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 .
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:
At step 1:
Claims 1-20 is directed to a “chain of thought machine learning model debiasing” and thus directed to a statutory category.
At step 2A, Prong One:
The claim 1 recites the following limitation directed to an abstract ideas:
“generating, by the processing device, a prompt including the context data, the query, and a chain-of-though prompt” recites a mental process as generatinga prompt including the context data, the query, and a chain-of-though prompt.
“receiving, by the processing device, a candidate result based on processing of the prompt using a machine-learning model, the candidate result including a candidate answer and a chain-of-thought result describing reasoning indicated by the machine-learning model as used in generating the candidate answer” recites a mental process as under Alice/Mayo framework such as calculations, and evaluation that can theoretically be performed in the human mind care classified as abstract idea.
“presenting, by the processing device, the candidate result including the candidate answer and the chain-of-thought result for output” recites the mental process presenting the candidate result including the candidate answer and the chain-of-thought result for output.
With respect to claims 2-9, the claims 2-3 recite the limitations that are classified as abstract ideas and can be perform a mental process.
The claim 10 recites the following limitation directed to an abstract ideas:
“generating a factual prompt including a query, a chain-of-thought prompt, and factual context data and a counterfactual prompt including the query, the chain-of-thought prompt, and counterfactual context data” recite the mental process generating a factual prompt including a query, a chain-of-thought prompt, and factual context data and a counterfactual prompt including the query, the chain-of-thought prompt, and counterfactual context data;
“receiving a factual candidate result based on processing of the factual prompt using a machine-learning model” recites a mental process as under Alice/Mayo framework such as calculations, and evaluation that can theoretically be performed in the human mind care classified as abstract idea.
“receiving a counterfactual candidate result based on processing of the counterfactual context data using the machine-learning model” recite the mental process receiving a counterfactual candidate result based on processing of the counterfactual context data using the machine-learning model like Alice/Mayo case.
“receiving a counterfactual candidate result based on processing of the counterfactual context data using the machine-learning model” recite the mental process receiving a counterfactual candidate result based on processing of the counterfactual context data using the machine-learning model like Alice/Mayo case
“estimating bias in an internal knowledge source of the machine-learning model based on the factual candidate result and the counterfactual candidate result.“ recites a mental process as under Alice/Mayo framework such as calculations, and evaluation that can theoretically be performed in the human mind care classified as abstract idea.
Claims 11-15, claims 11-15 recite the limitations that are classified as abstract ideas and can be perform a mental process.
The claim 16 recites the following limitation directed to an abstract ideas:
“generating, by a processing device, a plurality of prompts respectively including a query, context data based on the query, and a chain-of-though prompt;” recites a mental process as generating a plurality of prompts respectively including a query, context data based on the query, and a chain-of-though prompt.
“generating, by the processing device, a plurality of candidate results by processing the plurality of prompts using a machine-learning model, each said candidate result including a candidate answer and a chain-of-thought result describing reasoning indicated by the machine-learning model as used in generating the candidate answer” recite the mental process receiving a counterfactual candidate result based on processing of the counterfactual context data using the machine-learning model like Alice/Mayo case
“estimating, by the processing device, a causal effect of the context data on operation of the machine-learning model based on the plurality of candidate results” recite the mental process receiving a counterfactual candidate result based on processing of the counterfactual context data using the machine-learning model like Alice/Mayo case.
“mediating, by the processing device, subsequent operation of the machine-learning model based on the estimating” recite the mental process receiving a counterfactual candidate result based on processing of the counterfactual context data using the machine-learning model like Alice/Mayo case.
Claims 17-20, claims 17-20 recite the limitations that are classified as abstract ideas and can be perform a mental process.
At step 2A, Prong Two:
The claims recite the following additional elements:
That the content management system includes “processing device” “computer readable medium”, which are high level recitation of generic computer component s and functions and represent mere instruction to apply to a computer as in MPEP 2106.05 (f) which does not provide integration into a practical application.
At step 2B
The conclusions for the mere implementation using a generic computer and mere field of use are carried over and to not provide significantly more.
Claim Rejections - 35 USC § 102
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.
Claim(s) 1-2, 4 and 9 is/are rejected under 35 U.S.C. 102(a)(1) as being anticipated by Bernstein et al. (U.S. Pub. 2025/0021842 A1).
With respect to claim 1, Bernstein et al. discloses a method comprising:
receiving, by a processing device, a query; producing, by the processing device, context data based on the query;
generating, by the processing device, a prompt including the context data, the query, and a chain-of-though prompt (i.e., “ a prompt can include prompt context where appropriate. Prompt context can include prior prompts and generated responses (or portions thereof), if they exist. Such context is often referred to as occurring in a session. A new session starts with no prompt context.”(0017)“ The prompt and response may be translated into an input format expected by the expert system. In other words, a prompt and response can be translated into a first input format configured for processing by a first expert system and also into a second input format configured for processing by a second expert system. This enables the expert systems to operate without change. For example, a prompt may be translated to, or treated as, a query for evaluation by the knowledge engine and the response to the prompt may be translated to, or treated as, a candidate document to be scored by the knowledge engine in response to the query”(0018), (i.e., “Such analysis can inform retraining of the model. Remedial action can also include chain of thought/tree of thought approaches, where the model is asked the same questions differently and the expert systems can be used to inform the tree of thought approach”(0058), “The remedial action can include changing the quality threshold, e.g., if it is decided that the threshold is too sensitive. The remedial action can include further training and/or chain of thought/tree of thought approaches, as described with regard to FIG. 3. Remedial action can include keeping a version of the previous model running, identifying prompts similar to the quality backstop prompts that failed to meet the quality threshold, and sending those prompts to the previous model”(0077)));
receiving, by the processing device, a candidate result based on processing of the prompt using a machine-learning model (i.e., “The large language model 230 is an example of a large language model. The large language model 230 is trained to generate conversational responses to prompts (e.g., response 255) that have some level of creativity. In some implementations, the response can include an image. The image may be relevant to an entity in the prompt and/or the response…This makes the response 255 different from search results obtained in response to a search query. Search results include relevant text (snippets) taken directly from a source. Some search result pages include short answers. The short answers are conventionally taken from a search result or from an API of a service (e.g., such as weather data from weather.com). Some search result pages include a knowledge panel, with information taken from an entity repository, such as a knowledge graph. ”(0040)), the candidate result including a candidate answer and a chain-of-thought result describing reasoning indicated by the machine-learning model as used in generating the candidate answer (i.e., “ At step 304, the system evaluates first responses generated by a large language model for prompts in the set of prompts. In other words, the first responses are responses generated by a prior version of the model for the prompts in the set of prompts. In some implementations, these first responses may be generated in response to the benchmarking, or in other words as part of the current benchmark testing. In some implementations, these first responses may have been previously generated and evaluated.”(0056) and “a prompt may be translated to, or treated as, a query for evaluation by the knowledge engine and the response to the prompt may be translated to, or treated as, a candidate document to be scored by the knowledge engine in response to the query.”(0018)); and
presenting, by the processing device, the candidate result including the candidate answer and the chain-of-thought result for output (i.e., ) and “a prompt may be translated to, or treated as, a query for evaluation by the knowledge engine and the response to the prompt may be translated to, or treated as, a candidate document to be scored by the knowledge engine in response to the query.”(0018) and “Such analysis can inform retraining of the model. Remedial action can also include chain of thought/tree of thought approaches, where the model is asked the same questions differently and the expert systems can be used to inform the tree of thought approach.”(0058)); and
With respect to claim 2, Bernstein et al. discloses wherein the producing of the context data is performed from an external knowledge source independently of an internal knowledge source utilized by the machine-learning model (i.e., “ the knowledge engine may determine whether a fact (entity attribute or entity relationship) in the document differs from a fact represented in the database of entities.”(0029) and “ the search system 120 identifies the resources 105 by crawling and indexing the resources 105 provided on web sites 104. Data about the 105 resources can be indexed. The indexed and, optionally, cached copies of the resources 105 are stored in a search index 122, e.g., as indexed resources 126.”(0028) and fig. 2 shows the external is website resources 105 and the internal is expert system 150 ).
With respect to claim 4, Bernstein et al. discloses wherein the generating includes generating: a factual prompt including the query (i.e., “ A factual query is a query that includes an entity and requests information about an entity. The system may use this existing process of a knowledge engine to obtain the score by providing the first response as a document and the prompt as a query. Thus, as with the veracity score, the existing knowledge engine process can be used to parse the prompt and response as a query and candidate responsive document”(0062)), the chain-of-thought prompt (i.e., “Such analysis can inform retraining of the model. Remedial action can also include chain of thought/tree of thought approaches, where the model is asked the same questions differently and the expert systems can be used to inform the tree of thought approach”(0058), “The remedial action can include changing the quality threshold, e.g., if it is decided that the threshold is too sensitive. The remedial action can include further training and/or chain of thought/tree of thought approaches, as described with regard to FIG. 3. Remedial action can include keeping a version of the previous model running, identifying prompts similar to the quality backstop prompts that failed to meet the quality threshold, and sending those prompts to the previous model”(0077)), and factual context data (i.e.,” The response generator 250 may have a similar architecture as other conversational large language models (e.g., GLaM, LaMDA, PaLM, GPT-3, ChatGPT, etc.). The response generator 250 is capable of generating responses with high creativity, for example in response to open-ended prompts (e.g., “give me five first date ideas”) and prompts that lack any factual context (e.g., “how are you?”). In some implementations, the response generator 250 generates responses to factual questions informed by the additional context 237, by “memorized” facts that are learned from the training data, and by potential hallucinations that might be related to the conversation context or the stochasticity of the generation process.”(0043)); and a counterfactual prompt including the query, the chain-of-thought prompt, and counterfactual context data (i.e., (e.g., GLaM, LaMDA, PaLM, GPT-3, ChatGPT, etc.). The response generator 250 is capable of generating responses with high creativity, for example in response to open-ended prompts (e.g., “give me five first date ideas”) and prompts that lack any factual context (e.g., “how are you?”). In some implementations, the response generator 250 generates responses to factual questions informed by the additional context 237, by “memorized” facts that are learned from the training data, and by potential hallucinations that might be related to the conversation context or the stochasticity of the generation process.”(0043);
(i.e., (e.g., GLaM, LaMDA, PaLM, GPT-3, ChatGPT, etc.). The response generator 250 is capable of generating responses with high creativity, for example in response to open-ended prompts (e.g., “give me five first date ideas”) and prompts that lack any factual context (e.g., “how are you?”). In some implementations, the response generator 250 generates responses to factual questions informed by the additional context 237, by “memorized” facts that are learned from the training data, and by potential hallucinations that might be related to the conversation context or the stochasticity of the generation process.”(0043);
With respect to claim 9, Bernstein et al. discloses wherein the machine-learning model is a large language model (i.e., “A system is disclosed that uses expert systems to monitor and evaluate quality in a large language model”(abstract)).
Allowable Subject Matter
Claim 3 is objected to as being dependent upon a rejected base claim, but would be allowable if rewritten in independent form including all of the limitations of the base claim and any intervening claims and overcoming the rejection under 35 U.S.C. 101 since the prior art of record and considered pertinent to the applicant’s disclosure does not teach or suggest the claimed further comprising estimating bias from the machine-learning model in generating the candidate result based on irrelevant information included in the context data
Claim 5 is objected to as being dependent upon a rejected base claim, but would be allowable if rewritten in independent form including all of the limitations of the base claim and any intervening claims and overcoming the rejection under 35 U.S.C. 101 since the prior art of record and considered pertinent to the applicant’s disclosure does not teach or suggest the claimed wherein the generating in the counterfactual prompt includes replacing an entity specified in the factual context data with another entity as the counterfactual context data
Claims 6-8 are objected to as being dependent upon a rejected base claim, but would be allowable if rewritten in independent form including all of the limitations of the base claim and any intervening claims and overcoming the rejection under 35 U.S.C. 101 since the prior art of record and considered pertinent to the applicant’s disclosure does not teach or suggest the claimed comprising estimating a causal effect of the context data based on a factual candidate result generated by the machine-learning model based on the factual prompt and a counterfactual candidate result generated by the machine-learning model based on the counterfactual prompt model, wherein the estimating is performed by comparing a factual candidate answer and a factual chain-of-though result of the factual candidate result with a counterfactual candidate answer and a counterfactual chain-of-though result of the counterfactual candidate result; wherein the causal effect is an average causal effect.
Claims 10-20 are allowances contingent upon overcoming the rejection under 35 U.S.C. 101.
With respect to claims 10-15, Bernstein et al. discloses a computing device comprising:
a processing device; and a computer-readable storage medium storing instructions that, responsive to execution by the processing device, causes the processing device to perform operations including:
generating a factual prompt including a query (i.e., “ A factual query is a query that includes an entity and requests information about an entity. The system may use this existing process of a knowledge engine to obtain the score by providing the first response as a document and the prompt as a query. Thus, as with the veracity score, the existing knowledge engine process can be used to parse the prompt and response as a query and candidate responsive document”(0062)), a chain-of-thought prompt (i.e., “Such analysis can inform retraining of the model. Remedial action can also include chain of thought/tree of thought approaches, where the model is asked the same questions differently and the expert systems can be used to inform the tree of thought approach”(0058), “The remedial action can include changing the quality threshold, e.g., if it is decided that the threshold is too sensitive. The remedial action can include further training and/or chain of thought/tree of thought approaches, as described with regard to FIG. 3. Remedial action can include keeping a version of the previous model running, identifying prompts similar to the quality backstop prompts that failed to meet the quality threshold, and sending those prompts to the previous model”(0077)), and factual context data (i.e.,” The response generator 250 may have a similar architecture as other conversational large language models (e.g., GLaM, LaMDA, PaLM, GPT-3, ChatGPT, etc.). The response generator 250 is capable of generating responses with high creativity, for example in response to open-ended prompts (e.g., “give me five first date ideas”) and prompts that lack any factual context (e.g., “how are you?”). In some implementations, the response generator 250 generates responses to factual questions informed by the additional context 237, by “memorized” facts that are learned from the training data, and by potential hallucinations that might be related to the conversation context or the stochasticity of the generation process.”(0043)) and a counterfactual prompt including the query (i.e., (e.g., GLaM, LaMDA, PaLM, GPT-3, ChatGPT, etc.). The response generator 250 is capable of generating responses with high creativity, for example in response to open-ended prompts (e.g., “give me five first date ideas”) and prompts that lack any factual context (e.g., “how are you?”). In some implementations, the response generator 250 generates responses to factual questions informed by the additional context 237, by “memorized” facts that are learned from the training data, and by potential hallucinations that might be related to the conversation context or the stochasticity of the generation process.”(0043) or “Such prompts require a high level of creativity. The large language model 230 can also generate responses for complex questions (e.g., “what causes poverty?”), opinion questions (“is baseball better than cricket?”), etc. These responses may have a high level of creativity while including statements that can be verified (i.e.,factual statements). Because a large language model, such as large language model 230, generates responses with some level of creativity, the responses can include factual information that is incorrect. Such incorrect factual information is referred to as a hallucination’(0040)), the chain-of-thought prompt (i.e., “Such analysis can inform retraining of the model. Remedial action can also include chain of thought/tree of thought approaches, where the model is asked the same questions differently and the expert systems can be used to inform the tree of thought approach.”(0058)), and counterfactual context data (i.e., (e.g., GLaM, LaMDA, PaLM, GPT-3, ChatGPT, etc.). The response generator 250 is capable of generating responses with high creativity, for example in response to open-ended prompts (e.g., “give me five first date ideas”) and prompts that lack any factual context (e.g., “how are you?”). In some implementations, the response generator 250 generates responses to factual questions informed by the additional context 237, by “memorized” facts that are learned from the training data, and by potential hallucinations that might be related to the conversation context or the stochasticity of the generation process.”(0043);
receiving a factual candidate result based on processing of the factual prompt using a machine-learning model (i.e., “ Model quality is measured by scores derived from application of one or more of the expert systems 150 to responses generated for prompts. The scores can reflect things like veracity (factuality), topicality, relevance, similarity with a response generated by an expert system, etc. To evaluate modification made to the large language model 230, the model evaluation system 140 may include a prompt collection 252. … if a particular topic is identified as needing additional training, before performing that additional training the large language model (e.g., the query generator 240) can be used to generate prompts related to that topic. This enables the system to have a large number of prompts for benchmarking the additional training”(0049)); receiving a counterfactual candidate result based on processing of the counterfactual context data using the machine-learning model (i.e., (e.g., GLaM, LaMDA, PaLM, GPT-3, ChatGPT, etc.). The response generator 250 is capable of generating responses with high creativity, for example in response to open-ended prompts (e.g., “give me five first date ideas”) and prompts that lack any factual context (e.g., “how are you?”). In some implementations, the response generator 250 generates responses to factual questions informed by the additional context 237, by “memorized” facts that are learned from the training data, and by potential hallucinations that might be related to the conversation context or the stochasticity of the generation process.”(0043); but Bernstein et al. does not discloses estimating bias in an internal knowledge source of the machine-learning model based on the factual candidate result and the counterfactual candidate result.
With respect to claims 16-20, Bernstein et al. discloses a method comprising:
generating, by a processing device, a plurality of prompts respectively including a query (i.e., “ A factual query is a query that includes an entity and requests information about an entity. The system may use this existing process of a knowledge engine to obtain the score by providing the first response as a document and the prompt as a query. Thus, as with the veracity score, the existing knowledge engine process can be used to parse the prompt and response as a query and candidate responsive document”(0062)), context data based on the query i.e.,” The response generator 250 may have a similar architecture as other conversational large language models (e.g., GLaM, LaMDA, PaLM, GPT-3, ChatGPT, etc.). The response generator 250 is capable of generating responses with high creativity, for example in response to open-ended prompts (e.g., “give me five first date ideas”) and prompts that lack any factual context (e.g., “how are you?”). In some implementations, the response generator 250 generates responses to factual questions informed by the additional context 237, by “memorized” facts that are learned from the training data, and by potential hallucinations that might be related to the conversation context or the stochasticity of the generation process.”(0043)) , and a chain-of-though prompt (i.e., “Such analysis can inform retraining of the model. Remedial action can also include chain of thought/tree of thought approaches, where the model is asked the same questions differently and the expert systems can be used to inform the tree of thought approach”(0058), “The remedial action can include changing the quality threshold, e.g., if it is decided that the threshold is too sensitive. The remedial action can include further training and/or chain of thought/tree of thought approaches, as described with regard to FIG. 3. Remedial action can include keeping a version of the previous model running, identifying prompts similar to the quality backstop prompts that failed to meet the quality threshold, and sending those prompts to the previous model”(0077)),);
generating, by the processing device, a plurality of candidate results by processing the plurality of prompts using a machine-learning model answer (i.e., “ At step 304, the system evaluates first responses generated by a large language model for prompts in the set of prompts. In other words, the first responses are responses generated by a prior version of the model for the prompts in the set of prompts. In some implementations, these first responses may be generated in response to the benchmarking, or in other words as part of the current benchmark testing. In some implementations, these first responses may have been previously generated and evaluated.”(0056) and “a prompt may be translated to, or treated as, a query for evaluation by the knowledge engine and the response to the prompt may be translated to, or treated as, a candidate document to be scored by the knowledge engine in response to the query.”(0018)); but Bernstein et al. does not discloses each said candidate result including a candidate answer and a chain-of-thought result describing reasoning indicated by the machine-learning model as used in generating the candidate answer; estimating, by the processing device, a causal effect of the context data on operation of the machine-learning model based on the plurality of candidate results; and mediating, by the processing device, subsequent operation of the machine-learning model based on the estimating.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to HUNG T VY whose telephone number is (571)272-1954. The examiner can normally be reached M-F 8-5.
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, Tony Mahmoudi can be reached at (571)272-4078. 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.
/HUNG T VY/Primary Examiner, Art Unit 2163 July 29, 2026
Prosecution Insights