Privacy enhanced language model prompt

1 . A computing device, comprising:

a memory; and

at least one processor coupled to the memory and configured to:

receive a user prompt from a user;

process the user prompt to generate a privacy enhanced user prompt in which privacy information or other information that may cause a large generative artificial intelligence model (LXM) to provide a response that reveals the privacy information is modified, substituted, or removed;

provide the privacy enhanced prompt to the LXM;

receive a plurality of sample responses from the LXM;

for each sample response of the plurality of sample responses, determine a ranking score for the sample response, the ranking score based, at least in part, on:

a relevance of the sample response to the user prompt;

a relevance of the sample response to physical context information indicative of a current physical condition of the user or of an environment of the user;

user background information comprising data indicative of characteristics, preferences, or demographics of the user;

a model confidence score indicative of a likelihood that the sample response is accurate;

privacy preservation based on evaluating whether the sample response would reveal privacy information; or

a combination thereof;

select, as a best output, a sample response of the plurality of sample responses based, at least in part, on the ranking scores; and

provide a response to the user based on the best output.

2 . The computing device of claim 1 , wherein the at least one processor is further configured to process the user prompt to generate the privacy enhanced user prompt using a trained model that has been trained to recognize the privacy information.

3 . The computing device of claim 1 , wherein the at least one processor is further configured to contextualize the best output to include privacy information in the user response.

4 . The computing device of claim 1 , wherein the at least one processor is further configured to:

obtain user context information from a source of physical context information and user background information;

identify user context information and user background information that could pose a privacy risk; and

generate the privacy enhanced user prompt based on substitution, modification, or removal of user context information and user background information that is identified to pose the privacy risk.

5 . The computing device of claim 1 , wherein the at least one processor is further configured to:

determine whether the user is in a location where the response to the user could be viewed by others; and

avoiding disclosure of privacy information by:

preventing presentation of the response to the user based on determining that the user is in a location where the response to the user could be viewed by others and further based on determining that the response to the user includes privacy data.

6 . The computing device of claim 5 , wherein the at least one processor is further configured to notify the user that the response to the user is not being presented to prevent the disclosure of privacy information.

7 . The computing device of claim 4 , wherein the at least one processor is further configured to:

obtain local context information from a data source available on a local context database; and

generate the privacy enhanced user prompt based, at least in part, on the obtained context information.

8 . A method of interacting with a large generative artificial intelligence model (LXM), comprising:

receiving a user prompt from a user;

processing the user prompt to generate a privacy-enhanced user prompt in which privacy information or other information that may cause the LXM to provide a response that will reveal the privacy information is modified, substituted, or removed;

providing the privacy-enhanced prompt to the LXM;

receiving a plurality of sample responses from the LXM;

for each sample response of the plurality of sample responses, determine a ranking score for the sample response, the ranking score based, at least in part, on:

a relevance of the sample response to the user prompt;

a relevance of the sample response to physical context information indicative of a current physical condition of the user or of an environment of the user;

user background information comprising data indicative of characteristics, preferences, or demographics of the user;

a model confidence score indicative of a likelihood that the sample response is accurate;

privacy preservation based on evaluating whether the sample response would reveal privacy information; or

a combination thereof;

select, as a best output, a sample response of the plurality of sample responses based, at least in part, on the ranking scores; and

providing a response to a user based on the best output.

9 . The method of claim 8 , wherein generating the privacy-enhanced user prompt comprises processing the user prompt using a trained model that has been trained to recognize the privacy information.

10 . The method of claim 8 , further comprising:

contextualizing the best output to include privacy information in the user response.

11 . The method of claim 8 , further comprising:

obtaining user context information from a source of physical context information and user background information; and

identifying user context information and user background information that could pose a privacy risk; and

generating the privacy enhanced user prompt based on substitution, modification, or removal of user context information and user background information that is identified as posing a privacy risk.

12 . The method of claim 8 , further comprising:

determining whether the user is in a location where the response to the user could be viewed by others; and

avoiding disclosure of privacy information preventing presentation of the response to the user based on determining that the user is in a location where the response to the user could be viewed by others and further based on determining that the response to the user includes privacy data.

13 . The method of claim 12 , further comprising notifying the user that the response to the user not being presented to prevent the disclosure of privacy information.

14 . The method of claim 11 , further comprising:

obtaining local context information from a data source available on a local context database; and

generating the privacy enhanced prompt based, at least in part, on the obtained user context information.

15 . A computing device, comprising:

means for receiving a user prompt from a user;

means for processing the user prompt to generate a privacy enhanced user prompt in which privacy information or other information may cause a large generative artificial intelligence model (LXM) to provide a response that reveals the privacy information is modified, substituted, or removed;

means for providing the privacy enhanced prompt to the LXM;

means for receiving a plurality of sample responses from the LXM;

means for determining a ranking score for each sample response the plurality of sample responses, the ranking score based, at least in part, on:

a relevance of the sample response to the user prompt;

a relevance of the sample response to physical context information indicative of a current physical condition of the user or of an environment of the user;

user background information comprising data indicative of characteristics, preferences, or demographics of the user;

a model confidence score indicative of a likelihood that the sample response is accurate;

privacy preservation based on evaluating whether the sample response would reveal privacy information; or

a combination thereof;

means for selecting, as a best output, a sample response of the plurality of sample responses based, at least in part, on the ranking scores; and

means for providing a response to the user based on the best output.

16 . The computing device of claim 15 , wherein the means for processing the user prompt to generate the privacy enhanced user prompt comprises means for processing the user prompt in a trained model that has been trained to recognize privacy information.

17 . The computing device of claim 15 , further comprising:

means for contextualizing the best output to include privacy information in the user response.

18 . The computing device of claim 15 , further comprising:

means for obtaining user context information from a source of physical context information and user background information; and

means for generating the privacy enhanced user prompt based, at least in part, on the user context information and user background information.

19 . The computing device of claim 15 , further comprising:

means for determining whether the user is in a location where the response to the user could be viewed by others; and

means for preventing presentation of the response to the user based on determining that the user is in a location where the response to the user could be viewed by others and further based on determining that the response to the user includes privacy data.

20 . A non-transitory processor readable media having stored thereon processor-executable instructions configured to cause at least one processor of a computing device to perform operations comprising:

receiving a user prompt from a user;

processing the user prompt to generate a privacy enhanced user prompt in which privacy information or other information that may cause large generative artificial intelligence model (LXM) to provide a response that reveals the privacy information is modified, substituted, or removed;

providing the privacy enhanced prompt to the LXM;

receiving a plurality of sample responses from the LXM;

for each sample response of the plurality of sample responses, determine a ranking score for the sample response, the ranking score based, at least in part, on:

a relevance of the sample response to the user prompt;

a relevance of the sample response to physical context information indicative of a current physical condition of the user or of an environment of the user;

user background information comprising data indicative of characteristics, preferences, or demographics of the user;

a model confidence score indicative of a likelihood that the sample response is accurate;

privacy preservation based on evaluating whether the sample response would reveal privacy information; or

a combination thereof;

select, as a best output, a sample response of the plurality of sample responses based, at least in part, on the ranking scores; and

providing a response to the user based on the best output.