Granted Patent B2

US 12,530,591 · App. 18/750,394 · Granted Jan 20, 2026

Inventor: Lawrence Marc Ausubel (Washington, DC)

Assignee: ChatIP LLC

G06N3/091G06N3/096G06N3/098

Abstract

Claims (84)

1 . A computer-implemented method for intervening in an artificial intelligence (AI) model, the method comprising:

obtaining a first prompt from a user computer, the first prompt comprising first data;

providing the first prompt as input to a first AI model, wherein the first AI model is configured to output keywords based on the input to the first AI model;

obtaining a first response to the first prompt from the first AI model, wherein the first response comprises a set of one or more keywords;

using i) the set of keywords included in the first response to the first prompt from the first AI model and ii) a database associating intervention information with keywords to obtain from the database first intervention information associated with one or more of the keywords included in the set of keywords;

generating second data using the first intervention information that was obtained from the database using the set of keywords included in the first response to the first prompt from the first AI model;

generating a second prompt by combining or concatenating the first data and the second data generated using the first intervention information that was obtained from the database using the set of keywords included in the first response to the first prompt from the first AI model;

providing as input to a second AI model the second prompt generated from the first data from the first prompt and the second data generated using the first intervention information that was obtained from the database using the set of keywords included in the first response to the first prompt from the first AI model;

obtaining a second response to the second prompt from the second AI model; and

sending the second response towards the user computer, wherein

the second AI model comprises at least a first transformer model,

the method further comprises a model training or fine-tuning process that comprises: feeding preprocessed data to the first transformer model, calculating a loss, and using the calculated loss to update parameters of the first transformer model, and

the model training or fine-tuning process is iterated at least twice.

2 . The method of claim 1 , wherein

using i) the set of keywords included in the first response to the first prompt from the first AI model and ii) the database to obtain intervention information from the database comprises using the set of keywords to retrieve the first intervention information from the database.

3 . The method of claim 1 , further comprising, prior to obtaining the first intervention information using the set of keywords:

obtaining from a stakeholder computer the first intervention information; and

adding the first intervention information to the database.

4 . The method of claim 1 , wherein the first intervention information is based upon or comprises a rating and/or a comment.

5 . The method of claim 1 , wherein the first intervention information is based upon or comprises a bid.

6 . The method of claim 1 , wherein

a first weight is associated with the first intervention information, and

the first weight indicates an amount by which the second data is to be weighted by the second AI model.

7 . The method of claim 6 , further comprising

incorporating the first weight into the second prompt.

8 . The method of claim 7 , further comprising:

incorporating into the second prompt a second weight for organic information contained in the second AI model, wherein the organic information comprises information available to the second AI model in response to the first prompt.

9 . The method of claim 1 , further comprising:

identifying a first portion of the second response comprising an option associated with the first intervention information and a second portion of the second response comprising an option not associated with the first intervention information; and

applying a first label to the first portion of the second response and a second label to the second portion of the second response before sending the second response towards the user computer.

10 . The method of claim 9 , wherein the first label comprises:

a first color different from a second color used in the second label,

a first typeface different from a second typeface used in the second label,

a first symbol different from a second symbol used in the second label, and/or

a first text character different from a second text character used in the second label.

11 . The method of claim 10 , wherein the first portion of the second response comprises a hyperlink associated with the option associated with the first intervention information.

12 . The method of claim 1 , wherein the second AI model is a large language model.

13 . The method of claim 1 , further comprising:

masking the second prompt from the user computer.

14 . The method of claim 1 , wherein the first prompt is a search request.

15 . The method of claim 1 , wherein the second AI model is comprised in a search engine system.

16 . An apparatus for intervening in an artificial intelligence (AI) model, the apparatus comprising:

processing circuitry; and

memory storing instructions, executable by the processing circuitry, for configuring the apparatus to perform a method comprising:

obtaining a first prompt from a user computer, the first prompt comprising first data;

providing the first prompt as input to a first AI model, wherein the first AI model is configured to output keywords based on the input to the first AI model;

obtaining a first response to the first prompt from the first AI model, wherein the first response comprises a set of one or more keywords;

using i) the set of keywords included in the first response to the first prompt from the first AI model and ii) a database associating intervention information with keywords to obtain from the database first intervention information associated with one or more of the keywords included in the set of keywords;

generating second data using the first intervention information that was obtained from the database using the set of keywords included in the first response to the first prompt from the first AI model;

generating a second prompt by combining or concatenating the first data and the second data generated using the first intervention information that was obtained from the database using the set of keywords included in the first response to the first prompt from the first AI model;

providing as input to a second AI model the second prompt generated from the first data from the first prompt and the second data generated using the first intervention information that was obtained from the database using the set of keywords included in the first response to the first prompt from the first AI model;

obtaining a second response to the second prompt from the second AI model; and

sending the second response towards the user computer, wherein

the second AI model comprises at least a first transformer model,

the method further comprises a model training or fine-tuning process that comprises: feeding preprocessed data to the first transformer model, calculating a loss, and using the calculated loss to update parameters of the first transformer model, and

the model training or fine-tuning process is iterated at least twice.

17 . The apparatus of claim 16 , wherein

using i) the set of keywords included in the first response to the first prompt from the first AI model and ii) the database to obtain intervention information from the database comprises using the set of keywords to retrieve the first intervention information from the database.

18 . The apparatus of claim 16 , wherein the method further comprises, prior to obtaining the first intervention information using the set of keywords:

obtaining from a stakeholder computer the first intervention information; and

adding the first intervention information to the database.

19 . The apparatus of claim 16 , wherein the first intervention information is based upon or comprises a rating and/or a comment.

20 . The apparatus of claim 16 , wherein the first intervention information is based upon or comprises a bid.

21 . The apparatus of claim 16 , wherein

a first weight is associated with the first intervention information, and

the first weight indicates an amount by which the second data is to be weighted by the second AI model.

22 . The apparatus of claim 21 , wherein the method further comprises

incorporating the first weight into the second prompt.

23 . The apparatus of claim 22 , wherein the method further comprises:

incorporating into the second prompt a second weight for organic information contained in the second AI model, wherein the organic information comprises information available to the second AI model in response to the first prompt.

24 . The apparatus of claim 16 , wherein the method further comprises:

identifying a first portion of the second response comprising an option associated with the first intervention information and a second portion of the second response comprising an option not associated with the first intervention information; and

applying a first label to the first portion of the second response and a second label to the second portion of the second response before sending the second response towards the user computer.

25 . The apparatus of claim 24 , wherein the first label comprises:

a first color different from a second color used in the second label,

a first typeface different from a second typeface used in the second label,

a first symbol different from a second symbol used in the second label, and/or

a first text character different from a second text character used in the second label.

26 . The apparatus of claim 25 , wherein the first portion of the second response comprises a hyperlink associated with the option associated with the first intervention information.

27 . The apparatus of claim 16 , wherein the second AI model is a large language model.

28 . The apparatus of claim 16 , wherein the method further comprises:

masking the second prompt from the user computer.

29 . The apparatus of claim 16 , wherein the first prompt is a search request.

30 . The apparatus of claim 16 , wherein the second AI model is comprised in a search engine system.

Assignments (1)

Continuity (6)

References Cited (38)