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 .
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103 is incorrect, any correction of the statutory basis for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
Claim Objection
Claim 12 is objected to for failing to provide antecedent basis for “the assistance data”. For purpose of interpretation, “the assistance data” is understood as referring to “a series of assistance questions” in claim 11.
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) NOVELTY; PRIOR ART.—A person shall be entitled to a patent unless—
(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; or
(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.
PNG
media_image1.png
18
19
media_image1.png
Greyscale
(b) EXCEPTIONS.—
(1) DISCLOSURES MADE 1 YEAR OR LESS BEFORE THE EFFECTIVE FILING DATE OF THE CLAIMED INVENTION.—A disclosure made 1 year or less before the effective filing date of a claimed invention shall not be prior art to the claimed invention under subsection (a)(1) if—
(A) the disclosure was made by the inventor or joint inventor or by another who obtained the subject matter disclosed directly or indirectly from the inventor or a joint inventor; or
(B) the subject matter disclosed had, before such disclosure, been publicly disclosed by the inventor or a joint inventor or another who obtained the subject matter disclosed directly or indirectly from the inventor or a joint inventor.
(2) DISCLOSURES APPEARING IN APPLICATIONS AND PATENTS.—A disclosure shall not be prior art to a claimed invention under subsection (a)(2) if—
(A) the subject matter disclosed was obtained directly or indirectly from the inventor or a joint inventor;
(B) the subject matter disclosed had, before such subject matter was effectively filed under subsection (a)(2), been publicly disclosed by the inventor or a joint inventor or another who obtained the subject matter disclosed directly or indirectly from the inventor or a joint inventor; or
(C) the subject matter disclosed and the claimed invention, not later than the effective filing date of the claimed invention, were owned by the same person or subject to an obligation of assignment to the same person.
Claims 11-19 are rejected under 35 USC 102(a)(2) as being anticipated by Sharma et al. (US 12423519 B1).
Regarding Claims 11 and 15, Sharma discloses a system (Fig. 1) comprising:
at least one processor (Col 21, Rows 60-62, one or more processors); and
at least one memory storing instructions that, when executed by the at least one processor (Col 21, Rows 56-66, computer executable instructions stored on non-transitory computer readable media and can be executed by one or more processors), cause the system to:
receive an input prompt / instructions to perform an analysis on an input data (Col 5, Rows 47-49 and Col 30, Rows 26-28, user supplied question “What are the activities, risks, controls, and monitors associated with implementing a vector database?”);
compose, based on the input prompt / instructions, a series of incremental prompts / assistance questions (Col 20, Rows 27-30, breaking large complex prompts into smaller, comparatively more tailored (i.e., compose) prompts that each focus on individual segments in a question chain; Col 30, Rows 15-18, breaking / separate large, complex prompts into smaller, comparatively more tailored / composed, prompts that focused on individual segments in an Activity-Risk-Control-Monitoring (A-R-C-M per Col 20, Rows 50-52) chain can enable implementation of cascading, smaller AI models; Col 31, Rows 29-32, construct a series of downstream prompts in A-R-C-M chain in a cascading manner), wherein each of the series of assistance questions / incremental prompts corresponds to a restricted set of responses and response format expected from a large language model (LLM) (Col 23, Rows 30-35, AI models can include LLMs that can be utilized to generate sets of outputs for A-R-C-M chains and to assess the quality of the generated outputs; e.g., Col 4, Rows 4-18, a first instance of LLM specifically trained to output activities required to perform the process (i.e., activity response format), a second instance of the LLM which is trained to identify potential vulnerabilities associated with that activity (i.e., risk response format), a third instance of the LLM which is trained to predict control tools for mitigating these vulnerabilities (i.e., control tools response format), and a fourth instance of the LLM which is trained to recommend monitoring tools for overseeing the implementation of the control tools (i.e., monitoring tools response format); Col 11, Row 65 – Col 12, Row 5, upon receiving a query about implementing a vector database, generate responses tailored to address different aspects of the query);
iteratively perform steps comprising:
provide, to one or more LLMs, an assistance question / incremental prompts of the series of assistance questions / incremental prompts (Col 4, Row 43 – Col 5, Row 11, for “What are the activities, risks, controls, and monitors associated with implementing a vector database?”, input a command prompt based on the query into the first instance of the LLM to output activities associated with the process of implementing a database (e.g., Col 6, Rows 10-15, “list the activities required to implement a vector database”), input one of the activities into the second instance of the LLM to output vulnerabilities associated with the activity (i.e., asking / prompting LLM for the risk of an activity implementing the vector database), input one of the vulnerabilities into the third instance of the LLM to select control tools for controlling the vulnerability, from among the control tools available to the system (i.e., asking / prompting LLM for control tools of a corresponding risk), and input control tools into the fourth instance of the LLM to select monitoring tools available to the system);
analyze a response from the one or more LLMs (Col 7, Rows 34-49, input first activity-related natural language response associated with the first activity into the second instance of the LLM to output one or more vulnerabilities associated with the first activity; e.g., Col 8, Rows 3-12, for the activity “import and vectorize the data”, LLM outputs “there is a risk of a data breach during the import and vectorization process, which may occur if sensitive data is not properly secured or if unauthorized access is gained”; Col 8, Rows 26-32, when analyzing “import and vectorize the data”, LLM outputs probabilities for the risks of data breach, data corruption, and unauthorized access); and
update the series of assistance questions / incremental prompts based on the response (Col 8, Rows 45-49, modify the display previously used to display the activities to show the one or more vulnerabilities associated with the first activity to include a second set of selectable indicators corresponding to the vulnerabilities; Col 8, Rows 50-67, when user selected a first vulnerability (i.e., user selected “data breach”), input the first vulnerability related natural language response into the third instance of the LLM to output control tools for addressing the first vulnerability; e.g., for the tailored prompt asking the third instance of the LLM for a control tool for a selected vulnerability / risk of implementing the vector database, modify the tailored prompt to specifically ask for control tools for user selected “data breach” where the third instance of the LLM would output “audit logging” and “data encryption”); and
determine a final response based on the responses to the series of assistance questions / incremental prompts (Col 11, Row 56 – Col 12, Rows 5, upon receiving a query about implementing a vector database, generate an activity related response, a vulnerability related response, a control related response, and a monitoring related response, each of the responses tailored to address different aspects of the query, providing a comprehensive approach to managing the process; e.g., Col 12, Rows 6-12, “To implement a vector database, you should first install the vector database, be aware of the risk of data breach, mitigate this risk by implementing data encryption, and ensure the effectiveness of this control by regularly performing encryption tool patching”).
Further regarding claim 15, Sharma discloses at least one non-transitory, computer-readable storage medium comprising instructions recorded thereon, wherein the instructions, when executed by at least one data processor of a system, cause the system to implement the system of claim 11 (Col 21, Rows 56-66, computer executable instructions stored on non-transitory computer readable media and can be executed by one or more processors).
Regarding Claim 12, Sharma discloses wherein the assistance data comprises a portion of the input data (Col 5, Rows 48-50, “What are the activities, risks, controls, and monitors associated with implementing a vector database”; Col 6, Rows 1-13, translate the query into a format that the LLM may process effectively: command prompt for the first instance of the LLM “List the activities required to implement a vector database”).
Regarding Claim 13, Sharma discloses wherein composing a series of assistance questions comprises: generating one or more assistance questions that include a portion of the input data and/or a portion of the instructions (Col 30, Rows 25-28, Rows 43-45, and Col 31, Rows 29-32, Rows 47-49 for a user supplied question “What are the activities, risks, controls, and monitors associated with implementing a vector database?”, parse the user supplied question to extract or generate various input items to construct a series of downstream prompts comprising appropriate items (e.g., risk item, control item, monitoring item)).
Regarding Claim 14, Sharma discloses wherein composing a series of assistance questions comprises: selecting, from a set of predetermined series of assistance questions, a series of assistance questions based on the instructions (Col 11, Row 60 – Col 12, Row 4, generate first activity related natural language response, first control related natural language response, and first monitoring related natural language response tailored to address different aspects of the query; see further Col 31, Rows 1-16, generating input items for a prompt by cross-referencing a data store or an ontology to transform “What is” to “define”, “What are” to “generate a list”, and map “vector database” to “storage”; i.e., the tailored prompts / assistance questions comprising pre-determined input items such as “install a vector database”, “data breach”, “data encryption”, “encryption tool patching” to generate final response per Col 12, Rows 6-12).
Regarding Claim 16, Sharma discloses wherein:
the series of incremental prompts includes at least two potential subseries of incremental prompts (Col 31, Rows 29-32 and Rows 47-49, construct the series of downstream prompts comprising appropriate items (e.g., risk, control, monitoring); see Col 6, Rows 10-15, “list the activities required to implement a vector database” as input to the first instance of the LLM to generate activities “install a vector database”, “define a data schema”, “import and vectorize data”, “define database optimizations” and “maintain the database” per Col 6, Rows 30-35 as potential outputs (i.e., subseries of incremental prompts) to the second instance of the LLM); and
updating the series of incremental prompts comprises removing prompts that are not included in one particular subseries of incremental prompts (Col 7, Rows 19-33 and Col 8, Rows 1-20, user selects “import and vectorize the data” as output / incremental prompt to the second instance of the LLM such that “install a vector database”, “define a data schema”, “define database optimizations” and “maintain the database” are removed as output prompts to the second instance of the LLM).
Regarding Claim 17, Sharma discloses in response to analyzing the response from the one or more LLMs (Col 7, Rows 34-49, the first instance of the LLM outputs a first activity related natural language response as an input into the second instance of the LLM in a cascading manner, where the second instance of the LLM predicts one or more vulnerabilities associated with the first activity; i.e., analyzing the first activity related natural language response to predict one or more vulnerabilities), determine assistance data to be supplied to the one or more LLMs (Col 31, Rows 1-15, generating input items for a prompt includes cross-referencing a data store, cross-referencing a user directory, truncating an extracted value, and generating a derivative value based on the extracted value; e.g., Col 31, Rows 29-38, construct a series of downstream prompts in a cascading manner so that for a particular activity A1, generate a set of risks including risk items R1 through Rn; Col 31, Rows 42-46, output items (i.e., activity A1 from the first instance of the LLM) can be replaced with a synonym or ontology item; i.e., analyzing the output of the first instance of the LLM comprising activity item for synonym or ontology item by cross referencing the ontology (“assistance data”) as input prompt to the second instance of the LLM); and supply assistance data to the one or more LLMs (Col 31, Rows 34-46, replace output items with a synonym or ontology item; i.e., generate the downstream prompt for the second instance of the LLM comprising activity items and associated synonym / ontology items).
Regarding Claim 18, Sharma discloses wherein analyzing a response comprises:
supplying the response to a second one or more LLMs (Col 7, Rows 34-49, the first instance of the LLM outputs a first activity related natural language response as an input into the second instance of the LLM in a cascading manner; Col 23, Rows 30-58, AI models 114 include LLMs where a first instance of AI model 114 (first LLM) structured to generate activities, a second instance of the AI model 114 (second LLM) structured to generate risks, a third instance of the AI model 114 (third LLM) structured to generate control or monitoring); and
determining, based on a response of the second one or more LLMs, that the response from the one or more LLMs conforms to a response format corresponding to the incremental prompt (Col 7, Rows 39-49, input the first activity into the second instance of the LLM, which was trained to identify potential vulnerabilities associated with that activity (“expected format” being vulnerabilities associated with a particular activity), to cause the second instance of the LLM to output the one or more vulnerabilities associated with the first activity; Col 7, Row 66 – Col 8, Row 20, LLM generates vulnerability related outputs using formats associated with respective vulnerability related natural language response).
Regarding Claim 19, Sharma discloses wherein updating the series of incremental prompts comprises:
determining that the response does not conform to a response format corresponding to the incremental prompt (Col 28, Row 55 – Col 29, Row 3, component risk corrector 420 correct a faulty A-R-C-M chain by generating suggested missing information and facilitating user acceptance: quality assessment engine 116 causes a trained first instance of an AI model 114 to be executed on system generated data (i.e., activity, risk, control tools cascading prompts to the second, third, and fourth instance of the LLM) to flag various missing items); and
including in the series of incremental prompts a correcting prompt that is directed towards generating a response to the incremental prompt that conforms to the corresponding response format (Col 29, Rows 1-10, provide the flagged data to the user for manual correction and provide the manually corrected and annotated unit of system generated data 134 to the command set generator 112 to generate the additional related items in the A-R-C-M chain).
Claim Rejections - 35 USC § 103
The following is a quotation of the appropriate paragraphs of 35 U.S.C. 103 that form the basis for the rejections under this section made 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.
Claims 1, 3-8, and 10 are rejected under 35 USC 103(a) as being unpatentable over Sharma et al. (US 12423519 B1) in view of Siebel et al. (US 12265570 B2).
Regarding Claim 1, Sharma discloses a computer-implemented method for generating an output from a large language model (LLM) (Col 5, Row 63 – Col 6, Row 2, a system for translating query into a structured command prompt in a format suitable for input into a LLM for effective processing), the method comprising:
receiving an input prompt related to an analysis to be performed on an input data (Col 5, Rows 47-49 and Col 30, Rows 26-28, user supplied question “What are the activities, risks, controls, and monitors associated with implementing a vector database?”);
separating the input prompt into a series of assistance questions having incremental complexities (Col 20, Rows 27-30, breaking (i.e., separating) large complex prompts into smaller, comparatively more tailored prompts that each focus on individual segments in a question chain; Col 30, Rows 15-18, breaking / separate large, complex prompts into smaller, comparatively more tailored / composed, prompts that focused on individual segments in an Activity-Risk-Control-Monitoring (A-R-C-M per Col 20, Rows 50-52) chain can enable implementation of cascading, smaller AI models; Col 31, Rows 29-32, construct a series of downstream prompts in A-R-C-M chain in a cascading manner), wherein each of the series of assistance questions corresponds to a restricted set of responses expected from an LLM (Col 23, Rows 30-35, AI models can include LLMs that can be utilized to generate sets of outputs for A-R-C-M chains and to assess the quality of the generated outputs; e.g., Col 4, Rows 4-18, a first instance of LLM specifically trained to output activities required to perform the process (i.e., LLM restricted to output activities), a second instance of the LLM which is trained to identify potential vulnerabilities associated with that activity (i.e., LLM restricted to output vulnerabilities), a third instance of the LLM which is trained to predict control tools for mitigating these vulnerabilities (i.e., LLM restricted to output control tools), and a fourth instance of the LLM which is trained to recommend monitoring tools for overseeing the implementation of the control tools (i.e., LLM restricted to output monitoring tools); Col 11, Row 65 – Col 12, Row 5, upon receiving a query about implementing a vector database, generate responses tailored to address different aspects of the query);
iteratively performing:
providing an assistance question of the series of assistance questions to the LLM (Col 4, Row 43 – Col 5, Row 11, for “What are the activities, risks, controls, and monitors associated with implementing a vector database?”, input a command prompt based on the query into the first instance of the LLM to output activities associated with the process of implementing a database (e.g., Col 6, Rows 10-15, “list the activities required to implement a vector database”), input one of the activities into the second instance of the LLM to output vulnerabilities associated with the activity (i.e., asking / prompting LLM for the risk of an activity implementing the vector database), input one of the vulnerabilities into the third instance of the LLM to select control tools for controlling the vulnerability, from among the control tools available to the system (i.e., asking / prompting LLM for control tools of a corresponding risk), and input control tools into the fourth instance of the LLM to select monitoring tools available to the system);
analyzing a response from the LLM (Col 7, Rows 34-49, input first activity-related natural language response associated with the first activity into the second instance of the LLM to output one or more vulnerabilities associated with the first activity; e.g., Col 8, Rows 3-12, for the activity “import and vectorize the data”, LLM outputs “there is a risk of a data breach during the import and vectorization process, which may occur if sensitive data is not properly secured or if unauthorized access is gained”; Col 8, Rows 26-32, when analyzing “import and vectorize the data”, LLM outputs probabilities for the risks of data breach, data corruption, and unauthorized access); and
updating the series of assistance questions based on the response (Col 8, Rows 45-49, modify the display previously used to display the activities to show the one or more vulnerabilities associated with the first activity to include a second set of selectable indicators corresponding to the vulnerabilities; Col 8, Rows 50-67, when user selected a first vulnerability (i.e., user selected “data breach”), input the first vulnerability related natural language response into the third instance of the LLM to output control tools for addressing the first vulnerability; e.g., for the tailored prompt asking the third instance of the LLM for a control tool for a selected vulnerability / risk of implementing the vector database, modify the tailored prompt to specifically ask for control tools for user selected “data breach” where the third instance of the LLM would output “audit logging” and “data encryption”); and
determining a final response based on responses to the series of assistance questions (Col 11, Row 56 – Col 12, Rows 5, upon receiving a query about implementing a vector database, generate an activity related response, a vulnerability related response, a control related response, and a monitoring related response, each of the responses tailored to address different aspects of the query, providing a comprehensive approach to managing the process; e.g., Col 12, Rows 6-12, “To implement a vector database, you should first install the vector database, be aware of the risk of data breach, mitigate this risk by implementing data encryption, and ensure the effectiveness of this control by regularly performing encryption tool patching”).
Sharma does not disclose the output is a deterministic output.
Siebel discloses prompting LLM to generate a deterministic response (Abstract, generating a set of potential responses to a prompt using one or more data models in an enterprise information environment that includes access controls to select a deterministic response; see Col 5, Rows 3-5, implement machine learning models such as LLM)
It would’ve been obvious to one ordinarily skilled in the art before the effective filing date of the invention to generate a deterministic response / output from the LLM to avoid hallucination or information leakage (Siebel, Col 5, Rows 33-37) in an architecture where users and systems are able to trust the responses (Siebel, Col 6, Rows 20-23).
Regarding Claim 3, Sharma discloses wherein at least one assistance question corresponds to a restricted set of responses comprising predetermined response categories (Col 23, Rows 30-35, AI models can include LLMs that can be utilized to generate sets of outputs for A-R-C-M chains and to assess the quality of the generated outputs; e.g., Col 4, Rows 4-18, a first instance of LLM specifically trained to output activities required to perform the process (i.e., LLM restricted to output activities), a second instance of the LLM which is trained to identify potential vulnerabilities associated with that activity (i.e., LLM restricted to output vulnerabilities), a third instance of the LLM which is trained to predict control tools for mitigating these vulnerabilities (i.e., LLM restricted to output control tools), and a fourth instance of the LLM which is trained to recommend monitoring tools for overseeing the implementation of the control tools (i.e., LLM restricted to output monitoring tools)).
Regarding Claim 4, Sharma discloses wherein updating the series of assistance questions based on the response includes incorporating a portion of the response into one or more assistance questions (Col 4, Row 43 – Col 5, Row 11, for “What are the activities, risks, controls, and monitors associated with implementing a vector database?”, input a command prompt based on the query into the first instance of the LLM to output activities associated with the process of implementing a database (e.g., Col 6, Rows 10-15, “list the activities required to implement a vector database”), input one of the activities into the second instance of the LLM to output vulnerabilities associated with the activity (i.e., asking / prompting LLM for the risk of an activity implementing the vector database), input one of the vulnerabilities into the third instance of the LLM to select control tools for controlling the vulnerability, from among the control tools available to the system (i.e., asking / prompting LLM for control tools of a corresponding risk), and input control tools into the fourth instance of the LLM to select monitoring tools available to the system).
Regarding Claim 5, Sharma discloses after analyzing the response from the LLM (Col 7, Rows 34-49, the first instance of the LLM outputs a first activity related natural language response as an input into the second instance of the LLM in a cascading manner, where the second instance of the LLM predicts one or more vulnerabilities associated with the first activity; i.e., analyzing the first activity related natural language response to predict one or more vulnerabilities), determining that the response includes a request for assistance data (Col 31, Rows 1-15, generating input items for a prompt includes cross-referencing a data store, cross-referencing a user directory, truncating an extracted value, and generating a derivative value based on the extracted value; e.g., Col 31, Rows 29-38, construct a series of downstream prompts in a cascading manner so that for a particular activity A1, generate a set of risks including risk items R1 through Rn; Col 31, Rows 42-46, output items (i.e., activity A1 from the first instance of the LLM) can be replaced with a synonym or ontology item; i.e., analyzing the output of the first instance of the LLM comprising activity item for synonym or ontology item by cross referencing the ontology (“assistance data”) as input prompt to the second instance of the LLM); and
supplying assistance data to the LLM (Col 31, Rows 34-46, replace output items with a synonym or ontology item; i.e., generate the downstream prompt for the second instance of the LLM comprising activity items and associated synonym / ontology items).
Regarding Claim 6, Sharma discloses wherein updating the series of assistance questions comprises:
selecting a series of subsequent questions from a set of potential series of subsequent questions, wherein a subsequent question is a question from the series of assistance questions that has not yet been supplied to the LLM (Col 11, Row 60 – Col 12, Row 4, generate first activity related natural language response, first control related natural language response, and first monitoring related natural language response tailored to address different aspects of the query; e.g., Col 8, Rows 45-55, display to show one or more vulnerabilities associated with the first activity including a second set of selectable indicators corresponding to the vulnerabilities for a user to select a selectable indicator associated with the first vulnerability; Col 8, Rows 56-61, input a first vulnerability related natural language response into the third instance of the LLM; Col 10, Rows 14-29, display a third set of selectable indicators corresponding to control tools for user to select a selectable indicator associated with a first control tool and input the first control-related natural language response corresponding to the first control tools into the four instance of the LLM).
Regarding Claim 7, Sharma discloses wherein separating the input prompt into a series of assistance questions comprises selecting, from a set of predetermined series of assistance questions, a series of assistance questions based on the input prompt (Col 31, Rows 29-32, construct a series of downstream prompts in A-R-C-M chain in a cascading manner; in particular, Col 11, Row 60 – Col 12, Row 4, generate first activity related natural language response, first control related natural language response, and first monitoring related natural language response tailored to address different aspects of the query; see further Col 31, Rows 1-16, generating input items for a prompt by cross-referencing a data store or an ontology to transform “What is” to “define”, “What are” to “generate a list”, and map “vector database” to “storage”; i.e., selecting pre-determined input items such as “install a vector database”, “data breach”, “data encryption”, “encryption tool patching” to generate tailored prompts in a cascading manner).
Regarding Claim 8, Sharma discloses wherein at least one assistance question includes a portion of the input data (Col 5, Rows 48-50, “What are the activities, risks, controls, and monitors associated with implementing a vector database”; Col 6, Rows 1-13, translate the query into a format that the LLM may process effectively: command prompt for the first instance of the LLM “List the activities required to implement a vector database”).
Regarding Claim 10, Sharma discloses wherein analyzing a response comprises:
supplying the response to a second LLM (Col 7, Rows 34-49, the first instance of the LLM outputs a first activity related natural language response as an input into the second instance of the LLM in a cascading manner; Col 23, Rows 30-58, AI models 114 include LLMs where a first instance of AI model 114 (first LLM) structured to generate activities, a second instance of the AI model 114 (second LLM) structured to generate risks, a third instance of the AI model 114 (third LLM) structured to generate control or monitoring); and
determining, based on a response of the second LLM, that the response from the LLM conforms to a restricted set of responses corresponding to the assistance question (Col 7, Rows 39-49, input the first activity into the second instance of the LLM, which was trained to identify potential vulnerabilities associated with that activity (“restricted set of responses” being vulnerabilities associated with a particular activity), to cause the second instance of the LLM to output the one or more vulnerabilities associated with the first activity; Col 7, Row 66 – Col 8, Row 20, LLM generates vulnerability related outputs using formats associated with respective vulnerability related natural language response).
Claim 2 is rejected under 35 USC 103(a) as being unpatentable over Sharma et al. (US 12423519 B1) and Siebel et al. (US 12265570 B2) as applied to claim 1, in view of Cameron et al. (US 2025/0378280 A1).
Regarding claim 2, Sharma does not disclose wherein at least one assistance question corresponds to a restricted set of responses comprising Boolean answers.
Cameron discloses a machine learning framework comprising a cascade of multiple layers of nonlinear processing units for implicit feature extraction and transformation where each successive layer uses output (i.e., responses) from the previous layer as input (i.e., questions) (¶204), and wherein a restricted response comprises a Boolean answer requiring responses to be either positive or negative (¶212, train AI model 1230 and apply the AI model to feature vector to output indications of a probability that the feature vector has a particular Boolean property).
It would’ve been obvious to one ordinarily skilled in the art before the effective filing date of the invention to implement at least one assistance question corresponds to a restricted set of responses comprising Boolean answers to output indications of whether the assistance question corresponding to the restricted set of responses has an associated desired property (Cameron, ¶212; compare Sharma, Col 16, Rows 40-42, input to a language model is typically natural language parsed into tokens; Col 16, Row 67 – Col 17, Row 10, token sequence is converted into an embedding vector that captures semantic meaning of the text segment represented by the token; Col 17, Rows 32-40, the embeddings are encoded into feature vectors; i.e., the combination applies the LLM to feature vectors representing the series of downstream prompts in the A-R-C-M chain to output probability that the feature vector of a corresponding prompt has a particular Boolean property).
Claim 9 is rejected under 35 USC 103(a) as being unpatentable over Sharma et al. (US 12423519 B1) and Siebel et al. (US 12265570 B2) as applied to claim 1, in view of Myers et al. (US 12182539 B1).
Regarding claim 9, Sharma does not disclose wherein at least one assistance question comprises one-shot or few-shot prompting.
Myers discloses chain of thought prompting a LLM by breaking a problem down into small logical steps (i.e., one or more assistance questions) through zero shot or few-shot prompting (Col 13, Rows 50-54).
It would’ve been obvious to one ordinarily skilled in the art before the effective filing date of the invention to implement at least one assistance question comprises one-shot or few-shot prompting in order to break a problem down into small logical steps (Myers, Col 13, Rows 50-54; compare Sharma, Col 20, Rows 27-30, breaking large complex prompts into smaller, comparatively more tailored (i.e., compose) prompts that each focus on individual segments in a question chain).
Claim 20 is rejected under 35 USC 103(a) as being unpatentable over Sharma et al. (US 12423519 B1) in view of Cameron et al. (US 2025/0378280 A1).
Regarding claim 20, Sharma does not disclose wherein at least one response format corresponding to an incremental prompt comprises a Boolean format requiring responses to be either positive or negative.
Cameron discloses a machine learning framework comprising a cascade of multiple layers of nonlinear processing units for implicit feature extraction and transformation where each successive layer uses output (i.e., responses) from the previous layer as input (¶204), and wherein a response format comprises a Boolean format requiring responses to be either positive or negative (¶212, train AI model 1230 and apply the AI model to feature vector to output indications of a probability that the feature vector has a particular Boolean property).
It would’ve been obvious to one ordinarily skilled in the art before the effective filing date of the invention to implement at least one response format corresponding to an incremental prompt comprises a Boolean format requiring responses to be either positive or negative to output indications of whether the incremental prompt has an associated desired property (Cameron, ¶212; compare Sharma, Col 16, Rows 40-42, input to a language model is typically natural language parsed into tokens; Col 16, Row 67 – Col 17, Row 10, token sequence is converted into an embedding vector that captures semantic meaning of the text segment represented by the token; Col 17, Rows 32-40, the embeddings are encoded into feature vectors; i.e., the combination applies the LLM to feature vectors representing the series of downstream prompts in the A-R-C-M chain to output probability that the feature vector of a corresponding prompt has a particular Boolean property).
Conclusion
Prior art made of record and not relied upon is considered pertinent to applicant's disclosure:
US 2025/0045028 A1 discloses a pre-trained language model divides a prompt comprising several separate requirements into smaller prompts / sub-prompts for AI processing (¶145).
US 12659376 B1 discloses real-time prompt modification capabilities to update and refine AI processing tasks (Abstract) comprising a task decomposition process of breaking complex prompts into discrete subtasks that can be handled by specialized AI agents (Fig. 4, task decomposition 403, Col 22, Rows 19-27).
US 12393575 B2 discloses decomposing a complex question into one or more sub-questions using co-reference resolution, dependency parsing, and chunking technique (Fig. 3, step 306) to generate and execute LLM code scripts for each sub-question (Fig. 3, steps 308-310).
Any inquiry concerning this communication or earlier communications from the examiner should be directed to examiner Richard Z. Zhu whose telephone number is 571-270-1587 or examiner’s supervisor Hai Phan whose telephone number is 571-272-6338. Examiner Richard Zhu can normally be reached on M-Th, 0730:1700.
Information regarding the status of an application may be obtained from the Patent Application Information Retrieval (PAIR) system. Status information for published applications may be obtained from either Private PAIR or Public PAIR. Status information for unpublished applications is available through Private PAIR only. For more information about the PAIR system, see http://pair-direct.uspto.gov. Should you have questions on access to the Private PAIR system, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative or access to the automated information system, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/RICHARD Z ZHU/Primary Examiner, Art Unit 2654 09/16/2026
