Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Continued Examination Under 37 CFR 1.114
A request for continued examination under 37 CFR 1.114, including the fee set forth in 37 CFR 1.17(e), was filed in this application after final rejection. Since this application is eligible for continued examination under 37 CFR 1.114, and the fee set forth in 37 CFR 1.17(e) has been timely paid, the finality of the previous Office action has been withdrawn pursuant to 37 CFR 1.114. Applicant's submission filed on 5 Feb 2026 has been entered.
Detailed Action
Claims 1-20 are pending. Claims 1, 11, and 20 are independent.
Claims 2-10 depend from Claim 1.
Claims 12-19 depend from Claim 11.
This Application was published as U.S. 2025/0200298.
Response to Amendment
Examiner thanks Applicant for the response filed on 30 Apr 2026 which has been correspondingly accepted and considered in this office action. Claims 1- 20 are pending.
Response to Arguments
With regards to Claims 7-8 and 17-18 Rejected under 35 USC § 112, Applicant has provided arguments, see page 8, filed 30 Apr 2026 and amended claims 7-8 and 17. As a result, amendments to claims and arguments have been fully considered and are persuasive. The rejections of 35 USC § 112 have been withdrawn.
With regards to Claims 1-4, 6-14, and 16-19 Rejected under 35 USC § 103, Applicant has provided arguments, see pages 8-17, filed 30 Apr 2026 and amended claims 1, 7-8, 11, 13, and 17. Arguments and amendments have been considered but they are not persuasive. See below.
Claim 1:
Applicant argues on Page 12 of arguments received on 30 Apr 2026 the following:
Foley's labels therefore identify model provenance, not task identity. That difference matters because a provenance label answers which model generated a response, while the amended claims require a task label identifying what task the language model has been trained to perform and to which the pair relates. Foley also trains an ''attribution model'' to attribute a fine-tuned model to a source foundation model. However, Foley does not use task-labeled prompt-response pairs to train a specialized language model to perform the identified task, as now recited in amended claim 1.
Under MPEP 2111, claims should be interpreted using broadest reasonable interpretation (BRI) in light of the specification.
Here, claim refers to a task that is performed by a language model. Claim does not define what kind of task the model can perform. Under BRI, a task could be the identification of foundation models that are associated with the respective prompt-response pair. Applicant admits that Foley teaches training a “an ''attribution model'' to attribute a fine-tuned model to a source foundation model” (Applicant arguments page 8) which is a task which relates to the respective prompt-response pair.
On the other hand, Applicant argues that “Foley's labels therefore identify model provenance, not task identity.” (Applicant arguments page 8) This is an overly narrow interpretation of Foley. While Foley can provide model provenance, as discussed above Foley can also identify task, such as identifying foundation model that relates to a respective prompt-response pair.
For at least the aforementioned provided reasons, Examiner respectfully notes that said Appellant’s arguments are found unpersuasive. Therefore, the Rejections of Claim 1 and associated dependent claims 2-4 and 6-10 under 35 U.S.C. § 103 be respectfully will be sustained.
Claim 11:
Applicant argues on Page 10 of arguments received on 30 Apr 2026 the following:
With respect to amended claim 11, Carbune also does not disclose training a
specialized language model using prompt-response pairs from an analysis results store that are tagged as relating to the particular task. Carbune' s stored positive and negative examples are used for later selection, prompt reuse, RLHF, and tuning. Carbune does not disclose an analysis results store of task-tagged prompt-response pairs that feeds training of a specialized language model for the tagged task.
Under MPEP 2111, claims should be interpreted using broadest reasonable interpretation (BRI) in light of the specification.
Here, claim refers to an analysis results store. Claim does not define what is an analysis results store. Under BRI, an analysis results store is anything that can store analytical results as long as the data is tagged somehow and relates to a particular task. Carbune teaches business LLM that includes an external datastore (165) that provides services, such as requesting the “assistant LLM 150 to search for business LLMs having service capabilities specified by the discovery search” (Par [0042]) (i.e. booking flights on Delta) for the prompt-response pair relating to a particular task.
On the other hand, Applicant does not provide any evidence other than the conclusory assertion that “Carbune does not disclose an analysis results store of task-tagged prompt-response pairs.” (Applicant arguments page 10) This statement alone without any additional evidence is not sufficient to conclude that Carbune does not teach a analysis results store.
For at least the aforementioned provided reasons, Examiner respectfully notes that said Appellant’s arguments are found unpersuasive. Therefore, the Rejection of Claim 11 and associated dependent claims 12-14 and 16-19 under 35 U.S.C. § 103 be respectfully will be sustained.
With regards to Claims 5, 15, and 20 rejected under 35 USC § 103, Applicant has provided arguments, see pages 8-17, filed 30 Apr 2026 and amended claims 20. Arguments and amendments have been fully considered and are persuasive. The 35 USC § 103 rejections of claims 5, 15, and 20 have been withdrawn.
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 1-3, 7-13, and 17-19 are rejected under 35 U.S.C. 103 as being unpatentable over Carbune et al.(US2025/0069617 hereinafter Carbune) in view of Foley et al. (US2025/0028992 hereinafter Foley)
With regards to claim 1, Carbune teaches:
A method comprising: obtaining, by a device, prompt-response pairs of prompts for input to a language model trained to perform a plurality of tasks and their corresponding responses from the language model; [Carbune Fig 1 Par [0030] teaches user device 110 obtains prompt (152) and response (162) which are prompt-response pairs from language model (business LLM item 160) used for input into language model (150) where each business LLM performs at least one task]
training, by the device, a specialized language model to perform a particular task using a training set comprising the prompt-response pairs assigned a particular task label identifying the particular task; and [Carbune Fig 1 teaches specialized language model (150) that uses business LLM (160) that “perform the action/task on behalf of the user 10” (Par [0038]) where each business LLM identifies the particular tasks, for example a single task business LLM identifies the particular task label for that task. Specialized LLM (150) can be trained using “positive examples” (Par [0061]) and “negative examples” (Par [0063]) by a “training mode” (Par [0061]) where positive examples include a training set that comprises “respective prompts 152 created and issued to the business LLMs, the respective response content 162” (Par [0059]) which are the prompt-response pairs assigned to the corresponding to the particular task which can be labeled “successful interactions” (Par [0061]) which label the particular business LLM task as being successful or not. Similarly negative labels (Par [0060, 63]) are also task labels for unsuccessful business LLM tasks.]
causing, by the device, the specialized language model to be deployed for use to perform the particular task. [Carbune Fig 1 teaches “user 10 inputs, via a user device 110, a natural language query 116 to the assistant interface 150 specifying a particular action the user 10 wants the assistant interface 150 to perform on behalf of the user 10”(Par[0030]) to perform the particular task of the associated business LLM]
With regards to claim 1, Carbune fails to teach:
classifying, by the device, each of the prompt-response pairs as relating to one or more tasks by applying a machine-learning classifier to a respective prompt-response pair to assign one or more task labels to that respective prompt-response pair, each task label identifying a task of the plurality of tasks that the language model has been trained to perform and to which that respective prompt-response pair relates;
With regards to claim 1, Foley teaches:
classifying, by the device, each of the prompt-response pairs as relating to one or more tasks by applying a machine-learning classifier to a respective prompt-response pair to assign one or more task labels to that respective prompt-response pair, each task label identifying a task of the plurality of tasks that the language model has been trained to perform and to which that respective prompt-response pair relates; [Foley Fig 3 teaches classifying by a device (Par [0052,59]) where module (330) uses “prompts and their corresponding responses” (Par [0063]) which are respective prompt-response pairs related to the task of training a model, and “Module 330 labels a generated prompt response with the model that generated the prompt response.” (Par [0062]).
While Foley teaches the specific task of identifying a foundation model and labeling the foundation model and associated prompt-response pair, a broader reading of Foley is that each label identifies a task (task of identifying a foundation model) of the plurality of tasks (plurality of foundation models to be identified) that the language model (i.e. LLM or BERT (Par [0063])) has been trained to perform, and to which the respective prompt-response pair relates (foundation model that relates to the prompt-response pair (Par [0062])) where Foley Par [0064,65] identify different embodiments of classifying the foundation model.
It would be obvious to one of ordinary skill in the art at the time of applicant’s filing to combine the method of enabling multiple business LLMs from a user as taught by Carbune using the method of training models using prompt-response pairs as taught by Foley. The motivation to combine the teachings of Carbune with Foley is because “model is being trained to classify LLMs, … [and] image processing models” (Par [0063]) which increases the capabilities of the invention of Carbune to train to new models based on the new prompting functions]
With regards to claim 2, Carbune in view of Foley teaches:
All the limitations of claim 1
wherein the language model is a large language model trained to perform a plurality of tasks. [Carbune Fig 1 teaches language model (150) is an assistant LLM trained to perform tasks using business LLMs (160a-n) where “each multiple different business LLMs 160 that span a diverse set of LLM capabilities” (Par [0036]) which means each LLM can perform a plurality of tasks]
With regards to claim 3, Carbune in view of Foley teaches:
All the limitations of claim 1
wherein obtaining the prompt-response pairs comprises: intercepting, by the device, the prompts for input to the language model and their corresponding responses from the language model. [Carbune Fig 1 teaches the device has a user interface (170) that interacts with the assistant LLM (150) to receive prompts for input to the language model (items 152 and 160) and corresponding responses (item 162)]
With regards to claim 7, Carbune in view of Foley teaches:
All the limitations of claim 1
wherein obtaining the prompt-response pairs includes associating a response with a prompt based on one or more of session identifiers associated with the prompt and the response, timing information associated with the prompt and the response, or source and destination information associated with the prompt and the response. [Carbune Fig 1 teaches session identifier associated with the user (10) and business LLM (160) which are source and destination information associated with the prompt and the response]
With regards to claim 8, Carbune in view of Foley teaches:
All the limitations of claim 1
wherein, for at least one of the prompt-response pairs, the one or more task labels assigned to that prompt-response pair indicate a plurality of tasks that include the particular task, the method further comprising: [Carbune Fig 1 Par [0030] teaches user device 110 obtains prompt (152) and response (162) which are prompt-response pairs from language model (business LLM item 160) used for input into language model (150) where each business LLM performs at least one task that includes a particular task]
training, by the device, a plurality of language models that include the specialized language model to each perform one of the plurality of tasks. [Carbune Fig 1 teaches business LLMs items (160a-n) where “each multiple different business LLMs 160 that span a diverse set of LLM capabilities” (Par [0036]) which means each specialized LLM can perform one of the plurality of tasks]
With regards to claim 9, Carbune in view of Foley teaches:
All the limitations of claim 1
wherein the prompts for input to the language model are received via a user interface. [Carbune Fig 1 item 170]]
With regards to claim 10, Carbune in view of Foley teaches:
All the limitations of claim 9
wherein the language model is cloud-hosted. [Carbune teaches “business LLM 160 that is backed by a particular cloud service provider” (Par [0035])]
With regards to claim 11, Carbune teaches:
An apparatus, comprising: one or more network interfaces; [Carbune Fig 1 teaches the “network 130 may be wired, wireless, or a combination thereof, and may include private networks and/or public networks, such as the Internet” (Par [0033])]
a processor coupled to the one or more network interfaces and configured to execute one or more processes; and [Carbune Fig 1 teaches remote computing system which includes “data processing hardware” and Fig 5 teaches processor (510) for computing device (Par [0072])]
a memory configured to store a process that is executable by the processor, the process when executed configured to: [Carbune Fig 5 teaches memory (520) executable by processor (510) (Par [0072])]
obtain prompt-response pairs of prompts for input to a language model and their corresponding responses from the language model; [Carbune Fig 1 Par [0030] teaches user device 110 obtains prompt (152) and response (162) from language model (business LLM item 160)]
train a specialized language model to perform a particular task using prompt-response pairs from an analysis results store that are tagged as relating to the particular task; and [Carbune Fig 1 teaches specialized language model (150) that uses business LLM (160) that “perform the action/task on behalf of the user 10” (Par [0038]) which are particular tasks. Specialized LLM (150) can be trained using “positive examples” (Par [0061]) and “negative examples” (Par [0063]) by a “training mode” (Par [0061]) where positive examples include a training set that comprises “respective prompts 152 created and issued to the business LLMs, the respective response content 162” (Par [0059]) which are the prompt-response pairs assigned to the corresponding to the particular task which can be labeled “successful interactions” (Par [0061]) which label the particular business LLM task as being successful or not. Similarly negative labels (Par [0060, 63]) are also task labels for unsuccessful business LLM tasks. Furthermore, Carbune teaches business LLM that includes an external datastore (165) that provides services, such as requesting the “assistant LLM 150 to search for business LLMs having service capabilities specified by the discovery search” (Par [0042]) (i.e. booking flights on Delta) for the prompt-response pair relating to a particular task.]
cause the specialized language model to be deployed for use to perform the particular task. [Carbune Fig 1 teaches “presentation content 180 based on the response content 162 returned provided by each business LLM 160 that performed a corresponding portion of the action on behalf of the user 10”]
With regards to claim 11, Carbune fails to teach:
classify each of the prompt-response pairs as relating to one or more tasks by applying a machine-learning classifier to a respective prompt-response pair to assign one or more task labels to that respective prompt-response pair;
With regards to claim 11, Foley teaches:
classify each of the prompt-response pairs as relating to one or more tasks by applying a machine-learning classifier to a respective prompt-response pair to assign one or more task labels to that respective prompt-response pair; [Foley Fig 3 teaches classifying by a device (Par [0052,59]) where module (330) uses “prompts and their corresponding responses” (Par [0063]) which are prompt-response pairs related to the task of training a model, and “Module 330 labels a generated prompt response with the model that generated the prompt response.” (Par [0062]) where module (330) is a classifier model such as an LLM (Par [0063])
It would be obvious to one of ordinary skill in the art at the time of applicant’s filing to combine the method of enabling multiple business LLMs from a user as taught by Carbune using the method of training models using prompt-response pairs as taught by Foley. The motivation to combine the teachings of Carbune with Foley is because “model is being trained to classify LLMs, … [and] image processing models” (Par [0063]) which increases the capabilities of the invention of Carbune to train to new models based on the new prompting functions]
Claim 12 is a system claim with limitations corresponding to the limitations of method Claim 2 and is rejected under similar rationale.
Claim 13 is a system claim with limitations corresponding to the limitations of method Claim 3 and is rejected under similar rationale.
With regards to claim 17, Carbune in view of Foley teaches:
All the limitations of claim 11
Wherein, for at least one of the prompt-response pairs, the one or more task labels assigned to that prompt-response pair indicate a plurality of tasks that include the particular task. [Foley Fig 3 teaches classifying by a device (Par [0052,59]) where module (330) uses “prompts and their corresponding responses” (Par [0063]) which are prompt-response pairs related to the task of training a model, and “Module 330 labels a generated prompt response with the model that generated the prompt response.” (Par [0062]) where module (330) is a classifier model such as an LLM (Par [0063])
Furthermore, as previously discussed, Foley teaches the task of identifying a foundation model and labeling the foundation model and associated prompt-response pair. For at least on task label assigned to the prompt-response pair indicating a plurality of tasks (plurality of foundation models to be identified) that include the particular task (the particular foundation model associated with the prompt-response pair) (Par [0062])]
Claim 18 is a system claim with limitations corresponding to the limitations of method Claim 8 and is rejected under similar rationale.
Claim 19 is a system claim with limitations corresponding to the limitations of method Claim 9 and is rejected under similar rationale.
Claims 4, 6, 14, and 16 are rejected under 35 U.S.C. 103 as being unpatentable over Carbune et al.(US2025/0069617) in view of Foley et al. (US2025/0028992) in further view of Mimassi (US2022/0383859) hereinafter Mimassi.
With regards to claim 4, Carbune in view of Foley teaches:
All the limitations of claim 1
With regard to claim 4, Carbune in view of Foley fails to teach:
wherein the specialized language model is deployed to an edge node for execution.
With regard to claim 4, Mimassi teaches:
wherein the specialized language model is deployed to an edge node for execution. [Mimassi Fig 2 Par [0049] teaches “edge devices may operate a default local version of the RNN 205 based upon the received model parameters.”
It would be obvious to one of ordinary skill in the art at the time of applicant’s filing to combine the method of enabling multiple business LLMs from a user as taught by Carbune and Foley with execution of a local version of the LLM on an edge device as taught by Mimassi. The motivation to combine the teachings of Carbune and Foley with the teachings of Mimassi is because “system 100 can leverage the computing power of edge devices to train the local models operating on the edge devices, and update the central language models 204 periodically” (Mimassi Par [0049]) which increases the capabilities of the invention of Carbune and Foley to adapt to new training based on the new prompting functions]
With regards to claim 6, Carbune in view of Foley teaches:
All the limitations of claim 1
With regard to claim 6, Carbune in view of Foley fails to teach:
wherein the particular task comprises generating a configuration or script for use by a networking device.
With regard to claim 6, Mimassi teaches:
wherein the particular task comprises generating a configuration or script for use by a networking device. [Mimassi Fig 11 teaches computing device (10) uses program instructions or “for example, scripts written in Python, Perl, Ruby, Groovy, or any other scripting language” (Par [0089]) for use by a networking device.
It would be obvious to one of ordinary skill in the art at the time of applicant’s filing to combine the method of enabling multiple business LLMs from a user as taught by Carbune and Foley with execution of a local version of the LLM on an edge device using scripts as taught by Mimassi. The motivation to combine the teachings of Carbune and Foley with the teachings of Mimassi is because “system 100 can leverage the computing power of edge devices to train the local models operating on the edge devices, and update the central language models 204 periodically” (Mimassi Par [0049]) which increases the capabilities of the invention of Carbune in view of Foley to adapt to new training based on the new prompting functions]
Claim 14 is a system claim with limitations corresponding to the limitations of method Claim 4 and is rejected under similar rationale.
Claim 16 is a system claim with limitations corresponding to the limitations of method Claim 6 and is rejected under similar rationale.
Allowable Subject Matter
Claim 20 allowed.
Claims 5 and 15 objected to as being dependent upon a rejected base claim, but would be allowable if rewritten in independent form including all of the limitations of the base claim and any intervening claims.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to Joseph J Yamamoto whose telephone number is (571)272-4020. The examiner can normally be reached M-F 1000-1800 EST.
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, Bhavesh Mehta can be reached at 571-272-7453. 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.
JOSEPH J. YAMAMOTO
Examiner
Art Unit 2656
/BHAVESH M MEHTA/Supervisory Patent Examiner, Art Unit 2656
Prosecution Insights