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 .
DETAILED ACTION
1. Claims 1 - 20 are pending. Claims 1, 9, 17 are independent.
2. This application was filed on 8-22-2024.
Claim Rejections - 35 USC § 103
3. 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.
4. Claims 1, 9, 10, 12, 15 - 18, 20 are rejected under 35 U.S.C. 103 as being unpatentable over Dong et al. (US PGPUB No. 20240184912) in view of Flesher et al. (US PGPUB No. 20090055477).
Regarding Claims 1, 9, 17, Dong discloses a computer-implemented method for controlled and validated interactions with one or more predictive language models and a computing system for controlled and validated interactions with one or more predictive language models, the computer-implemented method comprising:
a) receiving, via one or more processors, user input including a text prompt string; (see Dong paragraph [0016]: operable to receive first text data 105, which includes context-aware labels in place of sensitive text.)
b) processing, via the one or more processors, the text prompt string to generate a sanitized text prompt string; (see Dong paragraph [0016]: operable to receive first text data 105 and output sanitized second text data 145, which includes context-aware labels in place of sensitive text.) and
d) processing, via the one or more processors, the text output string to generate a sanitized text output string. (see Dong paragraph [0016]: pipeline 100 includes sensitive token identification module 110, context data structure generation module 120, context-based token classification module 130, and post-processing module 140. Generally speaking, pipeline 100 is operable to receive first text data 105 and output sanitized second text data 145, which includes context-aware labels in place of sensitive text.)
Dong does not specifically disclose for c) receiving text output string corresponding to processing of sanitized text prompt string using the one or more predictive language models, and for e) causing the sanitized text output string to be transmitted via an electronic network.
However, Flesher discloses:
c) receiving, via the one or more processors, a text output string corresponding to processing of the sanitized text prompt string using the one or more predictive language models; e) causing, via the one or more processors, the sanitized text output string to be transmitted via an electronic network. (see Flesher paragraph [0023]: sanitizing the input message to generate a first sanitized message for transmission to the first recipient; and sanitizing the input message to generate a second sanitized message, different than the first sanitized message, for transmission to the second potential recipient. In accordance with the present invention, a substantially unlimited number of recipients can be accommodated in this regard.)
It would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to modify Dong for c) receiving text output string corresponding to processing of sanitized text prompt string using the one or more predictive language models, and for e) causing the sanitized text output string to be transmitted via an electronic network as taught by Flesher. One of ordinary skill in the art would have been motivated to employ the teachings of Flesher for the benefits achieved from enhanced security from masking of secure information during the processing of text based information. (see Flesher paragraph [0023])
Furthermore, for Claim 9, Dong discloses wherein one or more processors, one or memories having stored thereon computer-executable instructions that when executed cause the computing system to perform operations. (see Dong paragraph [0080]: Program instructions may be stored on a “non-transitory, computer-readable storage medium” or a “non-transitory, computer-readable medium.” The storage of program instructions on such media permits execution of the program instructions by a computer system.; paragraph [0100]: A structure can be said to be “configured to” perform some task even if the structure is not currently being operated. Thus, an entity described or recited as being “configured to” perform some task refers to something physical, such as a device, circuit, a system having a processor unit and a memory storing program instructions executable to implement the task,)
Furthermore, for Claim 17, Dong discloses wherein a non-transitory computer readable medium having stored thereon computer-executable instructions that when executed cause a computer to perform operations. (see Dong paragraph [0080]: Program instructions may be stored on a “non-transitory, computer-readable storage medium” or a “non-transitory, computer-readable medium.” The storage of program instructions on such media permits execution of the program instructions by a computer system.; paragraph [0100]: A structure can be said to be “configured to” perform some task even if the structure is not currently being operated. Thus, an entity described or recited as being “configured to” perform some task refers to something physical, such as a device, circuit, a system having a processor unit and a memory storing program instructions executable to implement the task,)
Regarding Claims 10, 18, Dong discloses the computing system of claim 9 and the non-transitory computer readable medium of claim 17, the memories having stored thereon instructions that when executed cause the computing system to: identify sensitive, private, or harmful data in the text prompt string. (see Dong paragraph [0026]: Named entity recognition (NER) module 210 is operable to perform named entity recognition and identify sensitive tokens within textual data that are predicted to correspond to sensitive data.; paragraph [0025]: “Sensitive data” refers to any type or class of data for which there are additional restrictions on handling relative to non-sensitive data. Sensitive data can commonly include personally identifying data for a user (e.g., Social Security Number, passport number, driver's license number), financial information for a user (e.g., bank account number, credit card number), health information for a user, confidential data about a company, etc.)
Regarding Claims 12, 20, Dong discloses the computing system of claim 9 and the non-transitory computer readable medium of claim 17, the memories having stored thereon instructions that when executed cause the computing system to perform operations, wherein identify task-specific information from user input text prompts and generate task-specific templates by customizing configurations of the one or more predictive language models with pre-configured prompts tailed to specific tasks with role-based and instruction indicators, variable placeholders, and structuring elements. (see Dong paragraph [0100]: different entities (which may variously be referred to as “units,” “circuits,” other components, etc.) may be described or claimed as “configured” to perform one or more tasks or operations. This formulation-[entity] configured to [perform one or more tasks]—is used herein to refer to structure (i.e., something physical). More specifically, this formulation is used to indicate that this structure is arranged to perform the one or more tasks during operation. A structure can be said to be “configured to” perform some task even if the structure is not currently being operated. Thus, an entity described or recited as being “configured to” perform some task refers to something physical, such as a device, circuit, a system having a processor unit and a memory storing program instructions executable to implement the task,; paragraph [0101]: various units/circuits/components may be described herein as performing a set of task or operations. It is understood that those entities are “configured to” perform those tasks/operations,)
Regarding Claim 15, Dong discloses the computing system of claim 9, the memories having stored thereon instructions that when executed cause the computing system to: receive the user input and pair it with the text prompt string when processing the text prompt string to generate the sanitized text prompt string using the one or more predictive language models. (see Dong paragraph [0033]: Question-answer generation module 220, in the illustrated embodiment, is thus operable to generate, for the set of tokens identified as sensitive, questions and corresponding answers, which may be referred to as question-answer pairs.; paragraph [0036]: In some situations, when a more diverse set of questions is desired, NQG may be used over question templates given that NQG generates different questions using different random seeds. In some embodiments, NQG utilizes prompt-based methods to generate questions. Prompt-based methods utilize language models that model the probability of text directly; the original input text is modified using a template to generate a textual string prompt that has some unfilled information, and the language model is used to probabilistically fill the unfilled information to obtain a final string, from which an output can be derived.)
5. Claims 2, 11, 19 is rejected under 35 U.S.C. 103 as being unpatentable over Dong in view of Flesher and further in view of Dunagen et al. (US PGPUB No. 20090228431).
Regarding Claim 2, Dong discloses the computer-implemented. method of claim 1.
Dong does not specifically disclose for a) aligning one on more outputs with a user-provided knowledge base, wherein the user-provided knowledge base comprises, historical customer interactions, product preferences, and feedback, serves as reference data to which model output must align with to ensure output consistency, and for b) adjusting one or more parameters of the predictive language models using retrieval augmentation generation or recurrent binary embedding to enhance model performance by adjusting model parameters to ensure consistent outputs.
However, Dunagan discloses further comprising:
a) aligning, via the one or more processors, one on more outputs with a user-provided knowledge base, wherein the user-provided knowledge base comprises, historical customer interactions, product preferences, and feedback, serves as reference data to which model output must align with to ensure output consistency; b) adjusting, via the one or more processors, one or more parameters of the one or more predictive language models using retrieval augmentation generation or recurrent binary embedding to enhance model performance by adjusting model parameters to ensure consistent outputs with the user-provided knowledge base. (see Dunagan paragraph [0066]: Each CEDR operator 800 in an applications and data space can include the consistency monitor 822 and the operational module 824. The consistency monitor 822 decides whether to block the input stream in an alignment buffer 826 until output may be produced which upholds the desired level of consistency. The operational module 824 can compute the output stream based on incoming tuples and current operator state 828.; (output consistency); paragraph [0085]: Information may be collected for use in answering subsequent queries. For example, a reservoir of historical information may be maintained for levels of a stack. An application developer can be aware of such a reservoir and chose to collect information that is consistent across reservoirs.; paragraph [0086]: The knowledge base 1050 may be maintained in a query/strategy space or in an applications/data space. A knowledge base 1050 may be maintained in a distributed manner, for example, across a plurality of reservoirs.)
It would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to modify Dong for a) aligning one on more outputs with a user-provided knowledge base, wherein the user-provided knowledge base comprises, historical customer interactions, product preferences, and feedback, serves as reference data to which model output must align with to ensure output consistency, and for b) adjusting one or more parameters of the predictive language models using retrieval augmentation generation or recurrent binary embedding to enhance model performance by adjusting model parameters to ensure consistent outputs as taught by Dunagan. One of ordinary skill in the art would have been motivated to employ the teachings of Dunagan for the benefits achieved from the flexibility of a system to align outputs enabling consistent outputs enabling consistent outputs with a knowledge base information. (see Dunagan paragraph [0066])
Regarding Claims 11, 19, Dong discloses the computing system of claim 9 and the non-transitory computer readable medium of claim 17, the memories having stored thereon instructions that when executed cause the computing system to perform operations. (see Dong paragraph [0080]: Program instructions may be stored on a “non-transitory, computer-readable storage medium” or a “non-transitory, computer-readable medium.” The storage of program instructions on such media permits execution of the program instructions by a computer system.; paragraph [0100]: A structure can be said to be “configured to” perform some task even if the structure is not currently being operated. Thus, an entity described or recited as being “configured to” perform some task refers to something physical, such as a device, circuit, a system having a processor unit and a memory storing program instructions executable to implement the task,)
Dong does not specifically disclose align outputs generated by the predictive language models with a user-provided knowledge base.
However, Dunagan discloses wherein align outputs generated by the one or more predictive language models with a user-provided knowledge base. (see Dunagan paragraph [0066]: Each CEDR operator 800 in an applications and data space can include the consistency monitor 822 and the operational module 824. The consistency monitor 822 decides whether to block the input stream in an alignment buffer 826 until output may be produced which upholds the desired level of consistency. The operational module 824 can compute the output stream based on incoming tuples and current operator state 828.; (output consistency); paragraph [0085]: Information may be collected for use in answering subsequent queries. For example, a reservoir of historical information may be maintained for levels of a stack. An application developer can be aware of such a reservoir and chose to collect information that is consistent across reservoirs.; paragraph [0086]: The knowledge base 1050 may be maintained in a query/strategy space or in an applications/data space. A knowledge base 1050 may be maintained in a distributed manner, for example, across a plurality of reservoirs.)
It would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to modify Dong for align outputs generated by the predictive language models with a user-provided knowledge base as taught by Dunagan. One of ordinary skill in the art would have been motivated to employ the teachings of Dunagan for the benefits achieved from the flexibility of a system to align outputs enabling consistent outputs with a knowledge base information. (see Dunagan paragraph [0066])
6. Claim 3 is rejected under 35 U.S.C. 103 as being unpatentable over Dong in view of Flesher and further in view of Lin et al. (US PGPUB No. 20230379138) and Dunagen et al. (US PGPUB No. 20090228431) and Luft et al. (US PGPUB No. 20090034426).
Regarding Claim 3, Dong discloses the computer-implemented method of claim 1.
Dong does not specifically disclose for a) applying type and structure constraints to outputs of the predictive language models to ensure conformity, and for b) identifying instances of lack of factuality, toxicity, prohibited type or prohibited structure in the text prompt string, and for f) processing type and structure constraints to the outputs of the predictive language models to ensure conformity, and for g) implementing privacy controls by generating an updated user input string through filtering personally identifiable information, sensitive information, intellectual property, or prompt injection vulnerabilities from user input, and for i) regulating a desired factuality, factual consistency, toxicity, and consistency in output generated by the predictive language models.
However, Lin discloses wherein processing the text output string to generate a sanitized text output string includes:
a) applying type and structure constraints to outputs of the one or more predictive language models to ensure conformity; b) identifying one or more instances of lack of factuality, toxicity, prohibited type or prohibited structure in the text prompt string; f) processing type and structure constraints to the outputs of the one or more predictive language models to ensure conformity; g) implementing privacy controls by generating an updated user input string through filtering personally identifiable information, sensitive information, intellectual property, or prompt injection vulnerabilities from user input, i) regulating a desired factuality, factual consistency, toxicity, and consistency in output generated by the one or more predictive language models. (see Lin paragraph [0052]: The Mixed-Radix Input Generator 240 transforms each input part 442 to the respective input digit 444 by adjusting (e.g., reducing) one or more of the input parts 442 of the input data 242 to adjusted values such that there are no gaps, e.g., no unused values corresponding to values prohibited by the schema, in the range of possible values between 0 and the number of possible values of the input part 442. Each input digit 444 can be determined by subtracting, from the respective data part, the number of invalid values that respective the input data part 442 cannot be equal to (as specified by the data schema 202) that are less than the input data part. Thus, if a schema part does not specify any prohibited values, then no adjustment of the respective input part 442 is performed, and the respective input digit 444 has the same value as the respective input part 442. Further, if a schema part specifies one or more prohibited values (e.g., a “not equal to 0” constraint) that are less than the respective input part 442, then the respective input digit 444 is set to the value of the respective input part 442 minus the number of prohibited values that are less than the respective input part 442.)
It would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to modify Dong for a) applying type and structure constraints to outputs of the predictive language models to ensure conformity, and for b) identifying instances of lack of factuality, toxicity, prohibited type or prohibited structure in the text prompt string, and for f) processing type and structure constraints to the outputs of the predictive language models to ensure conformity, and for g) implementing privacy controls by generating an updated user input string through filtering personally identifiable information, sensitive information, intellectual property, or prompt injection vulnerabilities from user input, and for i) regulating a desired factuality, factual consistency, toxicity, and consistency in output generated by the predictive language models as taught by Lin. One of ordinary skill in the art would have been motivated to employ the teachings of Lin for the benefits achieved from the flexibility of a system that enables multiple techniques such as implementing type and structure constraints during data processing. (see Lin paragraph [0052])
Dong does not specifically disclose for c) aligning predictive model outputs with knowledge bases as dictated by a user, and for e) decoding partial outputs from the one or more predictive language models, and for h) exercising arbitrary control over outputs generated by a second of one or more predictive language models, leveraging inputs from a first set of one or more predictive language models.
However, Dunagan discloses:
c) aligning predictive model outputs with knowledge bases as dictated by a user; (see Dunagan paragraph [0066]: Each CEDR operator 800 in an applications and data space can include the consistency monitor 822 and the operational module 824. The consistency monitor 822 decides whether to block the input stream in an alignment buffer 826 until output may be produced which upholds the desired level of consistency. The operational module 824 can compute the output stream based on incoming tuples and current operator state 828.; (output consistency); paragraph [0085]: Information may be collected for use in answering subsequent queries. For example, a reservoir of historical information may be maintained for levels of a stack. An application developer can be aware of such a reservoir and chose to collect information that is consistent across reservoirs.; paragraph [0086]: The knowledge base 1050 may be maintained in a query/strategy space or in an applications/data space. A knowledge base 1050 may be maintained in a distributed manner, for example, across a plurality of reservoirs.)
It would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to modify Dong for c) aligning predictive model outputs with knowledge bases as dictated by a user, and for e) decoding partial outputs from the one or more predictive language models, and for h) exercising arbitrary control over outputs generated by a second of one or more predictive language models, leveraging inputs from a first set of one or more predictive language models. as taught by Dunagan. One of ordinary skill in the art would have been motivated to employ the teachings of Dunagan for the benefits achieved from the flexibility of a system to align outputs enabling consistent outputs with a knowledge base information. (see Dunagan paragraph [0066])
Dong does not specifically disclose for d) enforcing quality controls on output from the one or more predictive language models.
However, Luft discloses:
d) enforcing quality controls on output from the one or more predictive language models. (see Luft paragraph [0025]: network service node 305 is an application and subscriber aware network element capable of implementing application specific policies on a per subscriber basis at line rates. For example, network service node 305 can perform quality of service ("QoS") tasks (e.g., traffic shaping, flow control, admission control, etc.) on a per subscriber, per application basis, while monitoring quality of experience ("QoE") on a per session basis. To enable QoS and QoE applications for a variety of network services (e.g., VoD, VoIP, IPTV, etc.), network service node 305 is capable of deep packet inspection all the way to the session and application layers of the OSI model. To provide this granularity of service to hundreds or thousands of unique subscribers requires leveraging parallel processing advantages of a distributed compute environment.)
It would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to modify Dong for d) enforcing quality controls on output from the one or more predictive language models as taught by Luft. One of ordinary skill in the art would have been motivated to employ the teachings of Luft for the benefits achieved from the flexibility of a system implementing multiple techniques such as enforcing quality controls during data processing. (see Luft paragraph [0025])
7. Claim 4 is rejected under 35 U.S.C. 103 as being unpatentable over Dong in view of Flesher and further in view of Lin et al. (US PGPUB No. 20230379138).
Regarding Claim 4, Dong discloses the computer-implemented method of claim 1, including generating sanitized prompt string. (see Dong paragraph [0016]: pipeline 100 includes sensitive token identification module 110, context data structure generation module 120, context-based token classification module 130, and post-processing module 140. Generally speaking, pipeline 100 is operable to receive first text data 105 and output sanitized second text data 145, which includes context-aware labels in place of sensitive text.)
Dong does not specifically disclose receiving user input including text prompt string includes determining control or constraint parameters to be paired with the text prompt string when processing text prompt string to generate the sanitized text prompt string using the one or more predictive language models.
However, Lin discloses wherein receiving the user input including the text prompt string includes determining control or constraint parameters to be paired with the text prompt string when processing the text prompt string to generate the sanitized text prompt string using the one or more predictive language models. (see Lin paragraph [0052]: The Mixed-Radix Input Generator 240 transforms each input part 442 to the respective input digit 444 by adjusting (e.g., reducing) one or more of the input parts 442 of the input data 242 to adjusted values such that there are no gaps, e.g., no unused values corresponding to values prohibited by the schema, in the range of possible values between 0 and the number of possible values of the input part 442. Each input digit 444 can be determined by subtracting, from the respective data part, the number of invalid values that respective the input data part 442 cannot be equal to (as specified by the data schema 202) that are less than the input data part. Thus, if a schema part does not specify any prohibited values, then no adjustment of the respective input part 442 is performed, and the respective input digit 444 has the same value as the respective input part 442. Further, if a schema part specifies one or more prohibited values (e.g., a “not equal to 0” constraint) that are less than the respective input part 442, then the respective input digit 444 is set to the value of the respective input part 442 minus the number of prohibited values that are less than the respective input part 442.)
It would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to modify Dong for receiving user input including text prompt string includes determining control or constraint parameters to be paired with the text prompt string when processing text prompt string to generate the sanitized text prompt string using the one or more predictive language models as taught by Lin. One of ordinary skill in the art would have been motivated to employ the teachings of Lin for the benefits achieved from the flexibility of a system that enables multiple techniques such as implementing type and structure constraints during data processing. (see Lin paragraph [0052])
8. Claim 5 is rejected under 35 U.S.C. 103 as being unpatentable over Dong in view of Flesher and further in view of Nixon et al. (Patent No. EP 2973242 B1).
Regarding Claim 5, Dong discloses the computer-implemented method of claim 1, wherein one or more predictive models are pre-configured and trained with one or more of predefined task templates, special tokens to optimize a performance of the one or more predictive language models with task-specific templates; the computer-implemented method further comprising:
a) training, via the one or more processors, the one or more predictive language models to extract task-specific information from user input text prompts and subsequently adjusting configurations of the one or more predictive language models; b) incorporating, via the one or more processors, task-specific data and feedback during training to create tailored templates for task-specific purposes; c) generating, via the one or more processors, task-specific templates by customizing the configurations of the one or more predictive language models with pre-configured prompts tailored to specific tasks with role-based and instruction indicators, variable placeholders, and structuring elements. (see Dong paragraph [0100]: different entities (which may variously be referred to as “units,” “circuits,” other components, etc.) may be described or claimed as “configured” to perform one or more tasks or operations. This formulation-[entity] configured to [perform one or more tasks]—is used herein to refer to structure (i.e., something physical). More specifically, this formulation is used to indicate that this structure is arranged to perform the one or more tasks during operation. A structure can be said to be “configured to” perform some task even if the structure is not currently being operated. Thus, an entity described or recited as being “configured to” perform some task refers to something physical, such as a device, circuit, a system having a processor unit and a memory storing program instructions executable to implement the task; paragraph [0101]: various units/circuits/components may be described herein as performing a set of task or operations. It is understood that those entities are “configured to” perform those tasks/operations,)
Dong does not specifically disclose iteratively reconfiguring the configurations of the predictive language models and performance based on user input, control parameters in various tasks including summarization, rephrasing, question answering, sentiment analysis, factual consistency detection, toxicity detection, chat, or machine translation.
However, Nixon discloses:
d) iteratively, via the one or more processors, reconfiguring the configurations of the one or more predictive language models and performance based on user input, control parameters in various tasks including summarization, rephrasing, question answering, sentiment analysis, factual consistency detection, toxicity detection, chat, or machine translation. (see Nixon paragraph [0022]: The data modeling studio as described herein provides a robust and efficient architecture for graphically creating and executing data models in a process plant environment, and is particularly useful in analyzing data stored in a big data machine. When the data modeling studio is implemented in a process plant utilizing a big data architecture that includes a big data appliance that collects and stores all (or almost all) process data and plant data collected, the data modeling studio enables the efficient creation, testing, and operation of models using that data. For example, the user may use the data modeling studio to graphically and iteratively create and edit data models without having to reconfigure the plant operation to collect additional data, which makes for a must faster model development environment. Moreover, in some cases, because all of the plant data is available in a big data machine, a model may be created to iterate or change itself in various manners to add new, more or different data within a modeling routine or to analyze new or different data as part of the modeling routine, testing itself and iteratively changing itself to develop a better or more accurate model that provides better or more accurate predictions, trend analysis, fault or abnormal situation detection, etc. In still another case, a user may create one or more data models which can be used in a knowledge discovery environment to analyze data within a big data machine to discover relationships, trends, etc. about that data and thus about the plant.)
It would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to modify Dong for iteratively reconfiguring the configurations of the predictive language models and performance based on user input, control parameters in various tasks including summarization, rephrasing, question answering, sentiment analysis, factual consistency detection, toxicity detection, chat, or machine translation as taught by Nixon. One of ordinary skill in the art would have been motivated to employ the teachings of Nixon for the benefits achieved from the flexibility of a system that enables repeated reconfiguration until an optimized configuration is achieved. (see Nixon paragraph [0022])
9. Claim 6 is rejected under 35 U.S.C. 103 as being unpatentable over Dong in view of Flesher and further in view of Luft et al. (US PGPUB No. 20090034426) and Schillace et al. (US PGPUB No. 20240202582).
Regarding Claim 6, Dong discloses the computer-implemented method of claim 1.
Dong does not specifically disclose for a) parallel processing for controlling a flow of user input through data processing steps based on detections performed to optimize performance, and for b) implementing user-configurable arbitrary or rule-based quality controls.
However, Luft discloses further comprising:
a) parallel processing, via the one or more processors, for controlling a flow of user input through data processing steps based on detections performed to optimize performance; b) implementing, via the one or more processors, user-configurable arbitrary or rule-based quality controls, via an API application. (see Luft paragraph [0025]: network service node 305 is an application and subscriber aware network element capable of implementing application specific policies on a per subscriber basis at line rates. For example, network service node 305 can perform quality of service ("QoS") tasks (e.g., traffic shaping, flow control, admission control, etc.) on a per subscriber, per application basis, while monitoring quality of experience ("QoE") on a per session basis. To enable QoS and QoE applications for a variety of network services (e.g., VoD, VoIP, IPTV, etc.), network service node 305 is capable of deep packet inspection all the way to the session and application layers of the OSI model. To provide this granularity of service to hundreds or thousands of unique subscribers requires leveraging parallel processing advantages of a distributed compute environment.)
It would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to modify Dong for a) parallel processing for controlling a flow of user input through data processing steps based on detections performed to optimize performance, and for b) implementing user-configurable arbitrary or rule-based quality controls as taught by Luft. One of ordinary skill in the art would have been motivated to employ the teachings of Luft for the benefits achieved from the flexibility of a system implementing multiple techniques such as enforcing quality controls during data processing. (see Luft paragraph [0025])
Dong does not specifically disclose for c) facilitating concurrent chaining of multiple generative AI or predictive model inferences, enabling an execution of tasks including text classifications, Boolean outputs, rephrasing, summarization, and various quality assessments on outputs of the one or more predictive language models, and for d) constructing responses for a client application from the API application after applying arbitrary controls in parallel during chaining of multiple generative AI or predictive model inferences, including encompassing error messages, flags, scores, or other relevant information.
However, Schillace discloses:
c) facilitating, via the one or more processors, concurrent chaining of multiple generative AI or predictive model inferences, enabling an execution of tasks including text classifications, Boolean outputs, rephrasing, summarization, and various quality assessments on outputs of the one or more predictive language models; and d) constructing, via the one or more processors, responses for a client application from the API application after applying arbitrary controls in parallel during chaining of multiple generative AI or predictive model inferences, including encompassing error messages, flags, scores, or other relevant information. (see Schillace paragraph [0054]: As a result of the multi-stage ML model chaining performed by ML models 204 and 206, model output 208 is generated. Thus, one or more model skills of a skill chain may each generate intermediate output (e.g., structured output), while a final skill of the skill chain (e.g., an ML model evaluation by ML model 206, as illustrated) may generate model output 208 based on such intermediate output. As an example, final model output produced by ML model 206 includes, but is not limited to, natural language output, speech and/or audio output, image output, video output, and/or programmatic output.; paragraph [0023]: the intermediate output comprises structured output, which may include one or more tags, key/value pairs, and/or metadata, among other examples. For example, a stream may be denoted according to an associated tag within such structured output. In examples, a prompt template of a skill defines or otherwise includes an indication relating to such structured output, thereby causing the generative ML model associated with the skill to produce structured output accordingly. As such, use of structured output may increase the degree to which model output is deterministic and may therefore improve reliability when chaining multiple ML model evaluations together according to aspects described herein. In other examples, intermediate output may be similar to ultimate model output that may otherwise be provided to a user or for further processing by an application, among other examples.)
It would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to modify Dong for c) facilitating concurrent chaining of multiple generative AI or predictive model inferences, enabling an execution of tasks including text classifications, Boolean outputs, rephrasing, summarization, and various quality assessments on outputs of the one or more predictive language models, and for d) constructing responses for a client application from the API application after applying arbitrary controls in parallel during chaining of multiple generative AI or predictive model inferences, including encompassing error messages, flags, scores, or other relevant information as taught by Schillace. One of ordinary skill in the art would have been motivated to employ the teachings of Schillace for the benefits achieved from the flexibility of a system that enables the chaining of multiple outputs from models in the generation of responses from learning models. (see Schillace paragraph [0054]; paragraph [0023])
10. Claims 7, 14 are rejected under 35 U.S.C. 103 as being unpatentable over Dong in view of Flesher and further in view of Manolache et al. (US PGPUB No. 20220327108).
Regarding Claims 7, 14, Dong discloses the computer-implemented. method of claim 1, further comprising:
a) modifying, via the one or more processors, data by removing, obfuscating, encrypting, substituting, or anonymizing specific information before exposing it to predictive models; (see Dong paragraph [0016]: operable to receive first text data 105 and output sanitized second text data 145, which includes context-aware labels in place of sensitive text.)
Dong does not specifically disclose for b) implementing a controlled decoder that initiates model generation with a prefix that systematically concatenates or enumerates all tokens from a vocabulary to the prefix, and for c) masking disallowed tokens with probabilities identifying the disallowed tokens to identify allowable tokens capable of matching a specified pattern, and for d) wherein finite state machines, or other suitable implementations, govern controlled decoding and state transitions that regulate a token generation process.
However, Manolache discloses:
b) implementing, via the one or more processors, a controlled decoder that initiates model generation with a prefix that is initially empty and systematically concatenates or enumerates all tokens from a vocabulary to the prefix; (see Manolache paragraph [0036]: input modifier 40 further comprises a token generator 41 configured to output a set of substitute tokens to replace the masked tokens within sequence 32. In the illustrated example, token generator outputs substitute tokens 35f and 35g to replace tokens 35c and 35e, respectively. A simple embodiment of generator 41 may be configured to randomly draw the substitute token from a reference pool. In more advanced embodiments, token generator 41 may comprise a dictionary/thesaurus and be configured, for each masked token, to output a synonym or an antonym of the respective token.) and
c) masking, via the one or more processors, disallowed tokens with probabilities set at 0.0 by identifying the disallowed tokens with regex or context-free patterns to identify allowable tokens capable of matching a specified pattern; (see Manolache paragraph [0036]: a token generator 41 configured to output a set of substitute tokens to replace the masked tokens within sequence 32; employ statistical language models to generate a substitute token according to a probability of occurrence of the respective substitute token within the context of the masked token.; paragraph [0037]: token generator 41 producing plausible substitute tokens comprises an AI system (e.g., set of deep neural networks) trained on a corpus of token sequences representative for the respective anomaly detection application. Such a version of generator 41 may output a substitute token according to a subsequence of tokens preceding a masked token within sequence 32.) and
d) wherein finite state machines, including deterministic finite automaton or other suitable implementations, govern controlled decoding and state transitions that regulate a token generation process. (see Manolache paragraph [0034]: an exemplary transformation 30 may remove tokens of type ‘create process’ from training sequence 32. An equivalent of paraphrasing in such embodiments may comprise replacing a target sequence of events with a substitute sequence of events that would bring the respective computer system to the same final state.)
It would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to modify Dong for b) implementing a controlled decoder that initiates model generation with a prefix that systematically concatenates or enumerates all tokens from a vocabulary to the prefix, and for c) masking disallowed tokens with probabilities identifying the disallowed tokens to identify allowable tokens capable of matching a specified pattern, and for d) wherein finite state machines, or other suitable implementations, govern controlled decoding and state transitions that regulate a token generation process as taught by Manolache. One of ordinary skill in the art would have been motivated to employ the teachings of Manolache for the benefits achieved from the flexibility of a system that enables the management of disallowed information in the processing of token information. (see Manolache paragraph [0036])
11. Claim 8, 16 is rejected under 35 U.S.C. 103 as being unpatentable over Dong in view of Flesher and further in view of Feng et al. (Patent No. WO 2022/266556 A1).
Regarding Claims 8, 16, Dong discloses the computer-implemented method of claim 1 and the computing system of claim 9.
Dong does not specifically disclose optimizing the one or more predictive language models for a specific hardware type of the one or more processors.
However, Feng discloses wherein further comprising: optimizing the one or more predictive language models for a specific hardware type of the one or more processors. (see Feng paragraph [0036]: processor 150 includes different types of processing units, such as central processing unit (CPU) 151 and neural processing unit (NPU) 152. Processor 150 may additionally include a graphic processing unit (GPU) for efficient image processing. Different types of processing units are optimized for different types of computations. For example, CPU 151 handles various types of system functions, such as managing cameras 180 and HMD tracking sensor(s) 175 and moving raw sensor data to memory 155. NPU 152 is optimized for convolutional neural networks and predictive models.)
It would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to modify Dong for optimizing the one or more predictive language models for a specific hardware type of the one or more processors as taught by Feng. One of ordinary skill in the art would have been motivated to employ the teachings of Fend for the benefits achieved from the flexibility of a system enabling implementation of optimization for data processing system. (see Feng paragraph [0036])
12. Claim 13 is rejected under 35 U.S.C. 103 as being unpatentable over Dong in view of Flesher and further in view of Schillace et al. (US PGPUB No. 20240202582).
Regarding Claim 13, Dong discloses the computing system of claim 9, the memories having stored thereon instructions that when executed cause the computing system to perform operations. (see Dong paragraph [0080]: Program instructions may be stored on a “non-transitory, computer-readable storage medium” or a “non-transitory, computer-readable medium.” The storage of program instructions on such media permits execution of the program instructions by a computer system.; paragraph [0100]: A structure can be said to be “configured to” perform some task even if the structure is not currently being operated. Thus, an entity described or recited as being “configured to” perform some task refers to something physical, such as a device, circuit, a system having a processor unit and a memory storing program instructions executable to implement the task,)
Dong does not specifically disclose facilitate concurrent chaining of multiple generative AI or model inferences, constructing responses while applying controls in parallel during chaining of multiple model inferences.
However, Schillace discloses wherein facilitate concurrent chaining of multiple generative AI or model inferences, constructing responses for a client application from an API application while applying controls in parallel during chaining of multiple model inferences. (see Schillace paragraph [0054]: As a result of the multi-stage ML model chaining performed by ML models 204 and 206, model output 208 is generated. Thus, one or more model skills of a skill chain may each generate intermediate output (e.g., structured output), while a final skill of the skill chain (e.g., an ML model evaluation by ML model 206, as illustrated) may generate model output 208 based on such intermediate output. As an example, final model output produced by ML model 206 includes, but is not limited to, natural language output, speech and/or audio output, image output, video output, and/or programmatic output.; paragraph [0023]: the intermediate output comprises structured output, which may include one or more tags, key/value pairs, and/or metadata, among other examples. For example, a stream may be denoted according to an associated tag within such structured output. In examples, a prompt template of a skill defines or otherwise includes an indication relating to such structured output, thereby causing the generative ML model associated with the skill to produce structured output accordingly. As such, use of structured output may increase the degree to which model output is deterministic and may therefore improve reliability when chaining multiple ML model evaluations together according to aspects described herein. In other examples, intermediate output may be similar to ultimate model output that may otherwise be provided to a user or for further processing by an application, among other examples.)
It would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to modify Dong for facilitate concurrent chaining of multiple generative AI or model inferences, constructing responses while applying controls in parallel during chaining of multiple model inferences as taught by Schillace. One of ordinary skill in the art would have been motivated to employ the teachings of Schillace for the benefits achieved from the flexibility of a system that enables the chaining of multiple outputs in the generation of responses from learning models. (see Schillace paragraph [0054]; paragraph [0023])
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to CARLTON JOHNSON whose telephone number is (571)270-1032. The examiner can normally be reached Work: 12-9PM (most days).
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, Shewaye Gelagay can be reached at 571-272-4219. 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.
/CJ/
June 1, 2026
/SHEWAYE GELAGAY/Supervisory Patent Examiner, Art Unit 2436
Prosecution Insights