLMs according to the instant specification which discloses: “a meta-language model (i.e., wherein model 210 is a meta-language model) trained to designate a single specialist model”[0030]).
Regarding claim 17, Madisetti and Balasubramaniam teaches the system of claim 15, Madisetti teaches wherein the selection module maps each of the plurality of steps to one of the plurality of specialized large language models using the generalist large language model (The base h-LLM model 104 is fine-tuned to generate multiple h-LLM models which are specialized to perform specific tasks such as Question Answering, Information Extraction, Sentiment Analysis, Image Captioning, Object Recognition, Instruction Following, Classification, Inferencing, and Sentence Similarity, for instance, col. 5, lines 47-52. The Examiner notes that base h-LLM model is the generalist large language model).
Balasubramaniam teaches wherein the selection module maps each of the plurality of steps to one of the plurality of specialized large language models (The load balancer 160 may select a LLM container from a group of LLM containers 150a-150n to process a prompt, col. 5, lines 59-60)
It would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention to have modified the method of Madisetti to incorporate the teachings of Balasubramaniam for the benefit of an advanced artificial intelligence system designed to process, understand, and generate human-like text based on massive amounts of data (Balasubramaniam, col. 15, lines 40-42)
Regarding claim 18, Madisetti and Balasubramaniam teaches the system of claim 14, Balasubramaniam teaches wherein the selection module comprises a model record identifying the corresponding domain of specialization (The load balancer 160 may receive (step 2) the prompt 140 ...The prompt 140 may include a session ID or the prompt 140 may be associated with a session ID for the dialog session, and the load balancer 160 may use the session ID to route the prompt 140 to a particular LLM container 150 (col. 5, lines 38-46); The Examiner notes prompt 140 including a session ID is a model record and the session ID identifies the domain of specialization) and at least one of an input format and an output format for each of the plurality of specialized large language models (This may be supported by the load balancer 160 routing a user input to a LLM container 150 that processed similar past user inputs (col. 9, lines 33-35); The Examiner notes processed similar past user inputs indicates input format of the user).
The same motivation to combine independent claim 14 applies here.
Regarding claim 19, Madisetti and Balasubramaniam teaches the system of claim 14, Madisetti teaches further comprising a database communicatively coupled with at least one of plurality of specialized large language models to provide context injection for that respective specialized large language model (An AI Broker Database 1020 stores the results along with the meta-data information such as the request path. AI Broker Database 1020 creates an index of “derived requests” that may be used in future to select which set of “derived requests” an incoming request may fall into for further processing, col. 8, lines 14-20, Fig. 11. The Examiner notes that AI Broker Database 1020 in Fig. 11 is communicatively coupled to at least one of plurality of specialized large language models).
Regarding claim 20, Madisetti and Balasubramaniam teaches the system of claim 14, Madisetti teaches wherein each of the plurality of specialized large language models is trained for its respective domain using different training data and/or parameters than all others of the plurality of specialized large language models (Referring now to FIG. 2 is an illustration of h-LLMs trained with different training sets (col. 5, lines 53-54); For example, as shown in FIG. 2, h-LLM-1 152 is trained with training set-1 150, h-LLM-2 156 is trained with training set-2 154, h-LLM-3 160 is trained with training set-3 158, and h-LLM-3_4 164 is trained with training set-3 158 and training set-4 162, col. 5, lines 64-67).
Regarding claim 21, Madisetti and Balasubramaniam teaches system of claim 20, Madisetti teaches wherein at least a subset of the specialized large language models are trained via fine tuning, transfer learning, or both (The base h-LLM model 104 is fine-tuned to generate multiple h-LLM models which are specialized to perform specific tasks such as Question Answering, Information Extraction, Sentiment Analysis, Image Captioning, Object Recognition, Instruction Following, Classification, Inferencing, and Sentence Similarity, col. 5, lines 47-52).
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to MORIAM MOSUNMOLA GODO whose telephone number is (571)272-8670. The examiner can normally be reached Monday-Friday 8:00am-5:00pm EST.
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, Michelle T. Bechtold can be reached on (571) 431-0762. 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.G./Examiner, Art Unit 2148
Prosecution Insights