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 .
Continued Examination Under 37 CFR 1.114
A request for continued examination under 37 CFR 1.114, including the fee set forth in 37 CFR 1.17(e), was filed in this application after final rejection. Since this application is eligible for continued examination under 37 CFR 1.114, and the fee set forth in 37 CFR 1.17(e) has been timely paid, the finality of the previous Office action has been withdrawn pursuant to 37 CFR 1.114. Applicant's submission filed on 04/14/26 has been entered. Currently claims 1, 2, 4, 6, 8-12, 14-18, 20, 21 are pending.
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.
Claims 1, 2, 4, 6, 8-12, 14-18, 20, 21, are 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.
For claims 1, 4, 8, 12, the claims recite that the trained machine learning transformer language model is used to provide responses to the received queries, but then the claim subsequently recites that the trained machine learning transformer language model is not being used for the received first query. The claims recites:
using by the one or more computing devices, the trained machine learning transformer language model to provide housing-related responses to the received queries based on classifications of some of the received queries,- including:
receiving, by the one or more computing devices, a first query of the received queries about at least one first housing topic from a first user;
responding, by the one or more computing devices, to the first query without using the trained machine learning transformer language model, including:
This is contradictory in nature because the claim recites that the queries are responded to by using the trained machine learning transformer language model that includes not using the trained model for the first query. One part of the claim recites that the received queries (all of them) are processed by using the trained machine learning transformer language model but the language for the first query seems to recite the opposite. Is the trained machine learning transformer language model used to respond to the queries or just some of the queries? The claim language creates confusion and appears to be contradictory, thereby rendering the claims indefinite.
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, 2, 4, 6, 8-12, 14-18, 20, 21, are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
The claims recite a method (1, 2, 4,6, 8-11), a system (12, 14-17), and a non-transitory computer readable medium (18, 20, 21); therefore, the claims pass step 1 of the eligibility analysis.
For step 2A, the claim(s) recite(s) an abstract idea of providing housing related information in response to a search query that relates to fair housing rule violations. This represents a certain method of organizing human activities type of abstract idea.
Using claim 1 as a representative example that is applicable to claims 4, 12, 18, the abstract idea is defined by the elements of:
analyzing contents of multiple documents with information related to multiple housing topics, including generating an encoded vector representation for each of the multiple documents of information related to housing in the contents of that document;
obtaining information about multiple defined tools each configured to provide information about at least one housing topic; and
classifies received queries as having violations of fair housing rules or not having violations of the fair housing rules, including:
retrieving a first group of positive query examples that include a plurality of first queries lacking fair housing rule violations
generating based at least in part on the plurality of first queries in the first group of positive query examples, a second group of negative query examples that include a plurality of second queries each having one or more fair housing rule violations, including supplying input to use a structure of one or more of the first queries and to generate semantic augmentations that create the second queries with fair housing violations based on legally protected classes and on a defined list of non-compliant phrases associated with violations of the fair housing rules: and
using, as training data, fair housing rules, and the defined list of non-compliant phrases as examples of having identified violations of fair housing rules, and the generated negative query examples as further examples of having identified violations of fair housing rules, and the positive query examples as examples of not having identified violations of fair housing rules:
provide housing-related responses to the received queries based on classifications of some of the received queries,- including:
receiving a first query of the received queries about at least one first housing topic from a first user;
responding to the first query without using a [the] trained machine learning transformer language model, including:
determining that the first query is associated with a violation of the fair housing rules based on comparing the first query to the defined list of non-compliant phrases; and
presenting, in response to the determining that the first query is associated with the violation of the fair housing rules, a first response to the first query indicating an inability to provide further information due to the fair housing rules;
receiving a second query of the received queries about at least one second housing topic from a second user;
responding to the second query by:
determining based at least in part on output that indicates the second query is associated with a violation of fair housing rules, to use information in a second response to the second query indicating an inability to provide further information due to the fair housing rules; and
presenting the second response;
receiving a third query of the received queries about at least one third housing topic from a third user; and
responding to the third query by:
receiving, an indication that the third query is not associated with a violation of the fair housing rules;
identifying, at least one of the multiple documents whose encoded vector representation differs from an encoded version of the third query by at most a defined threshold distance using a defined similarity metric;
generating, using one of the multiple defined tools that is selected as being associated with the third query, additional information from the selected one defined tool;
generating a prompt that includes the third query and the identified at least one document and the additional information and one or more predetermined query-response examples;
generating a third response; and
presenting the determined third response to the third query
The above limitations are reciting a process by which housing related information is being provided to a user. Providing housing related information such as information related to the Fair Housing Act of 1968 is considered to be a legal obligation type of abstract idea where information about the laws and rules regarding housing is provided to a user, where the response can indicate a violation of a fair housing rules. This represents a certain method of organizing human activities type of abstract idea that is simply the act of providing information to a user upon request. Before the invention of modern day computers, humans would ask other humans questions to elicit a response about a particular topic. The concept of receiving a search query (three queries) and providing a response, when it regards fair housing laws and housing rule violation related information, is something people can do and is a certain method of organizing human activities.
For claims 1, 4, 12, the additional elements are the recitation to the use of the one or more computing devices to perform the steps that define the abstract idea, the machine learning transformer model and using it for obtaining output based on input, the training of the machine learning transformer model, and the input into a trained large language model.
For claim 12, the additional elements (in addition to the above) are the one or more hardware processors of one or more computing devices and the claimed memory for storing instructions to perform the recited steps that are considered to define the abstract idea.
Claim 18 recites the additional element of a non-transitory computer readable medium that has stored instructions to perform the claimed steps upon execution by a computing device.
This judicial exception is not integrated into a practical application (2nd prong of eligibility test for step 2A) because the additional elements of the claim when considered individually and in combination with the claim as a whole, amount to the use of a computing device with a processor and memory (non-transitory CRM) that is being used as a tool to execute the abstract idea (MPEP 2106.05(f)), in combination with training of a machine learning model using training data (the training data itself is part of the abstract idea and is information per se) and the use of a large language model that has been trained (which is also machine learning). The claim is simply instructing one to practice the abstract idea by using a generically recited computing device with a processor and memory to perform steps that define the abstract idea and by using a trained machine learning model and by using a trained LLM, both of which are machine learning models that are trained using training data. This does not amount to more than a mere instruction to implement the abstract idea on a computer (MPEP 2106.05(f)) and can be taken as a link to a particular technological environment which is the field of machine learning, see MPEP 2106.05(h). The training of the machine learning transformer model is recited broadly and is not claiming any specific manner of training that would render the claims eligible. For the training limitation, the claims recite:
performing, by the one or more computing devices, the training of the machine learning transformer language model, including using, as training data, the fair housing rules, and the defined list of non-compliant phrases as examples of having identified violations of fair housing rules, and the generated negative query examples as further examples of having identified violations of fair housing rules, and the positive query examples as examples of not having identified violations of fair housing rules;
While the data that is used to train the model is claimed, the training of the machine learning model is itself recited at a high level of generality and does not amount to more than a general instruction for one to use machine learning that is trained on training data. Claiming the type of training data that is used to train the model does not amount to claiming a specific way to actually perform the training of the model such that the claim would contain more than a general link to training a machine learning model. This is indicative of the fact that the claim has not integrated the abstract idea into a practical application and therefore the claim is found to be directed to the abstract idea identified by the examiner.
For step 2B, the claim(s) does/do not include additional elements that are sufficient to amount to significantly more than the judicial exception when considered individually and in combination with the claim as a whole because they do not amount to more than simply instructing one to practice the abstract idea by use of a computing device with a processor and memory (non-transitory CRM) that is being used as a tool to execute the abstract idea (MPEP 2106.05(f)), in combination with training of a machine learning model using training data (the training data itself is part of the abstract idea and is information per se) and the use of a large language model that has been trained (which is also machine learning). The claim is simply instructing one to practice the abstract idea by using a generically recited computing device with a processor and memory to perform steps that define the abstract idea and by using a trained machine learning model and by using a trained LLM. This does not amount to more than a mere instruction to implement the abstract idea on a computer (MPEP 2106.05(f)) and can be taken as a link to a particular technological environment which is the field of machine learning, see MPEP 2106.05(h). The training of the machine learning transformer model is recited broadly and is not claiming any specific manner of training that would render the claims eligible. While the data that is used to train the model is claimed, the training of the machine learning model is itself recited at a high level of generality and does not amount to more than a general instruction for one to use machine learning that is trained on training data. Claiming the type of training data that is used to train the model does not amount to claiming a specific way to actually perform the training of the model such that the claim would contain more than a general link to training a machine learning model. What is claimed by way of the additional elements that are not part of the abstract idea does not amount to reciting significantly more.
For the above reasons claims 1, 4, 12, and 18 are not found to be eligible.
For claims 2, 17, the machine learning transformer language model being a bidirectional encoder representation from transformers model, is an additional element that is taken as a link to the use of a machine learning model, as was set forth for claims 1, 12. The use of machine learning as claimed is considered to be a general link to the field of machine learning, see MPEP 2106.05(f), (h). The claims do not recite any additional elements that provide for integration at the 2nd prong or that provide significantly more at step 2B. Therefore the claims are not considered to be eligible.
For claims 6, 20, the following elements are further defining the same abstract idea of claim 4:
Claims 6 (cl. 20): receiving a third query about a housing-related topic;
determining based on a comparison of the third query to the defined list of non-compliant phrases, that the third query does not include any of the non-compliant phrases of the defined list;
determining based on output from supplying the third query to the trained transformer language model, that the third query is not associated with a violation of fair housing rules;
generating a third response to the third query with information about the housing-related topic; and
providing the third response for the third query
The above steps are part of the abstract idea and are fully capable of being performed by people that are making comparisons (including mentally) and providing a response. The claimed additional element of the one or more computing devices have been treated in the same manner that was set forth for claim 4. The claims do not recite any additional elements that provide for integration at the 2nd prong or that provide significantly more at step 2B. Therefore the claims are not considered to be eligible.
For claim 8, the claimed query examples using few-shot examples in the one or more prompts for the generation of the negative query example is an element that is part of the abstract idea. The manner in which the negative query examples are generated is part of the abstract idea and is fully capable of being performed by a human, including mentally. The claimed use of the LLM has been treated in the same manner as set forth for claim 4, and does not render the claims eligible. The claim does not recite any additional elements that provide for integration at the 2nd prong or that provide significantly more at step 2B. Therefore the claim is not considered to be eligible.
For claims 9, 10, the recited step/function(s) are considered to be part of the abstract idea of the claims. Generating one or more structures for use in the negative query examples is reciting something that can be performed by a person who is mentally coming determining and manually writing down the one or more structures that is claimed. This includes the claimed types of structures recited in claim 10. The claimed additional element of the one or more computing devices and the training of the model (the training of the model is an additional element, the training data is part of the abstract idea) have been treated in the same manner that was set forth for claim 4 to which the applicant is referred. The training of the model is a general link to the field of machine learning and does not render the claims eligible for the same reasons already addressed. The claims do not recite any additional elements that provide for integration at the 2nd prong or that provide significantly more at step 2B. Therefore the claims are not considered to be eligible.
For claims 11, 21, using the language from claim 11 below as a representative example, the abstract idea is being further defined by the following elements:
receiving an additional query related to at least one housing topic;
determining, that the additional query is not associated with a violation of fair housing rules;
comparing the additional query to multiple prior queries from a plurality of users to determine whether any of the multiple prior queries match the additional query based on one or more defined matching criteria, each of the prior queries being associated with at least one document used in providing a prior response to that prior query, the at least one document being selected for use in responding to that prior query from a group of multiple documents having contents including housing-related information;
selecting one or more documents of the multiple documents to use in responding to the additional query, including, if one of the prior queries is determined to match the additional query, using the at least one document associated with that one prior query as the selected one or more documents, and otherwise determining the one or more documents by matching an encoded representation of the additional query to encoded representations of the contents of the multiple documents;
generating a further prompt to supply to a trained large language model that includes the additional query, and one or more predetermined query-response examples, and information about the selected one or more documents, and query-response generation instructions to refuse to provide responses to inputs with references to defined legally protected classes;
generating an additional response to the additional query; and
providing the generated additional response to the additional query with an indication of at least one of the selected one or more documents as a source for the generated additional response
The above elements are simply further defining the abstract idea of claim 4 and are reciting steps that are fully capable of being performed by people. The claimed steps that are taken to provide the response to the user are elements that are part of the abstract idea. The claimed additional element of the one or more computing devices have been treated in the same manner that was set forth for claim 4. The claims do not recite any additional elements that provide for integration at the 2nd prong or that provide significantly more at step 2B. Therefore the claims are not considered to be eligible.
For claim 14, the abstract idea is being further defined by the elements of:
receiving an additional query about a housing-related topic;
determining, based at least in part on use of the additional query, that the additional query is not associated with a violation of fair housing rules;
comparing the additional query to multiple prior queries from a plurality of users to determine whether any of the multiple prior queries match the additional query based on one or more defined matching criteria, each of the prior queries being associated with at least one document used in providing a prior response to that prior query, the at least one document being selected for use in responding to that prior query from a group of multiple documents having contents including housing-related information;
selecting one or more documents of the multiple documents to use in responding to the additional query, including, if one of the prior queries is determined to match the additional query, using the at least one document associated with that one prior query as the selected one or more documents, and otherwise determining the one or more documents by matching an encoded representation of the additional query to encoded representations of the contents of the multiple documents;
generating a prompt to supply to a trained large language model that includes the additional query, and one or more predetermined query-response examples, and information about the selected one or more documents, and query-response generation instructions to refuse to provide responses to inputs with references to defined legally protected classes;
generating an additional response to the additional query based at least in part on o the generated prompt; and
displaying the generated additional response to the additional query with an indication of at least one of the selected one or more documents as a source for the generated additional response
For claim 16, the abstract idea is being further defined by:
modifying, based on determining that the second query is associated with the detected violation of fair housing rules, the second query to remove one or more terms of the query associated with the violation;
selecting one or more documents to use in responding to the modified second query by matching an encoded representation of the modified second query to encoded representations of contents of multiple documents including housing-related information;
generating a prompt that includes the modified second query and further includes information about the selected one or more documents and further includes query-response generation instructions to refuse to provide responses to inputs with references to defined legally protected classes; and
generating the second response based at least in part on the generated prompt, and
wherein providing of the second response includes providing the second response with an indication of at least one of the selected one or more documents as a source for the second response
Claims 14 and 16 are further defining the abstract idea of the claims and are reciting steps that people can perform. Modifying a query to remove words can be done manually, selecting documents can be done by a person, generating a prompt can be done by a person also. The generation of the response can be performed by a person. The claim is simply providing more about the abstract idea. The additional elements of using the large language model for output and the stored instructions that cause a processor to perform the recited steps, that have been treated in the same manner as set forth for claim 12. The claims do not recite any additional elements that provide for integration at the 2nd prong or that provide significantly more at step 2B. Therefore the claims is/are not considered to be eligible.
For claim 15, the abstract idea is being further defined by the elements reciting the receiving of the queries and the display of the first and second responses. The claim recites the additional elements of the use of a chatbot with a GUI. A chatbot is a computer program that can interact with a user to receive input and provide a response via a GUI. All computers have an interface that allows for data input and data display. The claimed chatbot and use of the GUI is considered to be an instruction for one to use a computer with an interface and a chatbot to perform the abstract idea steps of receiving queries (input from a user) and displaying the responses to the queries. This is claiming a general link to computer implementation for the abstract idea, see MPEP 2106.05(f). The claim does not recite any additional elements that provide for integration at the 2nd prong or that provide significantly more at step 2B. Therefore the claim is not considered to be eligible.
Therefore, for the above reasons, claims 1, 2, 4, 6, 8-12, 14-18, 20, 21, are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Response to arguments
The traversal of the 35 USC 101 rejection is not persuasive.
On pages 17-18 of the reply the applicant argues that the claims are generally directed to particular ways of training a machine learning model. The argument is not persuasive. The examiner notes that the claims do not recite a particular way that the model is being trained. The extent of the training that is recited in the claims is set forth by the language “performing by the one or more computing devices, the training of the machine learning transformer language model including using as training data…”, where the applicant simply claims the training data that is used to train the model but is not otherwise claiming any particular way in which the model is being trained. The claims do not recite a particular way to train the model. The claims recite that the training data includes positive query examples that do not have fair housing rule violations and negative query examples that do contain fair housing rule violations; however, specifying the training data as being positive query responses and negative query responses is not claiming a particular way in which the transformer language model is being trained such that it would render the claims eligible (in view of an improvement to training a machine learning model, in view of the Desjardins decision). Machine learning models work by being trained, that is how they function. Machine learning models can be trained using different training data sets where the training data is something that depends on the field of use that the model is being used with, where the actual training of the machine learning model is performed the same way regardless of the what the training data represents. The argument that the claims recite a particular way to train the transformer language model is not persuasive because the claims simply recite that the model is trained using training data. The training data itself is ineligible subject matter that is part of the abstract idea and the training is not recited with any specificity. The claimed training of the model amounts to a general link to the field of machine learning.
The applicant argues that the claims recite specific techniques for generating the training data used in the training of the model. This is not persuasive. People can generate training data manually by performing the recited steps. The claim recites that positive query examples are used to generate negative query examples that each have one or more fair housing violations. The claimed manner in which the examples are generated is considered to be part of the abstract idea. The generation of the training data or deciding on what data to have as training data, is not a specific way in which the model is being trained that would render the claims eligible. Query examples are used to generate other query examples and they are used to train the model in a non-limiting manner that does not require any specific way of training. The central argument from the applicant that the claims recite a specific way to train a machine learning model is not persuasive because the training of the claims is not specific but is recited in a general sense. Again, the claim recites performing by the one or more computing devices, the training of the machine learning transformer language model including using as training data, which is not claiming a specific way in which the model is being trained. For this reason the argument and the reliance upon the Desjardins decision and memo is not persuasive. The claims do not recite a specific way in which a machine learning model is being trained such that the model is being improved.
On page 19, the applicant argues that particular ways of training a machine learning model improve the functioning of a computer by addressing problems of having sufficient training data by generating training data in an automated manner. The examiner notes that the generation of training data does not amount to a new way to train a model. The generation of a data collection can be done by people who populate a memory with data to be used to train a model. That process does not define anything specific to the training of the model by defining how the training data is generated. Automating the process of creating training data that can otherwise be created by people does not amount to more than an instruction for one to use machine learning as a tool to execute the abstract idea. For this reason the argument is not persuasive.
The applicant also argues that having a list of deny phrases improves the functioning of a computer by reducing computational resources needed to respond to a query. This is not persuasive. The language at issue for this limitation reads as follows:
responding to the first query without using the trained machine learning transformer language model, including:
determining that the first query is associated with a violation of the fair housing rules based on comparing the first query to the defined list of non-compliant phrases; and
presenting, in response to the determining that the first query is associated with the violation of the fair housing rules, a first response to the first query indicating an inability to provide further information due to the fair housing rules;
receiving a second query of the received queries about at least one second housing topic from a second user;
The claim recites that the trained model is not used for the first query, which is not doing anything to reduce the resources on the model other than the fact that the model is not being used for that given query. Nothing is claimed about a decision making process that results in the system deciding to not use the model, such that it would be acting as a filter as the applicant argues. The claim recites that the trained model is used to provide responses to the queries, and also recites that the trained model is not used for the first query. That itself is not improving technology in any manner and does not result in computational resources being saved. The specification also does not disclose anything more than a general allegation that by using the disclosed system of the specification, it improves technology. Paragraph 038 discloses:
The described techniques provide various benefits in various embodiments, including to significantly improve the identification and use of responsive information to specified queries for housing-related information, including queries specified in a natural language format. Such automated techniques allow such responsive answer information to be generated much more quickly and efficiently than previously existing techniques (e.g., using less storage and/or memory and/or computing cycles) and with greater accuracy, based at least in part on using the described techniques for restricting responses to particular housing-related topics and providing citations as to sources of the information in the responses, such as using a defined list of housing-related topics, a defined group of documents with contents related to those topics, a defined group of tools that provide information related to those topics, etc.
The above disclosure is general in nature and does not attribute any specific improvement in technology to any specific claim limitation or combination of limitations. The specification makes little more than a general allegation that by using the disclosed techniques (the claims do not include everything in the specification), less storage is needed and the system responds faster. The reason people use computers to do things is that they are faster and more accurate than people. That aspect of the argument nis nothing more than automation by a computer of the abstract idea, which is not persuasive, see MPEP 2106.05(f). As to the reduced storage needs, nothing is disclosed about how this result is achieved or what aspect of the invention is responsible for providing for reduced storage needs. Other than a general allegation that the specification and its teachings can improve technology by reducing storage needs, there is nothing else disclosed that equates anything in the claim to an improvement in technology. The argument is not persuasive.
On page 20, the applicant argues that the claims are directed to techniques for training a machine learning model before using the machine learning model to provide responses to queries. The applicant argues that this improves the operation of the computer system by using non-compliant phrases to respond to some of the queries without using the trained model. What is being argued is not reflected in the claims because the claims are just reciting that the trained model is not used for the first query. That is not the same as the argued use of non-compliant phrases and determining whether or not to use the model, and then either using the model or not using the model to respond to a query. The claim recites that the trained model is used to respond to queries, and also recites that the model is not used for the first query. The responding to the first query is positively claimed as being done by not using the trained model. The use of the non-compliant phrases and their comparison to the query is claimed as being done without using the model. That is not providing for an improvement to technology and is not claiming a decision making process of whether or not to use the model, it just recites not using the model for a first query but using the model for the 2nd and 3rd queries. For this reason the argument is not commensurate with the scope of the claims and is not persuasive.
On pages 21-22 of the reply the applicant argues that the claims recite a particular way to train a machine learning model that serves to improve the functioning of a computer system. As has already been addressed, this is not persuasive. The examiner again notes that the claims do not recite a particular way that the model is being trained. The extent of the training that is recited in the claims is set forth by the language “performing by the one or more computing devices, the training of the machine learning transformer language model including using as training data…”, where the applicant simply claims the training data that is used to train the model but is not otherwise claiming any particular way in which the model is being trained. The claims do not recite a particular way to train the model. The argument that the claims recite a particular way to train the transformer language model is not persuasive.
The applicant argues that example 39 of the USPTO examples shows that training of a model is eligible subject matter. This is not persuasive because the pending claims do not recite a specific way to train a model and because example 39 did not contain a judicial exception at step 2A, so the analysis stopped, whereas the pending claims do recite a judicial exception. The pending claims are only claiming the training in a general sense and do not recite an improved manner in which a machine learning model is being trained. Also, unlike example 39, the claimed invention recites to a judicial exception so the analysis moves to the 2nd prong and step 2B if necessary, and is more analogous to example 47 from the eligibility guidance than example 39. The pending claims are not similar to example 39 from the USPTO guidance. The applicant is referred to example 47 of the USPTO guidance and specifically to claims 2 and 3 for guidance on how the USPTO treats machine learning claims. Example 39 did not contain a judicial exception which was the reason it was considered to be eligible. Example claim 39 was not found to be eligible due to claiming an improved manner by which a model is trained. To argue such is not supported by the example itself that clearly states that because no judicial exception is recited at step 2A, the claim is eligible. The same cannot be said for the pending claims that do recite a judicial exception and that requires a 2nd prong and step 2B analysis, unlike example 39. The reliance upon example 39 is not persuasive to show that the claims are reciting eligible subject matter.
With respect to the claimed manner in which the training data is being generated, as has already been addressed, the claimed elements can be performed by people and the generation of the training data is not something that amounts to a particular way to train a model such that at the 2nd prong or step 2B the claims would be eligible. According to the specification, the generation of the negative query examples is performed by using a large language model, such as commercially available GPT-4. The applicant does not have a particular way in which the negative examples are created other than by using a commercially available LLM, which is nothing more than linking the creation of the negative query examples to machine learning.
Paragraph 031: “transformer language model compliance classifier that is trained using a positive set of queries (e.g., using actual received plugin queries, SEO queries, natural language search queries, NLS (National Language Support) queries, and real estate acronyms to create a positive set) and a negative set of queries (e.g., using GPT-4 or another LLM with detailed instructions about fair housing to generate a large quantity, such as ~10K, noncompliant queries using the structure of the positive queries but with LLM-based induced semantic augmentations for noncompliance, such as by using the deny list and legally protected classes for the noncompliance augmentations, and optionally using few-shot prompting) and optionally a positive augmentation dataset for the positive set of queries (e.g., using similar techniques with GPT-4 or another LLM for creating tricky but compliant queries for defined classes including disability, familial status, veteran status, receipt of public assistance, etc.).
With respect to the argument on page 22 for Example 39, the examiner again notes that example 39 was not found to be eligible due to claiming an improved manner by which a model is trained or due to a particular way to generate training data. Example 39 was found to be eligible because at step 2A it was determined that no judicial exception was recited. From example 39:
The claim does not recite any of the judicial exceptions enumerated in the 2019 PEG. For instance, the claim does not recite any mathematical relationships, formulas, or calculations. While some of the limitations may be based on mathematical concepts, the mathematical concepts are not recited in the claims. Further, the claim does not recite a mental process because the steps are not practically performed in the human mind. Finally, the claim does not recite any method of organizing human activity such as a fundamental economic concept or managing interactions between people. Thus, the claim is eligible because it does not recite a judicial exception.
The applicant is arguing example 39 as if it was eligible due to an improvement in technology that was flushed out at the 2nd prong or step 2B, which is not correct. Example 39 is used to show an example of where a claim can be eligible because at step 2A, if no judicial exception is recited in a claimed invention, the claim is eligible. The argument that the claim was eligible due to an improved way to generate training data and train a model is not correct and is mischaracterization of example 39 and why that hypothetical example was found to be eligible. The reliance upon example 39 is not persuasive to show that the instant pending claims are eligible. The arguments are not persuasive and the 101 rejection is being maintained.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to DENNIS WILLIAM RUHL whose telephone number is (571)272-6808. The examiner can normally be reached M-F 7am-3:30pm.
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, Jessica Lemieux can be reached at 5712703445. 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.
/DENNIS W RUHL/ Primary Examiner, Art Unit 3626
Prosecution Insights