22
DETAILED ACTION
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 .
Information Disclosure Statement
The information disclosure statement (IDS) submitted on 1/2/2026 is in compliance with the provisions of 37 CFR 1.97. Accordingly, the information disclosure statement is being considered by the examiner.
Claim Rejections - 35 USC § 101
35 U.S.C. 101 reads as follows:
Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title.
Claims 1-20 stand rejected:
Claims 1 (“method”), 14 (“Computer program product”), and 20 (“system”) are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more. The claims are basically about how to “compress[]” a “tool prompt” (Sp. ¶ 0042 S2: “tool prompt” “provides” “descriptions of a list of tools available” where a “tool” is Sp. ¶ 0017 last S: “e.g., calculators, application programming interfaces (APIs), search engines and the like”). The method begins by “segmenting” “the” “tool prompt” “into multiple text chunks” (e.g., by “identif[ing]” “tags” “keywords” “phrases” “structural elements specific to functional tool descriptions in tool prompts” (dep claim 11)).
Next a “semantic vector representation” of the resulting “text chunks” is “generat[ed]” (e.g., a text to vector representation). Using the “semantic vector representation”, a “first semantic distribution” is obtained (Sp. ¶ 0031 lines 2+: “generating the first semantic distribution comprises processing the at least one semantic vector representation via a Gaussian Mixture Model (GMM)”).
Next “a perturbed semantic vector representation” is obtained by “eliminating at least one text chunk from the multiple text chunks” (e.g. eliminating certain words (e.g. propositions) not determined important to the overall meaning of the “tool prompt” (tool description)). A “second semantic distribution” (also by using “Gaussian Mixture Model” (“GMM”)) is “generat[ed]” “based on the perturbed semantic vector representation”.
A “similarity metric” (e.g., Sp. ¶ 0028 S1; “using” “a first algorithm” which according to Sp. Par. 0029 S1: “Examples” of “first algorithm is a K-S test algorithm” (i.e., “Kolmogorov-Smirnov text (K-S text)”) calculation is performed between the “first” and the “second” “semantic distribution[s]” and if that “similarity” “exceed[s] a threshold similarity value”, a “compressed tool prompt” based on the textual representation of the “perturbed vector representation” (or “based on the subset of the multiple text chunks”) is “generat[ed]”.
Here other than reciting “generative machine learning model” (Claims 1, 14, and 20), and “program instructions stored on the computer readable storage medium to perform” claims steps (Claim 14), and a “processor set” (claim 20), nothing in the claim limitations precludes their limitations from practically being performed in the mind. For example, a person doing a text summarization of a “tool” (device) description (“tool prompt”); e.g., the could simply take an original description pertaining to the tool and begin eliminating words not determined very important to the overall meaning (semantic representation) of the “tool” description one by one and use his own technical knowledge pertaining to the “tool” as well as knowledge of grammar and vocabulary to assess if the description after the elimination imparts roughly the same meaning or not and if the answer is yes to replace the original “tool” description by the modified one and iteratively continue in compressing the “tool” description. The generation of “semantic vector representation” of the original “tool” description (“tool prompt”) requires nothing more than a lookup text to vector table. If a claim limitation, under its broadest reasonable interpretation, covers performance of the limitation in the mind but for the recitation of generic computer components and/or computer software, then it falls withing the “Mental Processes” grouping of abstract ideas. Accordingly, the claims recite an abstract idea.
This judicial exception is not integrated into a practical application. In particular, the claims recited additional elements (i.e., “a generative machine learning model” (claims 1, 14, 20), “processor set” in claim 20) to perform all the claim limitations and steps. But they are recited at a high-level of generality (i.e., to perform the “receiving” “segmenting” “generating” “performing” and “compress[ing]”) such that it amounts no more than mere instructions to apply the exception using a generic computer component and/or software. Accordingly, these additional elements do not integrate the abstract idea into a practical application because they do not impose any meaningful limits on practicing the abstract idea. The claims are thus directed to an abstract idea.
The claims do not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to integration of the abstract idea into a practical application, the additional element of using e.g., a “processor set” and/or “a generative machine learning model” to perform all the steps of “receiving” “segmenting” “generating” “performing” and “compress[ing]”, amounts to no more than mere instructions to apply the exception using a generic computer component. Mere instructions to apply an exception using a generic computer component cannot provide an inventive concept. The claims are not therefore patent eligible.
Regarding claims 2 (15), the person who generates the text summarization of the tool, could simply store it on a sheet of paper for later use.
Regarding claims 3 (16), the person who generates the text summarization, if the text is about e.g. how to perform a task, he could use that to engage in the task and generate an output.
Regarding claims 4 (17), the summarization could provide description of how to perform the task which could be something pertaining to a specific tool usage.
Regarding claims 5(18), the summary would certainly contain less words (2nd number of defining elements) than the original description word count (1st number of defining elements).
Regarding claims 6(19) and (claims 7-8) generating of “similarity metric” is directed to a mathematical operation abstract type.
Regarding claim 9 (and 10), determination of “largest difference …” is directed to a mathematical operation abstract type.
Regarding claim 11, the person doing summarization would be knowlegable of specific keywords associated with the prompt to help in parsing it.
Regarding claims 12 using Gaussian Mixture Model would amount to using a mathematical operation recipe.
Regarding claim 13 determining “semantic vectors” can be accomplished by a lookup table.
Claim 14 is rejected under 35 USC 101 because the claimed invention is directed to a nonstatutory subject matter. First, the “computer program product” recitation fails to state that the “product” is a medium that is useable or readable by a computer an hence encompasses a software printout, book, or other printed material that is not directly accessible by a computer. In such a case, utility patentability rests soley with the details of the “product”, which are not claimed, and thus would be anticipated by any book, software printout or other printed matter. Second, the “program comprising” “program instructions” is software per se, and thus is merely functional descriptive material unless claimed as embodied on a computer readable non-transitory medium, or other tangible and functional computer program product claimed as specifically enabling a computer or processor to process the claimed “code” and thus effect utility.
Claim Rejections - 35 USC § 102
The following is a quotation of the appropriate paragraphs of 35 U.S.C. 102 that form the basis for the rejections under this section made in this Office action:
A person shall be entitled to a patent unless –
(a)(2) the claimed invention was described in a patent issued under section 151, or in an application for patent published or deemed published under section 122(b), in which the patent or application, as the case may be, names another inventor and was effectively filed before the effective filing date of the claimed invention.
Claim(s) 1, 9, 11, 14, 20 is/are rejected under 35 U.S.C. 102(a)(2) as being anticipated by VENKITACHALAM et al. (US 2025/0342183).
Regarding claim 1, VENKITACHALAM et al. do teach a computer-implemented method (¶ 0002 S1: “Various aspects of the technology described herein are generally directed to systems, methods, and computer storage media for, among other things, using Shapley values to evaluate prompt generation parameters”)
comprising:
receiving an original tool prompt for a generative machine learning model; segmenting the original tool prompt into multiple text chunks (¶ 0032 S1-2: “A” “prompt compression tool” “refers to a prompt refinement tool that utilizes a model to paraphrase a prompt to more concise lengths without changing the meaning of the prompt” “e.g., by removing unnecessary words or letters” (a “prompt” (a received tool prompt) is segmented into “words or letters” (multiple text chunks) for processing by a “Sentence-Bidirectional Encoder representation from Transformers (SBERT)” (a generative machine learning model));
generating at least one semantic vector representation of the multiple text chunks; generating a first semantic distribution based on the at least one semantic vector representation (¶ 0032 S2: “For example, a prompt compression tool can be a model that is trained for prompt optimization through text compression using the measured quality” “e.g., cosine similarity between the Sentence-Bidirectional Encoder Representations from Transformers (SBERT) generated embeddings” “of the reduced prompt and original prompt to reduce token count, but maintain the quality of the prompt” (i.e., obtained “generated embeddings” (a semantic vector representation or first semantic distribution) of the “original prompt” (of the original tool prompt or all of its “words and letters” (the text chunks));
generating a perturbed semantic vector representation based on a subset of the multiple text chunks, the subset being generated by eliminating at least one text chunk from the multiple text chunks; generating a second semantic distribution based on the perturbed semantic vector representation (¶ 0032 S2: “For example, a prompt compression tool can be a model that is trained for prompt optimization through text compression using the measured quality” “e.g., cosine similarity between the Sentence-Bidirectional Encoder Representations from Transformers (SBERT) generated embeddings” “of the reduced prompt and original prompt to reduce token count, but maintain the quality of the prompt” (i.e., the obtained “generated embeddings” (a perturbed or second semantic vector representation associated with a second semantic distribution) corresponding to “the reduced prompt” (i.e., one obtained by “reduc[ing]” (i.e., eliminating) some of the “words and letters” (text chunks) associated with the “original prompt” (the plurality of text chunks) resulting in a subset of the text chunks));
performing a comparison of the first semantic distribution and the second semantic distribution to generate at least one similarity metric; and in response to the at least one similarity metric exceeding a threshold similarity value, generating a compressed tool prompt based on the subset of the multiple text chunks (¶ 0032 S2: “For example, a prompt compression tool can be a model that is trained for prompt optimization through text compression using the measured quality” “e.g., cosine similarity between the Sentence-Bidirectional Encoder Representations from Transformers (SBERT) generated embeddings” “of the reduced prompt and original prompt to reduce token count, but maintain the quality of the prompt”(a “cosine similarity” (a similarity metric) is used to compare “embeddings” “of the reduced prompt” (the second semantic distribution) and “embeddings” “of the” “original prompt” (the first semantic distribution) which “maintain[s] quality of prompt” (i.e., when the metric exceeds a threshold associated with “quality” will thus help designate “the reduced prompt” (a compressed tool prompt is validated)).
Regarding claim 9, VENKITACHALAM et al. do teach the method of claim 1, wherein the at least one similarity metric is generated by executing at least one of a first algorithm that measures a largest difference between the first semantic distribution and the second semantic distribution, and a second algorithm that measures how much the first semantic distribution and the second semantic distribution agree or differ (¶ 0032 S2: “For example, a prompt compression tool can be a model” (a second algorithm) “that is trained for prompt optimization through text compression using the measured quality” “e.g., cosine similarity” (for the similarity metric generated) “between the Sentence-Bidirectional Encoder Representation from Transformers (SBERT) generated embeddings” “of the reduced prompt and original prompt” (based how how much the first and the second semantic distributions agree or differ) “to reduce token count, but maintain quality of prompt”).
Regarding claim 11, VENKITACHALAM et al. do teach the method of claim 1, wherein segmenting the original tool prompt into multiple text chunks comprises parsing the original tool prompt and generating text chunks based on an identification of at least one of tags, key words, phrases, or structural elements specific to functional tool descriptions in tool prompts ((¶ 0032 S1-2: “A” “prompt compression tool” “refers to a prompt refinement tool that utilizes a model to paraphrase a prompt to more concise lengths without changing the meaning of the prompt” “e.g., by removing unnecessary words or letters” (the text chunks correspond to) “rephrasing synonyms” (e.g. functional tool descriptions associated with the “prompts” (tool prompts)).
Regarding claim 14, VENKITACHALAM et al. do teach a computer program product comprising: a computer readable storage medium; and program instructions stored on the computer storage medium to perform operations (¶ 0002 S1: “Various aspects of the technology described herein are generally directed to systems, methods, and computer storage media for, among other things, using Shapley values to evaluate prompt generation parameters”)
comprising:
receiving an original tool prompt for a generative machine learning model; segmenting the original tool prompt into multiple text chunks (¶ 0032 S1-2: “A” “prompt compression tool” “refers to a prompt refinement tool that utilizes a model to paraphrase a prompt to more concise lengths without changing the meaning of the prompt” “e.g., by removing unnecessary words or letters” (a “prompt” (a received tool prompt) is segmented into “words or letters” (multiple text chunks) for processing by a “Sentence-Bidirectional Encoder representation from Transformers (SBERT)” (a generative machine learning model));
generating at least one semantic vector representation of the multiple text chunks; generating a first semantic distribution based on the at least one semantic vector representation (¶ 0032 S2: “For example, a prompt compression tool can be a model that is trained for prompt optimization through text compression using the measured quality” “e.g., cosine similarity between the Sentence-Bidirectional Encoder Representations from Transformers (SBERT) generated embeddings” “of the reduced prompt and original prompt to reduce token count, but maintain the quality of the prompt” (i.e., obtained “generated embeddings” (a semantic vector representation or first semantic distribution) of the “original prompt” (of the original tool prompt or all of its “words and letters” (the text chunks));
generating a perturbed semantic vector representation based on a subset of the multiple text chunks, the subset being generated by eliminating at least one text chunk from the multiple text chunks; generating a second semantic distribution based on the perturbed semantic vector representation (¶ 0032 S2: “For example, a prompt compression tool can be a model that is trained for prompt optimization through text compression using the measured quality” “e.g., cosine similarity between the Sentence-Bidirectional Encoder Representations from Transformers (SBERT) generated embeddings” “of the reduced prompt and original prompt to reduce token count, but maintain the quality of the prompt” (i.e., the obtained “generated embeddings” (a perturbed or second semantic vector representation associated with a second semantic distribution) corresponding to “the reduced prompt” (i.e., one obtained by “reduc[ing]” (i.e., eliminating) some of the “words and letters” (text chunks) associated with the “original prompt” (the plurality of text chunks) resulting in a subset of the text chunks));
performing a comparison of the first semantic distribution and the second semantic distribution to generate at least one similarity metric; and generating in response to the at least one similarity metric exceeding a threshold similarity value, generating a compressed tool prompt based on the subset of the multiple text chunks (¶ 0032 S2: “For example, a prompt compression tool can be a model that is trained for prompt optimization through text compression using the measured quality” “e.g., cosine similarity between the Sentence-Bidirectional Encoder Representations from Transformers (SBERT) generated embeddings” “of the reduced prompt and original prompt to reduce token count, but maintain the quality of the prompt”(a “cosine similarity” (a similarity metric) is used to compare “embeddings” “of the reduced prompt” (the second semantic distribution) and “embeddings” “of the” “original prompt” (the first semantic distribution) which “maintain[s] quality of prompt” (i.e., when the metric exceeds a threshold associated with “quality” will thus help designate “the reduced prompt” (a compressed tool prompt is validated)).
Regarding claim 20, VENKITACHALAM et al. do teach a computer system comprising a processor set; one or more computer-readable storage media; and program instructions stored on the one or more computer-readable storage media to cause the processor set to perform operations (¶ 0002 S1: “Various aspects of the technology described herein are generally directed to systems, methods, and computer storage media for, among other things, using Shapley values to evaluate prompt generation parameters”; [0063] “The user device 102 can include one or more processors, and one or more computer-readable media. The computer-readable media may include computer-readable instructions executable by the one or more processors. The instructions may be embodied by one or more applications, such as application 110 shown in FIG. 1. Application 110 is referred to as single applications for simplicity, but its functionality can be embodied by one or more applications in practice”)
comprising:
receiving an original tool prompt for a generative machine learning model; segmenting the original tool prompt into multiple text chunks (¶ 0032 S1-2: “A” “prompt compression tool” “refers to a prompt refinement tool that utilizes a model to paraphrase a prompt to more concise lengths without changing the meaning of the prompt” “e.g., by removing unnecessary words or letters” (a “prompt” (a received tool prompt) is segmented into “words or letters” (multiple text chunks) for processing by a “Sentence-Bidirectional Encoder representation from Transformers (SBERT)” (a generative machine learning model));
generating at least one semantic vector representation of the multiple text chunks; generating a first semantic distribution based on the at least one semantic vector representation (¶ 0032 S2: “For example, a prompt compression tool can be a model that is trained for prompt optimization through text compression using the measured quality” “e.g., cosine similarity between the Sentence-Bidirectional Encoder Representations from Transformers (SBERT) generated embeddings” “of the reduced prompt and original prompt to reduce token count, but maintain the quality of the prompt” (i.e., obtained “generated embeddings” (a semantic vector representation or first semantic distribution) of the “original prompt” (of the original tool prompt or all of its “words and letters” (the text chunks));
generating a perturbed semantic vector representation based on a subset of the multiple text chunks, the subset being generated by eliminating at least one text chunk from the multiple text chunks; generating a second semantic distribution based on the perturbed semantic vector representation (¶ 0032 S2: “For example, a prompt compression tool can be a model that is trained for prompt optimization through text compression using the measured quality” “e.g., cosine similarity between the Sentence-Bidirectional Encoder Representations from Transformers (SBERT) generated embeddings” “of the reduced prompt and original prompt to reduce token count, but maintain the quality of the prompt” (i.e., the obtained “generated embeddings” (a perturbed or second semantic vector representation associated with a second semantic distribution) corresponding to “the reduced prompt” (i.e., one obtained by “reduc[ing]” (i.e., eliminating) some of the “words and letters” (text chunks) associated with the “original prompt” (the plurality of text chunks) resulting in a subset of the text chunks));
performing a comparison of the first semantic distribution and the second semantic distribution to generate at least one similarity metric; and generating, in response to the at least one similarity metric exceeding a threshold similarity value, generating a compressed tool prompt based on the subset of the multiple text chunks (¶ 0032 S2: “For example, a prompt compression tool can be a model that is trained for prompt optimization through text compression using the measured quality” “e.g., cosine similarity between the Sentence-Bidirectional Encoder Representations from Transformers (SBERT) generated embeddings” “of the reduced prompt and original prompt to reduce token count, but maintain the quality of the prompt”(a “cosine similarity” (a similarity metric) is used to compare “embeddings” “of the reduced prompt” (the second semantic distribution) and “embeddings” “of the” “original prompt” (the first semantic distribution) which “maintain[s] quality of prompt” (i.e., when the metric exceeds a threshold associated with “quality” will thus help designate “the reduced prompt” (a compressed tool prompt is validated)).
Claim Rejections - 35 USC § 103
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.
Claim(s) 2-5, 15-18 is/are rejected under 35 U.S.C. 103 as being unpatentable over VENKITACHALAM et al., and further in view of Fayyaz et al. (US Patent 12,566,765).
Regarding claim 2, VENKITACHALAM et al. do not specifically disclose the method of claim 1, further comprising storing the compressed tool prompt in a data storage that is accessible to an artificial intelligence agent that communicates with the generative machine learning model.
Fayyaz et al. do teach the method of claim 1, further comprising storing the compressed tool prompt in a data storage that is accessible to an artificial intelligence agent that communicates with the generative machine learning model (Col 17 lines 5+ : “acquiring a pre-generating task prompt” (tool prompt) “from a network-accessible data store” (stored in a data storage) “machine-trained model” (accessible to an artificial intelligence model) “includes a base machine-trained model” “that provides an initial task prompt, and a prompt-compressing machine-trained model” (that communicates with a generative machine learning model) “that reduces a size of the initial task prompt” (to produce a compressed tool prompt to be stored) “to produce the task prompt that is sent to the client device”).
It would have therefore been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to incorporate the methods of “task prompt” “prompt-compressing” of Fayyaz et al. into the “prompt compression tool” of VENKITACHALAM et al. would enable the combined systems and their associated methods to perform in combination as they do separately and to further enable VENKITACHALAM et al. to have a “data store” to store its “prompt[s]” and its resulting “compress[ed]” version to avoid having to repeat compressing the same prompt again.
Regarding claim 3, VENKITACHALAM et al. do not specifically disclose the method of claim 1, further comprising: adding the compressed tool prompt to a task prompt; inputting the task prompt into a generative machine learning model; and in response to the inputting, receiving a task output from the generative machine learning model.
Fayyaz et al. do teach:
adding the compressed tool prompt to a task prompt (Col. 1 lines 48+: “the client device produces a combined” (adding) “prompt that includes the task prompt” (compressed tool prompt) “and a query prompt” (and a task prompt) “that represents the query”; i.e., because according to Col. 4 lines 51-52: “The task prompt produced by the prompt-compressing model” (task prompt is compressed));
inputting the task prompt into a generative machine learning model; and in response to the inputting, receiving a task output from the generative machine learning model (Col. 1 lines 50-51: “The client model” (a generative model) “transforms the combined prompt” (that was inputted the task prompt) “into a model response” (produces a task output)).
It would have therefore been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to incorporate the methods of “task prompt” “prompt-compressing” of Fayyaz et al. into the “prompt compression tool” of VENKITACHALAM et al. would enable the combined systems and their associated methods to perform in combination as they do separately and to further enable VENKITACHALAM et al. to tailor its methods to a practical application.
Regarding claim 4, VENKITACHALAM et al. do not specifically disclose the method of claim 1, wherein the original tool prompt helps define a function tool and comprises one or more defining elements selected from a group consisting of:
a function declaration specifying an identifier of the function tool,
a function description that describes what the function tool does,
a parameter description that describes parameters used by the function tool, and
a return description that describes a type of output to be provided by the function tool in response to the function tool being invoked.
Fayyaz et al. do teach
wherein the original tool prompt helps define a function tool and comprises one or more defining elements selected from a group consisting of:
a function declaration specifying an identifier of the function tool,
a function description that describes what the function tool does (Col. 3 lines 37+: “The task description describes a particular task that is to be performed by the client device 104 using the client model 110. In operations (2) and (3), the main-system model 112 transform the task description” (a function description) “into a task prompt” (which describes the “task” (e.g. a tool’s function))),
a parameter description that describes parameters used by the function tool, and
a return description that describes a type of output to be provided by the function tool in response to the function tool being invoked.
For obviousness to combine VENKITACHALAM et al. and Fayyaz et al. see claim 3.
Regarding claim 5, VENKITACHALAM et al. do teach the method of claim 4, wherein the original tool prompt comprises a first number of the defining elements and the compressed tool prompt comprises a second number of the defining elements, the second number being smaller than the first number (¶ 0032 S1-2: “A” “prompt compression tool” “refers to a prompt refinement tool that utilizes a model to paraphrase a prompt to more concise lengths without changing the meaning of the prompt”(the original tool prompt) “e.g., by removing unnecessary words or letters” “in order to reduce the token size” (comprises “token size” also called “token count” (a first number of defining elements (¶ 0032 line 13)) which get “reduced” (converts to a second number of defining elements being smaller))).
Regarding claim 15, VENKITACHALAM et al. do not specifically disclose the computer program product of claim 14, further comprising storing the compressed tool prompt in a data storage that is accessible to an artificial intelligence agent that communicates with the generative machine learning model.
Fayyaz et al. do teach the computer program of claim 14, further comprising storing the compressed tool prompt in a data storage that is accessible to an artificial intelligence agent that communicates with the generative machine learning model (Col 17 lines 5+ : “acquiring a pre-generating task prompt” (tool prompt) “from a network-accessible data store” (stored in a data storage) “machine-trained model” (accessible to an artificial intelligence model) “includes a base machine-trained model” “that provides an initial task prompt, and a prompt-compressing machine-trained model” (that communicates with a generative machine learning model) “that reduces a size of the initial task prompt” (to produce a compressed tool prompt to be stored) “to produce the task prompt that is sent to the client device”).
It would have therefore been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to incorporate the methods of “task prompt” “prompt-compressing” of Fayyaz et al. into the “prompt compression tool” of VENKITACHALAM et al. would enable the combined systems and their associated methods to perform in combination as they do separately and to further enable VENKITACHALAM et al. to have a “data store” to store its “prompt[s]” and its resulting “compress[ed]” version to avoid having to repeat compressing the same prompt again.
Regarding claim 16, VENKITACHALAM et al. do not specifically disclose the computer program product of claim 14, wherein the operations further comprise: adding the compressed tool prompt to a task prompt; inputting the task prompt into a generative machine learning model; and in response to the inputting, receiving a task output from the generative machine learning model.
Fayyaz et al. do teach:
adding the compressed tool prompt to a task prompt (Col. 1 lines 48+: “the client device produces a combined” (adding) “prompt that includes the task prompt” (compressed tool prompt) “and a query prompt” (and a task prompt) “that represents the query”; i.e., because according to Col. 4 lines 51-52: “The task prompt produced by the prompt-compressing model” (task prompt is compressed));
inputting the task prompt into a generative machine learning model; and in response to the inputting, receiving a task output from the generative machine learning model (Col. 1 lines 50-51: “The client model” (a generative model) “transforms the combined prompt” (that was inputted the task prompt) “into a model response” (produces a task output)).
It would have therefore been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to incorporate the methods of “task prompt” “prompt-compressing” of Fayyaz et al. into the “prompt compression tool” of VENKITACHALAM et al. would enable the combined systems and their associated methods to perform in combination as they do separately and to further enable VENKITACHALAM et al. to tailor its methods to a practical application.
Regarding claim 17, VENKITACHALAM et al. do not specifically disclose the computer program product of claim 14, wherein the original tool prompt helps define a function tool and comprises one or more defining elements selected from a group consisting of:
a function declaration specifying an identifier of the function tool,
a function description that describes what the function tool does,
a parameter description that describes parameters used by the function tool, and
a return description that describes a type of output to be provided by the function tool in response to the function tool being invoked.
Fayyaz et al. do teach
wherein the original tool prompt helps define a function tool and comprises one or more defining elements selected from a group consisting of:
a function declaration specifying an identifier of the function tool,
a function description that describes what the function tool does (Col. 3 lines 37+: “The task description describes a particular task that is to be performed by the client device 104 using the client model 110. In operations (2) and (3), the main-system model 112 transform the task description” (a function description) “into a task prompt” (which describes the “task” (e.g. a tool’s function))),
a parameter description that describes parameters used by the function tool, and
a return description that describes a type of output to be provided by the function tool in response to the function tool being invoked.
For obviousness to combine VENKITACHALAM et al. and Fayyaz et al. see claim 3.
Regarding claim 18, VENKITACHALAM et al. do teach the computer program product of claim 17, wherein the original tool prompt comprises a first number of the defining elements and the compressed tool prompt comprises a second number of the defining elements, the second number being smaller than the first number (¶ 0032 S1-2: “A” “prompt compression tool” “refers to a prompt refinement tool that utilizes a model to paraphrase a prompt to more concise lengths without changing the meaning of the prompt”(the original tool prompt) “e.g., by removing unnecessary words or letters” “in order to reduce the token size” (comprises “token size” also called “token count” (a first number of defining elements (¶ 0032 line 13)) which get “reduced” (converts to a second number of defining elements being smaller))).
Claim(s) 6-8, 19 is/are rejected under 35 U.S.C. 103 as being unpatentable over VENKITACHALAM et al., and further in view of RANE et al. (US 2026/0119505).
Regarding claim 6, VENKITACHALAM et al. do not specifically disclose the method of claim 1, further comprising generating an associative tree data structure based on the multiple text chunks and at least one similarity metric, wherein the at least one similarity metric comprises a plurality of similarity metrics, and wherein connections between nodes of the associative tree data structure comprise corresponding similarity metrics, in the plurality of similarity metrics, specifying a similarity between nodes connected by a corresponding connection.
RANE et al. do teach the method of claim 1, further comprising generating an associative tree data structure based on the multiple text chunks and at least one similarity metric, wherein the at least one similarity metric comprises a plurality of similarity metrics, and wherein connections between nodes of the associative tree data structure comprise corresponding similarity metrics, in the plurality of similarity metrics, specifying a similarity between nodes connected by a corresponding connection (¶ 0093+: “one or more artificial intelligence models executes natural language processing on the text prompt” (processing of a tool prompt) “to extract the set of rules” (into multiple chunks)” [0094] 9. “The method embodiment 7 or embodiment 8, wherein an artificial intelligence model of the one or more artificial intelligence models comprises an autoencoder trained to generate a set of embeddings representing each rule of the set of rules; for each pair of embeddings of the set of embeddings: compute a similarity metric” (determine a plurality of similarity metrics) “indicating a similarity of the pair of embeddings. [0095] 10. The method of embodiment 9, further comprising: determining that a similarity metric computed using a first embedding representing a first rule of the set of rules and a second embedding representing a second rule of the set of rules satisfies a similarity condition” (using different rules) ¶ 0098: “generating an expression tree” (an associated tree) “representing the compound rule, wherein the expression tree comprises a plurality of nodes representing the operators and the set of rules” (corresponding to the similarity metrics pertaining to e.g., the “first rule” versus the “second rule”).
It would have therefore been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to incorporate the “tree” implementation for the “text prompt” of RANE et al. into the “prompt compression tool” of VENKITACHALAM et al. would enable the combined systems and their associated methods to perform in combination as they do separately and to further enable to use “the expression tree approach enables the complex expression to be evaluated, analyzed, and transformed systematically” as disclosed in RANE et al. ¶ 0005 S2.
Regarding claim 7, VENKITACHALAM et al. do not specifically disclose the method of claim 6, wherein generating the compressed tool prompt comprises pruning the associative tree data structure by removing nodes and paths which have only connections whose corresponding similarity metrics meet a predetermined criterion, to thereby generate a pruned associative tree data structure.
RANE et al. do teach the method of claim 6, wherein generating the compressed tool prompt comprises pruning the associative tree data structure by removing nodes and paths which have only connections whose corresponding similarity metrics meet a predetermined criterion, to thereby generate a pruned associative tree data structure (¶ 0095 last S: “and responsive to determining that the similarity condition” (similarity condition satisfied) “has been satisfied, removing” (removing nodes) “the second rule from the set of rules”).
For obviousness to combine VENKITACHALAM et al. and RANE et al. see claim 6.
Regarding claim 8, VENKITACHALAM et al. do teach the method of claim 7, wherein generating the compressed tool prompt comprises traversing the pruned associative tree data structure to reconstruct a tool prompt that comprises less textual content than the original tool prompt (¶ 0032 lines 5-6 with respect to “prompt compression” (compressed tool prompt) teach: “to reduce the token size of the prompt” (comprises less textual content)).
Regarding claim 19, VENKITACHALAM et al. do not specifically disclose the computer program product of claim 14, wherein the operation further comprise generating an associative tree data structure based on the multiple text chunks and at least one similarity metric, wherein the at least one similarity metric comprises a plurality of similarity metrics, and wherein connections between nodes of the associative tree data structure comprise corresponding similarity metrics, in the plurality of similarity metrics, specifying a similarity between nodes connected by a corresponding connection.
RANE et al. do teach the computer program product of claim 14, wherein the operation further comprise generating an associative tree data structure based on the multiple text chunks and at least one similarity metric, wherein the at least one similarity metric comprises a plurality of similarity metrics, and wherein connections between nodes of the associative tree data structure comprise corresponding similarity metrics, in the plurality of similarity metrics, specifying a similarity between nodes connected by a corresponding connection (¶ 0093+: “one or more artificial intelligence models executes natural language processing on the text prompt” (processing of a tool prompt) “to extract the set of rules” (into multiple chunks)” [0094] 9. “The method embodiment 7 or embodiment 8, wherein an artificial intelligence model of the one or more artificial intelligence models comprises an autoencoder trained to generate a set of embeddings representing each rule of the set of rules; for each pair of embeddings of the set of embeddings: compute a similarity metric” (determine a plurality of similarity metrics) “indicating a similarity of the pair of embeddings. [0095] 10. The method of embodiment 9, further comprising: determining that a similarity metric computed using a first embedding representing a first rule of the set of rules and a second embedding representing a second rule of the set of rules satisfies a similarity condition” (using different rules) ¶ 0098: “generating an expression tree” (an associated tree) “representing the compound rule, wherein the expression tree comprises a plurality of nodes representing the operators and the set of rules” (corresponding to the similarity metrics pertaining to e.g., the “first rule” versus the “second rule”).
It would have therefore been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to incorporate the “tree” implementation for the “text prompt” of RANE et al. into the “prompt compression tool” of VENKITACHALAM et al. would enable the combined systems and their associated methods to perform in combination as they do separately and to further enable to use “the expression tree approach enables the complex expression to be evaluated, analyzed, and transformed systematically” as disclosed in RANE et al. ¶ 0005 S2.
Claim(s) 10 is/are rejected under 35 U.S.C. 103 as being unpatentable over VENKITACHALAM et al., and further in view of HAJI SOLEIMANI (US 2025/0110971).
Regarding claim 10, VENKITACHALAM et al. do not specifically disclose the method of claim 9, wherein the first algorithm is a K-S test algorithm, and the second algorithm is a Jensen-Shannon divergence algorithm.
HAJI SOLEIMANI does teach the method of claim 9, wherein the first algorithm is a K-S test algorithm, and the second algorithm is a Jensen-Shannon divergence algorithm (¶ 0150 S2: “Specific and non-limiting distribution similarity/distance metrics include, Kullback-Leibler divergence, Jensen-Shannon divergence (also known as information radius—IRad), Kolmogorov-Smirnov distance, Bhattacharyya distance and Hellinger distance”).
It would have therefore been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to incorporate the “similarity” techniques of HAJI SOLEIMANI into the “cosine similarity” calculations of VENKTACHALAM et al. would enable the latter compatibility with using ready packages for numerous similarity calculations.
Claim(s) 12-13 is/are rejected under 35 U.S.C. 103 as being unpatentable over VENKITACHALAM et al., and further in view of Borgstrom et al. (US 2017/0236520).
Regarding claim 12, VENKITACHALAM et al. do not specifically disclose the method of claim 1, wherein generating the first semantic distribution comprises processing the at least one semantic vector representation via a Gaussian Mixture Model (GMM), and wherein generating the second semantic distribution comprises processing the perturbed semantic vector representation via the GMM.
Borgstrom et al. do teach the method of claim 1, wherein generating the first semantic distribution comprises processing the at least one semantic vector representation via a Gaussian Mixture Model (GMM), and wherein generating the second semantic distribution comprises processing the perturbed semantic vector representation via the GMM ( ¶ 0012 “Generating the universal background model for the prompt” (processing a tool prompt) “using the plurality of feature vectors for each speech unit in the plurality of speech units may involve generating a Gaussian mixture model” (using GMM) “for the universal background model using the expectation-maximization algorithm”).
It would have therefore been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to incorporate the “GMM” model for “prompt” analysis of Borgstrom et al. into the “prompt compression tool” of VENKITACHALAM et al. would enable the combined systems and their associated methods to perform in combination as they do separately and to further enable VENKITACHALAM et al. to benefit with the latter’s “expectation-maximization algorithm” as disclosed in Borgstrom et al. ¶ 0012 last S.
Regarding claim 13, VENKITACHALAM et al. do not specifically disclose the method of claim 1, wherein generating at least one semantic vector representation of the multiple text chunks comprises generating a separate semantic vector representation for each text chunk in the multiple text chunks, and wherein generating the perturbed semantic vector representation comprises generating a separate perturbed semantic vector representation for each text chunk in the multiple text chunks other than the eliminated at least one text chunk.
Borgstrom et al. do teach the method of claim 1, wherein generating at least one semantic vector representation of the multiple text chunks comprises generating a separate semantic vector representation for each text chunk in the multiple text chunks, and wherein generating the perturbed semantic vector representation comprises generating a separate perturbed semantic vector representation for each text chunk in the multiple text chunks other than the eliminated at least one text chunk (¶ 0012 “Generating the universal background model for the prompt” (processing a tool prompt) “using the plurality of feature vectors for each speech unit” (for each of the text chunks) “in the plurality of speech units may involve generating a Gaussian mixture model” (using GMM it generates a separate semantic vector representation) “for the universal background model using the expectation-maximization algorithm”).
For obviousness to combine VENKITACHALAM et al. and Borgstrom et al. see claim 12.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure. SUN XIAOXIAO (CN 117591726).
Any inquiry concerning this communication or earlier communications from the examiner should be directed to FARZAD KAZEMINEZHAD whose telephone number is (571)270-5860. The examiner can normally be reached 10:30 am to 11:30 pm.
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, Paras D. Shah can be reached at (571) 270-1650. 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.
/Farzad Kazeminezhad/
Art Unit 2653
July 25th 2026.
Prosecution Insights