Notice of Pre-AIA or AIA Status
1. The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Claim Objections
2. Claim 1 is objected to because of the following informalities:
Claim 1 recites “an reception unit” in line 4. Please change this to “a reception unit”
Appropriate correction is required.
Claim Rejections - 35 USC § 102
3. In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The following is a quotation of the appropriate paragraphs of 35 U.S.C. 102 that form the basis for the rejections under this section made in this Office action:
A person shall be entitled to a patent unless –
(a)(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.
4. Claims 1 and 9-10 are rejected under 35 U.S.C. 102(a)(2) as being anticipated by Karmakar (US 2026/0010963).
Regarding Claim 1:
Karmakar discloses an information processing apparatus (Karmakar: ¶14 discloses an automated query response information generation (AQRIG) system executing on one or more computing systems and interacting with user client devices) comprising:
a generation unit configured to generate non-response target information indicating a non-response target (Karmakar: ¶20 discloses using a large language model (LLM) to generate approximately 10,000 noncompliant queries using a deny list and legally protected classes as noncompliance augmentations. The generated noncompliant queries correspond to the nonresponse target information because they provide examples indicating queries that should be rejected);
an reception unit configured to receive information indicating a request of a user (Karmakar: ¶15 discloses that a user supplies a housing related query through a natural language input interface and that the interface provides the query to the fair housing query filter component, Karmakar ¶21 discloses that during operation the filter component receives user query 191, the received user query corresponds to information indicating a request of a user);
and a determination unit configured to determine, based on the non-response target information generated by the generation unit, whether the request indicated by the information received by the reception unit is a request concerning the non-response target (Karmakar: ¶20 discloses using a transformer LLM model classifier using negative query examples to fair house violations; ¶31 discloses generating those negative examples using a LLM. ¶21 discloses submitting a received user query to the trained classifier to determine whether the question should be rejected or accepted. The determination is based on the generated non-response target information because the generated noncompliant queries establish the negative training examples from which the classifier learns its rejection boundary).
Regarding Claim 9:
Claim 9 has been analyzed with regards to claim 1 (see rejection above) and is rejected for the same reasons of anticipation set forth above.
Regarding Claim 10:
Claim 10 has been analyzed with regards to claim 1 (see rejection above) and is rejected for the same reasons of anticipation set forth above.
Claim Rejections - 35 USC § 103
5. In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
The factual inquiries for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or nonobviousness.
6. Claims 2 and 3-4 are rejected under 35 U.S.C. 103 as being unpatentable over Karmakar in view of Tan (US 2020/0364300).
Regarding Claim 2:
Karmakar further discloses the information processing apparatus according to claim 1, wherein the generation unit generates the non-response target information based on non-response category information indicating a non-response category that is a category different from a designated category that is a category designated by the user (Karmakar: ¶15 discloses receiving a user’s housing related query and ¶16 discloses determining a housing related topic corresponding to the user query. The topic communicated through the query corresponds to the designated category designated by the user ¶31 discloses generating the noncompliant query examples using a deny list and legally protected classes as noncompliance augmentations. The deny-list and protected class information indicates categories different from the designated housing topic and the generated noncompliant queries correspond to the non-response target information)
Karmakar does not explicitly disclose:
the non-response category being correlated with the designated category in advance as a non-response category.
However, Tan discloses:
the non-response category being correlated with the designated category in advance as a non-response category (Tan: ¶46 discloses the off-topic classifier for the current domain is trained using negative samples that are not from the current domain or its child domains. Those negative samples represent categories different from the user designated active domain. Tan further discloses in ¶83 that sampling off topic examples from outside the current domain and selecting suitable off topic instances using semantic similarity and training the off-topic classifier for that particular domain using the selected negative examples. In other words, this establishes the association beforehand between the active domain and outside domains treated as off topic).
Karmakar and Tan are combinable because they are from the same field of endeavor, i.e., both disclose computerized conversational systems that receive natural language user requests, classify the request by topic or domain and control the corresponding responses. It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to disclose modifying Karmakar to organize its approved topics into predefined domains and for each user designated domain, generate its question rejection classifier using negative examples sampled from different domains outside the designated domain as taught by Tan. This modification would have improved the accuracy with which Karmakar distinguishes requests concerning the designated topic from requests concerning other topics, particularly where the categories are semantically similar, the motivation for doing so is explained in Tan ¶23: “different domains may share semantically-close classification labels, such that choosing a correct domain from which to classify an utterance impacts the accuracy of classification.”
Regarding Claim 3:
The proposed combination of Karmakar and Tan further discloses the information processing apparatus according to claim 2, wherein the generation unit generates the non-response target information using a language model (Karmakar: ¶31 discloses using GPT-4 or another LLM to generate approximately 10,000 noncompliant queries using a deny list and legally protected classes for non-argumentations. These LLM generated noncompliant queries are interpreted as the non-response target information under the same interpretation applied to claim 1).
Regarding Claim 4:
The proposed combination of Karmakar and Tan further discloses the information processing apparatus according to claim 3, wherein the generation unit generates information indicating a risk in the non-response category as the non-response target information using the language model (Karmakar ¶20 discloses negative query examples corresponding to fair housing rule violations which are used to train the compliance classifier to reject user queries presenting those violations. ¶31 discloses another LLM together with the deny list and legally protected class information to generate approximately 10k noncompliant queries through LLM induced noncompliance augmentations. Each generated noncompliant query constitutes information indicating a risk in the nonresponse category because it exemplifies a query presenting the risk of a fair housing violation associated with the deny list or protected class category. These LLM generated queries are the same nonresponse target information identified in claims 1 and 3 and are used to train the classifier guardrail to recognize and reject requests presenting that risk).
7. Claim 5 is rejected under 35 U.S.C. 103 as being unpatentable over Karmakar in view of Tan and further in view of Park et al. “Generative Agents: Interactive Simulacra of Human Behavior” herein Park.
Regarding Claim 5:
The proposed combination of Karmakar and Tan further discloses the information processing apparatus according to claim 4, wherein the generation unit includes:
a first generation processing unit configured to generate risk information indicating a risk in the non- response category using the language model (Karmakar: ¶20 discloses negative query examples corresponding to fair housing rule violation, ¶31 discloses using GPT-4 or another LLM, together with a deny list and legally protected class information to generate approximately 10,000 noncompliant queries through noncompliance augmentations. Each generated noncompliant query constitutes risk information indicating a risk in the nonresponsive category because it exemplifies a query presenting the risk of a fair housing violation associated with the deny list or protected class category);
The proposed combination of Karmakar and Tan does not explicitly disclose:
and a second generation processing unit configured to aggregate the risk information generated by the first generation processing unit into a preset number or less risk information as the non-response target information using the language model.
However, Park discloses:
and a second generation processing unit configured to aggregate the risk information generated by the first generation processing unit into a preset number or less risk information as the non-response target information using the language model (Park: Section 4, discloses a separate reflection processing unit, Section 4.2 discloses gathering multiple natural language records and prompting the language model to generate five high level insights from those records. The synthesis of multiple records into higher level insights corresponds to aggregating the information and the five requested insights correspond to a preset number or less information. Applied to Karmakar’s generated noncompliant risk examples and Park’s five higher level risk insights map to the nonresponse target information identifying the risks for which a response should be withheld).
Karmakar and Park are analogous art because both disclose using language models to process collections of natural language records into information that guides subsequent system behavior. It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify Karmakar by providing its generated noncompliant risk examples to Park’s reflection process and using GPT-3.5 turbo to aggregate those examples into five higher level risk statements as the non-response target information. Park provides the motivation because “the most relevant pieces of the agent’s memory are retrieved and synthesized when needed” in Section 4.
8. Claims 6 is rejected under 35 U.S.C. 103 as being unpatentable over Karmakar in view of Tan and further in view of Jalaluddin (US 2022/0171930).
Regarding Claim 6:
The proposed combination of Karmakar and Tan further discloses the information processing apparatus according to claim 3, wherein the generation unit generates information (Karmakar: ¶31 discloses using fair housing rules and protected classes to generate a deny list, this deny list is used in an LLM to generate approximately 10,000 noncompliant queries. Those queries train the rejection classifier. Wherein the non-response target information is interpreted as the 10,000 noncompliant queries).
Karmakar and Tan do not explicitly disclose:
partially including a feature word extracted from the non-response category information.
However, Kadam explicitly discloses:
partially including a feature word extracted from the non-response category information (Jalaluddin: ¶114 discloses identifying keywords within original utterances and generating negative examples containing each identified keyword. ¶129 and ¶133 further disclose generating out of domain (OOD) examples using a large language model. Applied to Karmakar a keyword would be extracted from the deny-list or protected class category information and the language model generates noncompliant query examples containing that keyword).
Karmakar and Jalaluddin are in the same field of endeavor, i.e., both disclose systems and methods for generating and using negative query examples to train chatbot utterance classifiers. It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to disclose modifying Karmakar’s language model to generate the noncompliant queries by partially identifying keywords from the deny list or protected class information and generate noncompliant queries containing the identified keywords as taught by Jalaluddin. The modification would improve the classifier’s ability to distinguish prohibited uses of a keyword from irrelevant or permissible context. Jalaluddin explains: “consequently, the machine learning model will better generalize and learn that keywords should only matter when the keywords are used in contexts similar to the training data sets of utterances” in ¶30.
9. Claim 7-8 are rejected under 35 U.S.C. 103 as being unpatentable over Karmakar in view of Tan and further in view of Carte (US 2025/0103624).
Regarding Claim 7:
The proposed combination of Karmakar and Tan further discloses the information processing apparatus according to claim 2, comprising a selection unit (Karmakar: ¶31 discloses generating the non response target information, the 10.000 examples and using those examples to train the compliance classifier) configured to
wherein the determination unit determines, based on (Karmakar: ¶20-21 disclose straining the compliance classifier with the generated negative query examples and subsequently using the trained classifier to accept or reject a received query).
The proposed combination of Karmakar and Tan does not explicitly disclose:
select, based on non-response accuracy for each combination of two or more pieces of the non-response target information among a plurality of pieces of the non-response target information generated by the generation unit, two or more pieces of the non-response target information used by the determination unit among the plurality of pieces of the non-response target information.
However, Carta discloses
select, based on non-response accuracy for each combination of two or more pieces of the non-response target information among a plurality of pieces of the non-response target information generated by the generation unit, two or more pieces of the non-response target information used by the determination unit among the plurality of pieces of the non-response target information and wherein the determination unit determines, based on the two or more pieces of non-response target information selected by the selection unit (Carta: ¶45-48 discloses candidate prompts from different combinations of examples and further evaluates each candidate using validation data and selects the lowest loss candidate. ¶49-50 repeats the process until multiple examples, three are incorporated into the prompt used by the target model. Applied to Karmakar, the selected combination of noncompliant queries is used by the prompt-based refusal guardrail to determine whether a user request should be refused).
Karmakar, Tan and Carta are analogous art because each concerns improving the accuracy with which natural language requests are processed by conversational or language model systems. Karmakar already generates a large plurality of noncompliant query examples, implements a prompt-based refusal guardrail and selects examples for inclusion in an enhanced large language model prompt. It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to disclose modifying Karmakar’s prompt generation process according to Carta by treating Karmakar’s generated noncompliant queries as candidate prompt examples, evaluating candidate multi example combinations on compliant and noncompliant validation requests using Tan’s disclosed accuracy metrics, selecting the best performing combination and employing that selected combination in the prompt used for compliance determination. This would reduce the inclusion of redundant or ineffective noncompliant examples while improving rejection accuracy and reducing prompt processing costs. Carta explicitly explains in ¶23, that this approach would “allow for exploration of textual space more efficiently and systematically, thereby increasing accuracy and speed during performance of a variety of tasks.”
Regarding Claim 8:
The proposed combination of Karmakar, Tan and Carta further discloses the information processing apparatus according to claim 7, comprising an evaluation unit configured to evaluate non-response accuracy for each combination of the two or more pieces of non-response target information, wherein the selection unit selects, based on an evaluation result by the evaluation unit, the two or more pieces of non-response target information used by the determination unit (Carta: ¶47-48 calculates a validation loss for each candidate combinatorial prompt and selects the candidate having the lowest loss. ¶52 evaluates each candidate prompt and replaces the current prompt with the best performing candidate).
Karmakar, Tan and Carta are analogous art because each concerns improving the accuracy with which natural language requests are processed by conversational or language model systems. Karmakar already generates a large plurality of noncompliant query examples, implements a prompt-based refusal guardrail and selects examples for inclusion in an enhanced large language model prompt. It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to disclose modifying Karmakar’s prompt generation process according to Carta by treating Karmakar’s generated noncompliant queries as candidate prompt examples, evaluating candidate multi example combinations on compliant and noncompliant validation requests using Tan’s disclosed accuracy metrics, selecting the best performing combination and employing that selected combination in the prompt used for compliance determination. This would reduce the inclusion of redundant or ineffective noncompliant examples while improving rejection accuracy and reducing prompt processing costs. Carta explicitly explains in ¶23, that this approach would “allow for exploration of textual space more efficiently and systematically, thereby increasing accuracy and speed during performance of a variety of tasks.”
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to IAN SCOTT MCLEAN whose telephone number is (703)756-4599. The examiner can normally be reached "Monday - Friday 8:00-5:00 EST, off Every 2nd Friday".
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, Hai Phan can be reached at (571) 272-6338. 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.
/IAN SCOTT MCLEAN/Examiner, Art Unit 2654
/HAI PHAN/Supervisory Patent Examiner, Art Unit 2654
Prosecution Insights