DETAILED ACTION
This communication is in response to the Amendments and Arguments filed on 1/27/2026.
Claims 1-4, 6, 8-11, 13, and 15-18 are pending and have been examined.
All previous objections / rejections not mentioned in this Office Action have been withdrawn by the examiner.
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 .
Response to Amendments
Applicant has amended independent claims 1, 8, and 15. Applicant has not provided remarks considering 35 U.S.C. § 102 and § 103 rejections. New references have been applied.
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.
Claims 1, 2, 4, 6, 8, 9, 11, 13, 15, 16, and 18 are rejected under 35 U.S.C. 103 as being unpatentable over Peng et al. (U.S. PG Pub No. 20240070394), hereinafter Peng, in view of Hajarnis et al. (U.S. PG Pub No. 20240330863), hereinafter Hajarnis, and further view of “Training language models to follow instructions with human feedback” by Ouyang et al., hereinafter Ouyang.
Regarding claim 1, 8, and 15 Peng teaches:
(Claim 1) A computer-implemented method comprising: (P0052, Soft Prompt Tuning module that may be used to implement and/or emulate the systems and models, and/or to implement any of the methods described further herein.)
(Claim 8) A system comprising: one or more processors; and memory storing thereon instructions that, as a result of being executed by the one or more processors, cause the system to perform operations comprising: (P0049, As shown in FIG. 4, computing device includes a processor coupled to memory.; P0052, Executable code that when run by one or more processors (e.g., processor) may cause the one or more processors to perform the methods described in further detail herein.)
(Claim 15) A non-transitory, computer-readable storage medium storing thereon executable instructions that, as a result of being executed by one or more processors of a computer system, cause the computer system to perform operations comprising: (P0052, Memory may include non-transitory, tangible, machine readable media that includes executable code that when run by one or more processors (e.g., processor) may cause the one or more processors to perform the methods described in further detail herein.)
accessing input data that includes initial content; (P0052, Ensembled Soft Prompt Tuning module may receive input such as an input training data. … Examples of the input data may include other types of natural language inputs such as a document, a text, etc.)
determining a content format of customized content to be generated by processing the input data, wherein the content format specifies how the customized content is to be formatted for a target recipient; (P0031, Ensembled soft prompts trained as described in FIG. 1 to train an attention module that generates a classification logit for a target input from a target training dataset. … The ensembled soft prompt tuning framework shown in FIG. 2 may be used to train an attention module so that the framework can make predictions for a target task.; P0052, Examples of the output data may include an answer, a summary, an intent classification label, and/or the like.)
applying tokenization algorithms to encode the input data and the content format into a feature vector; (P0077, The soft prompts may be a sequence of tokens or vectors and the input may also be a sequence of different tokens or vectors.)
processing the feature vector using a prompt machine-learning model to extract one or more insight prompts from the input data and the content format; (P0023, Pre-trained language model generates a task-specific output logit from a concatenation of the input data and soft prompt. Normalizing the task-specific output logit with a softmax operation generates a predicted source output Y corresponding to the respective source task. … The cross-entropy loss may then be used to backpropagate to update the soft prompt.)
applying a content machine-learning model to the one or more insight prompts to generate the customized content, wherein using the one or more insight prompts by the content machine-learning model improves performance of the content machine-learning model in accurately generating the customized content; (P0023, Then a pre-trained language model generates a task-specific output.)
outputting the customized content. (P0023, Then a pre-trained language model generates a task-specific output logit (e.g., 114, 124, 154) from a concatenation of the input data and soft prompt. Normalizing the task-specific output logit with a softmax operation generates a predicted source output Y (e.g., 118, 126, 156) corresponding to the respective source task.)
Peng does not specifically teach:
applying tokenization algorithms to encode the input data and the content format into a feature vector;
receiving in real-time text data provided by the target recipient, wherein the text data corresponds to modifications to the customized content; and
training the content machine-learning model based on the text data by updating parameters of the content machine-learning model via reinforcement learning, wherein the content machine-learning model is trained to enhance accuracy of generating future customized content for the target recipient.
Hajarnis, however, teaches:
applying tokenization algorithms to encode the input data and the content format into a feature vector; (P0053, Preprocessing layer may include one or more sub-layers of a token embedding layer, a segment embedding layer, and a position embedding layer. The token embedding layer may convert a word into a vector of token values, where the vector of token values has a predetermined dimension (e.g., a vector of values). The token embedding layer may tokenize the word using a certain tokenization method such as the WordPiece method which is a data-driven method.)
receiving in real-time text data provided by the target recipient, wherein the text data corresponds to modifications to the customized content; and (P0098, The output of the large language model can also be subject to the checks of inclusivity and exclusivity, including prompting on user interfaces for manual substitution and recommendation.)
It would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to apply tokenization and receive modifications from target recipient. It would have been obvious to combine the references because encoding using a tokenization is a known technique to yield a predictable result of creating embedding vectors for use in neural networks and receiving modifications provides a layer for inclusiveness check when LLM outputs exclusive words that should not be output. (Hajarnis P0081)
Peng in view of Hajarnis does not specifically teach:
training the content machine-learning model based on the text data by updating parameters of the content machine-learning model via reinforcement learning, wherein the content machine-learning model is trained to enhance accuracy of generating future customized content for the target recipient.
Ouyang, however, teaches:
training the content machine-learning model based on the text data by updating parameters of the content machine-learning model via reinforcement learning, wherein the content machine-learning model is trained to enhance accuracy of generating future customized content for the target recipient. (1. Introduction, We focus on fine-tuning approaches to aligning language models. Specifically, we use reinforcement learning from human feedback to fine-tune GPT-3 to follow a broad class of written instructions (see Figure 2).)
It would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to utilize reinforcement learning when training the content machine-learning model. It would have been obvious to combine the references because reinforcement learning allows for better alignment of language models with user intent on a wide range of tasks . (Ouyang, Abstract)
Regarding claim 2, 9, and 16 Peng in view of Hajarnis and further view of Ouyang teach claim 1, 8, and 15.
Peng does not specifically teach:
determining persona data that identifies characteristics associated with the target recipient, wherein the one or more insight prompts are generated further based on the persona data.
Hajarnis, however, teaches:
determining persona data that identifies characteristics associated with the target recipient, wherein the one or more insight prompts are generated further based on the persona data. (P0039, Processing device may rewrite and enhance the query prompt based on the one or more skills, where the prompt is designed to inquire the large language model. The one or more skills may be used to enrich the prompt. For example, if the one or more skills include “deep learning,” “TensorFlow,” “C++” in addition to Python, processing device may rewrite the user-specified prompt of “write a job description for a Data Scientist with two-year Python experience” to “write a job description for a Data Scientist with two-year experience of Python, deep learning, TensorFlow, and C++,” thus generating a more precise prompt for eliciting more accurate job description.)
It would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to determine persona data and generate prompt based on the persona data. It would have been obvious to combine the references because large language model produces more precise and unbiased answers if prompt is phrased with specific attributes and not broadly phrased. (Hajarnis P0032)
Regarding claim 4, 11, and 18 Peng in view of Hajarnis and further view of Ouyang teach claim 1, 8, and 15.
Peng further teaches:
wherein the initial content is previously generated by applying an initial machine-learning model to raw data. (P0099, In PLG, source prompts were tuned on target data sets with the few-shot target training data. Then the trained source prompts are used with the pre-trained language model to generate pseudo labels for each for the entire target dataset. The final pseudo label for each target sample is determined by majority voting of the 6 source tasks. The samples with final pseudo labels are then used to train the soft prompt of the target task before few-shot prompt tuning with the few shot target training data.)
Regarding claim 6 and 13 Peng in view of Hajarnis and further view of Ouyang teach claim 1 and 8.
Peng further teaches:
wherein the content format includes an email, a memorandum, a slide deck, an executive briefing, or mitigation strategies. (P0031, The ensembled soft prompt tuning framework shown in FIG. 2 may be used to train an attention module so that the framework can make predictions for a target task.; P0052, Examples of the output data may include an answer, a summary, an intent classification label, and/or the like.)
Claims 3, 10, and 17 are rejected under 35 U.S.C. 103 as being unpatentable over Peng, in view of Hajarnis, in view of Ouyang, and further view of "Prompt Learning for Domain Adaptation in Task-Oriented Dialogue" by Sreedhar et al., hereinafter Sreedhar.
Regarding claim 3, 10, and 17 Peng in view of Hajarnis and further view of Ouyang teach claim 1, 8, and 15.
Peng does not specifically teach:
determining domain data that identifies characteristics associated with a domain associated with the customized content, wherein the one or more insight prompts are generated further based on the domain data.
Sreedhar, however, teaches:
determining domain data that identifies characteristics associated with a domain associated with the customized content, wherein the one or more insight prompts are generated further based on the domain data. (1. Introduction, We observe that using such canonical forms as labels for the intent classification task allows the model to generalize better to domains that are adjacent, but not seen at train time (e.g., Flight Reservations Bus Bookings). We also find that it is beneficial to do a two-stage P-tuning for domain adaption, i.e., once we have a p-tuned large language model on a wide set of domains, we can continue p-tuning this model on a small set of labelled samples from the target domain to allow the model to generalize better.)
It would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to determine domain data and generate prompt based on the domain data. It would have been obvious to combine the references because language models such as BERT are not immune to the challenge of adapting models to new domains and labelling the prompt to the target domain allow the model to generalize better. (Sreedhar, 1. Introduction)
Conclusion
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to DANIEL WONSUK CHUNG whose telephone number is (571)272-1345. The examiner can normally be reached Monday - Friday (7am-4pm)[PT].
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, PIERRE-LOUIS DESIR can be reached at (571)272-7799. 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.
/DANIEL W CHUNG/Examiner, Art Unit 2659
/PIERRE LOUIS DESIR/Supervisory Patent Examiner, Art Unit 2659
Prosecution Insights