Multidimensional local large language model user authentication

1 . A method of authenticating a user in a computerized system, comprising:

collecting user data from a plurality of dimensions of user activity into a local large language model residing entirely on a user device;

generating a plurality of user authentication questions from two or more of the plurality of dimensions of user activity using the local large language model, wherein generated questions having intersecting times or locations are discarded prior to presentation and wherein, for each authentication question at least one plausible distractor answer is generated by the local large language model;

presenting the plurality of user authentication questions from two or more dimensions to the user;

receiving a user response to each of the plurality of user authentication questions;

comparing the user response to each of the plurality of user authentication questions to a known correct answer to each of the plurality of user authentication questions; and

selectively authenticating the user based on the comparison between the user response and correct answer to each of the plurality of user authentication questions being correct.

2 . The method of authenticating a user in a computerized system of claim 1 , wherein collecting user data comprises collecting user activity via a smartphone.

3 . The method of authenticating a user in a computerized system of claim 1 , wherein the large language model comprises one or more of a neural network, a recurrent neural network, or a generative pre-trained transformer.

4 . The method of authenticating a user in a computerized system of claim 1 , wherein the plurality of dimensions comprise one or more of geographic location, purchase history, social media engagement, calendar events, electronic messaging, app usage, and media history.

5 . The method of authenticating a user in a computerized system of claim 1 , wherein the plurality of user authentication questions from two or more of the plurality of dimensions comprise authentication questions that do not intersect.

6 . The method of authenticating a user in a computerized system of claim 1 , wherein at least one of the plurality of user authentication questions comprises a multiple choice question and receiving a response comprises selecting the correct answer from among multiple presented answers.

7 . The method of authenticating a user in a computerized system of claim 1 , wherein receiving a response comprises receiving a typed text response.

8 . The method of authenticating a user in a computerized system of claim 1 , wherein the plurality of user authentication questions are used only once, and are not reused.

9 . The method of authenticating a user in a computerized system of claim 1 , wherein selectively authenticating the user comprises a part of at least one of user login, user two-factor authentication, user password change, or third-party authentication.

10 . The method of authenticating a user in a computerized system of claim 1 , wherein the large language model is trained using the plurality of dimensions of user activity.

11 . The method of authenticating a user in a computerized system of claim 1 , wherein at least one of: generating a plurality of user authentication questions, presenting the plurality of user authentication questions to the user, receiving a user response to each of the plurality of user authentication questions, comparing the user response to each of the plurality of user authentication questions to a known correct answer, and selectively authenticating the user, are performed in a server remote to the user.

12 . A computing device, comprising:

a processor and a non-volatile storage, the non-volatile storage comprising instructions that when executed on the computing device cause the computing device to:

collect user data from a plurality of dimensions of user activity into a local large language model contained entirely on a user device;

generate a plurality of user authentication questions from two or more of the plurality of dimensions of user activity using the local large language model, wherein generated questions having intersecting times or locations are discarded prior to presentation and wherein, for each authentication question at least one plausible distractor answer is generated by the local large language model;

present the plurality of user authentication questions from two or more dimensions to a user;

receive a user response to each of the plurality of user authentication questions;

compare the user response to each of the plurality of user authentication questions to a known correct answer to each of the plurality of user authentication questions; and

selectively authenticate the user based on the comparison between the user response and correct answer to each of the plurality of user authentication questions being correct.

13 . The computing device of claim 12 , wherein collecting user data comprises collecting user activity via a smartphone.

14 . The computing device of claim 12 , wherein the large language model comprises one or more of a neural network, a recurrent neural network, or a generative pre-trained transformer.

15 . The computing device of claim 12 , wherein the plurality of dimensions comprise one or more of geographic location, purchase history, social media engagement, calendar events, electronic messaging, app usage, and media history.

16 . The computing device of claim 12 , wherein the plurality of user authentication questions from two or more of the plurality of dimensions comprise authentication questions that do not intersect.

17 . The computing device of claim 12 , wherein receiving a response comprises at least one of selecting the correct answer from among multiple presented answers and receiving a typed text response.

18 . The computing device of claim 12 , wherein the large language model is trained using the plurality of dimensions of user activity.

19 . The computing device of claim 12 , wherein at least one of: generating a plurality of user authentication questions, presenting the plurality of user authentication questions to the user, receiving a user response to each of the plurality of user authentication questions, comparing the user response to each of the plurality of user authentication questions to a known correct answer, and selectively authenticating the user, are performed in a server remote to the user.

20 . A method of authenticating a user in a computerized server, comprising:

collecting user data from a user device representing a plurality of dimensions of user activity into a large language model residing entirely on a user device;

generating a plurality of user authentication questions from two or more of the plurality of dimensions of user activity using the large language model, wherein generated questions having intersecting times or locations with are discarded prior to presentation and wherein, for each authentication question at least one plausible distractor answer is generated by the local large language model;

providing the plurality of user authentication questions from two or more dimensions to the user device for presentation to the user;

receiving a user response to each of the plurality of user authentication questions via the user device;

comparing the user response to each of the plurality of user authentication questions to a known correct answer to each of the plurality of user authentication questions in at least one of the computerized server and the user device; and

selectively authenticating the user based on the comparison between the user response and correct answer to each of the plurality of user authentication questions being correct.