DETAILED ACTION
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 .
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.
Claim(s) 1-20 are rejected under 35 U.S.C. 103 as being unpatentable over Carley US 2022/0067181 in view of Berglund et al US 2024/0403341(reference provisional application # 63,505408).
As per claim 1. Carley discloses a method for secure deployment of a Large Language Model (LLM), (0005 methods and systems for secure machine learning are disclosed. The system includes a processor and a non-transitory computer readable medium for storing instructions that when executed by the processor cause the system to execute the methods of this disclosure)comprising:
receiving a training dataset for the LLM that includes confidential data (fig.2, 0007 the labeled dataset, i.e. a training dataset, may be created by: receiving a corpus of unlabeled data, ,0020 (1) labeled data is kept confidential and parameters and artifacts of a machine learning models are kept confidential wherein the labeled data is the confidential data [0051] At 206, the system may receive from a user of a labeling device, labeled data units including labels suitably associated with the assigned data units), wherein the LLM comprises preset parameters ( 0008 training data from the labeled dataset based on one or more of random sampling, i.e. parameters);
training the LLM using the training dataset (0008 training data from the labeled dataset based on one or more of random sampling), including:
identifying one or more parameters changed during the training ( 0042 the training device 106a-n, can modify the machine learning model being trained to include a new or modified algorithm, new or modified hyperparameter(s), i.e. parameters changed, fine-tune a model, perform bug fixes, revise various parameters, i.e. parameters changed,), and encrypting the changed parameters( 0007 encrypting the labeled data units; encryption mechanisms to be used for encrypting the labeled data units and [0044] In certain embodiments, the integration module 126 may also include a model encryptor 126(c) configured to encrypt trained (fully or partially trained) machine learning models before storage in a model repository 140. );
receiving an input query for an LLM form a user (par 0018 a user search query and 0064 allows for receipt of data from input devices 45 and 0061 the secure machine learning pipeline discussed above prevents malicious attacks that may tamper with the integrity of data used for training the model and/or the model because they only allow limited access to data such as labeled data and evaluation metrics as well provide encrypted model repository with checkpoints);
determining the user has access rights to the confidential data(0007 determining an access level associated with a labeling device and 0008 types of data in the labeled dataset and 0044 the system encapsulates the trained encrypted model progress (e.g., parameters, etc.) for providing to a user requiring access for, for example, further training or updating of a stored model to prevent such users from having direct access to a trained model and 0054 [0054] At 304, the system may receive encrypted labeled data and decrypt the data. At 306, the system may identify a plurality of clusters (i.e., subsets) of the decrypted data for use in training of the machine learning model).
Carley does not disclose
receiving an input query for an LLM form a user;
determining if the user has access rights to the data;
in response to determining that the user has the access rights to the data, decode the encoded data of the LLM, and performing an LLM inference using the decode the data; and
in response to determining that the user does not have the access rights to the data, performing the LLM inference with the preset parameters without decode the encode data.
Berglund discloses
receiving an input query for an LLM form a user (0013 When a user submits a search query);
determining if the user has access rights to the data (0013 the search platform identifies relevant content and leverages the relevant content as a knowledge base or context for an LLM to generate an answer to the search query );
in response to determining that the user has the access rights to the data, decode the encoded data of the LLM, and performing an LLM inference using the decode the data( 0013/0014 A semantic search can be performed to identify content that are relevant to the submitted queries. The queries and the relevant content items are sent to the LLM to formulate an answer to the queries based on the content items. The output returned in response to the query can therefore include a portion of text generated by the LLM (such as a bulleted list of key takeaways from a product overview deck), in addition to, or instead of, a list of semantically matched content items ); and
in response to determining that the user does not have the access rights to the data, performing the LLM inference with the preset parameters without decode the encode data(0043 the search engine honors access rights of content in the content repository when executing a search. For example, if a particular user does not have access to a particular content item, the search engine does not return the content item to the user as a search result. Similarly, a content item to which a user does not have access may be excluded from a list of content sent to the LLM for generating an answer to the user's queries. However, in other cases, depending on access rights associated with each content item, a content item may be used as the knowledge base for the LLM to generate an answer even if the access rights of the content item do not permit the content item itself to be accessed by the user).
Carley and Berglund are both considered to be analogous to the claimed invention because they are in the same field of machine learning for dataset.
Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified Carley to incorporate the teachings of Berglund and provide access rights to obtain data from the LLM.
Doing so would provide protection for the content of the LLM, thereby increasing the protection of the content from the large Language model.
As per claim 2. Carley and Berglund disclose the method of claim 1, Carley discloses wherein encrypting the changed parameters includes encrypting at least one layer comprising the changed parameters ([0029] As shown in FIG. 1, the system also includes a secure machine learning platform 120 that receives encrypted labeled data 114 from the data store 152, and enables training devices 106a-n to access the labeled data 114, as described below. The machine learning platform 120 may include a data module 122, a source code repository 124, and an integration module 126 ).
As per claim 3. Carley and Berglund disclose the method of claim 1, Carley discloses wherein encrypting the changed parameters comprises encrypting a difference between a first state of the changed parameters prior to the training and a second state of the changed parameters after the training (0042 the training device 106a-n, can modify the machine learning model being trained to include a new or modified algorithm, new or modified hyperparameter(s), i.e. parameters changed, fine-tune a model, perform bug fixes, revise various parameters, i.e. parameters changed ).
As per claim 4. Carley and Berglund disclose the method of claim 3, Carley discloses wherein performing the LLM inference using the decrypted changed parameters comprises decrypting the difference and applying the decrypted difference to the first state of the changed parameters to determine the second state of the changed parameters ([0030] The data module 122 may receive encrypted labeled data and decrypt it using a suitable protocol (e.g., using a decryption key). The data module 122 may also divide the received training data into subsets such as, without limitation, training data, test data, and validation data for use in training a machine learning model, testing a trained model, and validating a trained model, ).
As per claim 5. Carley and Berglund disclose the method of claim 3, Carley discloses wherein the first state of the changed parameters is the preset parameters ( 0042 the training device 106a-n, can modify the machine learning model being trained to include a new or modified algorithm, new or modified hyperparameter(s), i.e. parameters changed, fine-tune a model, perform bug fixes, revise various parameters, i.e. parameters changed , 0007 encrypting the labeled data units; encryption mechanisms to be used for encrypting the labeled data units and [0044] In certain embodiments, the integration module 126 may also include a model encryptor 126(c) configured to encrypt trained (fully or partially trained) machine learning models before storage in a model repository 140. ).
As per claim 6. Carley and Berglund disclose The method of claim 3, Carley discloses further comprising: determining if a value of the difference between the changed parameters and prior parameters is less than a threshold amount, and in response to determining that the value of the difference is less than the threshold amount, reverting the changed parameters such that the changed parameters return to a state prior to the training with the training dataset, and not encrypting the changed parameters( 0042] The evaluation metrics generated by the evaluator may be transmitted back to a training device 106a-n or developer, via a metrics module 126(b) and may be used to determine whether to adjust the training process and/or to stop the training process. For example, the evaluation metrics can indicate that the machine learning model is performing poorly (e.g., has an error rate above a threshold value, has a statistical distribution that is not an expected or desired distribution (e.g., not a binomial distribution, a Poisson distribution, a geometric distribution, a normal distribution, Gaussian distribution, etc.), has an execution latency above a threshold value, has a confidence level below a threshold value)) and/or is performing progressively worse (e.g., the quality metric continues to worsen over time). In response, in some embodiments, the user, via the training device 106a-n, can modify the machine learning model being trained to include a new or modified algorithm, new or modified hyperparameter(s), fine-tune a model, perform bug fixes, revise various parameters, or the like. In certain embodiments, the metrics module 126(b) may only transmit a subset of the evaluation metrics to a training device in order to improve confidentiality of the training process and/or a trained model. For example, the metrics module 126(b) may only transmit evaluation metrics for a version that was trained by a particular user and/or using a particular training device. Optionally, the metrics module 126(b) may only transmit evaluation metrics for a portions of a model do not meet certain threshold requirements and that portion needs to be updated or retrained.).
As per claim 7. Carley and Berglund disclose The method of claim 1, Carley discloses wherein the changed parameters comprise weights and/or biases(0037a machine learning model, trained machine models may be updated regularly as new data comes in or utilizing a different range of data, and/or a trained machine learning model may be updated for improving convergence. As such, there will be multiple trained models created, each having differing predictive behaviors. These trained models may be saved in a source code repository 124. A saved trained model may be file(s) that include model data or model artifacts such as structure of a machine learning algorithm used for training, hyperparameters, learned parameters, coefficients, weights/coefficients for a particular ML algorithm (e.g., weights for a neural network), values serving as centroids for a model (e.g., a k-means clustering model), a shape of an equation, or other model-specific information allowing a model to be utilized (e.g., indicators of where tree splits exists for tree-based models, etc.). or the like. It should be noted that partially or fully trained models may be saved in the source code repository incrementally during the training phase and may be associated with a checkpoint when saved. Having regular checkpoints enables the training infrastructure/pipeline to be restarted without retraining on all of the data. It should be noted that trained model is ready for use in various applications (e.g., by an autonomous vehicle) when sufficient training has been made (e.g. convergence), and prior unusable versions of a model may also be saved in the source code repository).
As per clam 8. Carley and Berglund disclose the method of claim 1, Carley discloses wherein the LLM is a 1-bit large language model (LLM)(0044 a model encryptor 126(c) configured to encrypt trained (fully or partially trained) machine learning models before storage in a model repository 140. The encryption may be performed using any now or hereafter known encryption protocols such as, without limitation, block cipher technologies (e.g., advanced encryption standard “AES”, AES-256, data encryption standard “DES”, skipjack, etc.), post quantum cryptography, random bit generation,).
As per claim 9. Carley and Berglund disclose The method of claim 1, Carley discloses wherein the preset parameters are encrypted by a general encryption scheme(0005 The methods may further include receiving a trained machine learning model from the training device, evaluating the trained machine learning to determine whether the trained machine learning model satisfies an evaluation criterion, and encrypting the trained machine learning model if the trained machine learning model satisfies the evaluation criterion. The encrypted machine learning model may be deployed to make predictions using live data. Optionally, the encrypted machine learning may be stored).
As per claim 10. Carley and Berglund disclose The method of claim 1, Carley discloses wherein the user has the access rights to the confidential data when the user possesses a private encryption key for decrypting the encrypted parameters (0007 determining an access level associated with a labeling device and 0008 types of data in the labeled dataset and 0044 the system encapsulates the trained encrypted model progress (e.g., parameters, etc.) for providing to a user requiring access for, for example, further training or updating of a stored model to prevent such users from having direct access to a trained model and 0054 [0054] At 304, the system may receive encrypted labeled data and decrypt the data. At 306, the system may identify a plurality of clusters (i.e., subsets) of the decrypted data for use in training of the machine learning model).
As per claim 11. Carley and Berglund disclose The method of claim 10, Carley discloses wherein a first output value without private information is generated by the LLM when a user input query for the LLM inference is not provided with the private encryption key, and a second output value comprising the private information is generated by the LLM when the user input query is provided with the private encryption key( 0054] At 304, the system may receive encrypted labeled data and decrypt the data. At 306, the system may identify a plurality of clusters (i.e., subsets) of the decrypted data for use in training of the machine learning model. The clusters may be identified based on, for example, random sampling, similarities between data units of a cluster, required characteristics of data for training a machine learning model at a training device (e.g., training a machine learning model to perform ground segmentation may require training data that includes a first number of ground images and a second number of non-ground images), required number of data units for training a machine learning model at a training device, access levels associated with a user requesting the training data (e.g., certain training devices/users may have access to confidential information such as health information while other training devices/users may not), access levels associated with users/training devices to which the clusters will be transmitted for use, history of data already transmitted to users/training devices to which the clusters will be transmitted, types of data in the training subset (e.g., 3D point clouds may be clustered based on the LIDAR sensor used to capture the point cloud data, resolution, etc.), type of information included in the training data (e.g., financial information, health information, etc.), or the like. It should be noted that the clusters may be identified for each training device/user separately based on their respective properties or conditions.).
As per claim 12. Carley and Berglund disclose The method of claim 11, Carley discloses wherein the user is provided with one or more encryption keys based on a level of access to the confidential data such that all of the one or more encryption keys are needed to access all of the confidential data (0030] The data module 122 may receive encrypted labeled data and decrypt it using a suitable protocol (e.g., using a decryption key). The data module 122 may also divide the received training data into subsets such as, without limitation, training data, test data, and validation data for use in training a machine learning model, testing a trained model, and validating a trained model, 0045] Optionally, certain users with the requisite access defined by, for example, organizational policies, may be granted access to certain type of decrypted models and/or decryption keys for certain types of trained models ).
As per claim 13. Carley and Berglund disclose The method of claim 1, Carley discloses wherein associated parameters of each of one or more layers of the LLM is encrypted by a different encryption key(0027a data encryption module 130 may encrypt each labeled data unit generated by the labeling devices 104a-n before storage as labeled data 114. Encryption may be performed using any now or hereafter known methods or protocols (e.g., symmetric key encryption, asymmetric key encryption methods, etc.). Optionally, each labeled data unit may be encrypted using a unique encryption key/key pair, protocol, etc. to further enhance privacy protection of labeled data 114 ).
As per claim 14. Carley discloses A system for secure deployment of a Large Language Model (LLM), comprising: at least one memory; and at least one hardware processor coupled with the at least one memory and configured, individually or in combination (0005 methods and systems for secure machine learning are disclosed. The system includes a processor and a non-transitory computer readable medium for storing instructions that when executed by the processor cause the system to execute the methods of this disclosure), to:
receiving a training dataset for the LLM that includes confidential data (fig.2, 0007 the labeled dataset, i.e. a training dataset, may be created by: receiving a corpus of unlabeled data, ,0020 (1) labeled data is kept confidential and parameters and artifacts of a machine learning models are kept confidential wherein the labeled data is the confidential data [0051] At 206, the system may receive from a user of a labeling device, labeled data units including labels suitably associated with the assigned data units), wherein the LLM comprises preset parameters ( 0008 training data from the labeled dataset based on one or more of random sampling, i.e. parameters);
training the LLM using the training dataset (0008 training data from the labeled dataset based on one or more of random sampling), including:
identifying one or more parameters changed during the training ( 0042 the training device 106a-n, can modify the machine learning model being trained to include a new or modified algorithm, new or modified hyperparameter(s), i.e. parameters changed, fine-tune a model, perform bug fixes, revise various parameters, i.e. parameters changed,), and encrypting the changed parameters( 0007 encrypting the labeled data units; encryption mechanisms to be used for encrypting the labeled data units and [0044] In certain embodiments, the integration module 126 may also include a model encryptor 126(c) configured to encrypt trained (fully or partially trained) machine learning models before storage in a model repository 140. );
receiving an input query for an LLM form a user (par 0018 a user search query and 0064 allows for receipt of data from input devices 45 and 0061 the secure machine learning pipeline discussed above prevents malicious attacks that may tamper with the integrity of data used for training the model and/or the model because they only allow limited access to data such as labeled data and evaluation metrics as well provide encrypted model repository with checkpoints);
determining the user has access rights to the confidential data(0007 determining an access level associated with a labeling device and 0008 types of data in the labeled dataset and 0044 the system encapsulates the trained encrypted model progress (e.g., parameters, etc.) for providing to a user requiring access for, for example, further training or updating of a stored model to prevent such users from having direct access to a trained model and 0054 [0054] At 304, the system may receive encrypted labeled data and decrypt the data. At 306, the system may identify a plurality of clusters (i.e., subsets) of the decrypted data for use in training of the machine learning model).
Carley does not disclose
receiving an input query for an LLM form a user;
determining if the user has access rights to the data;
in response to determining that the user has the access rights to the data, decode the encoded data of the LLM, and performing an LLM inference using the decode the data; and
in response to determining that the user does not have the access rights to the data, performing the LLM inference with the preset parameters without decode the encode data.
Berglund discloses
receiving an input query for an LLM form a user (0013 When a user submits a search query);
determining if the user has access rights to the data (0013 the search platform identifies relevant content and leverages the relevant content as a knowledge base or context for an LLM to generate an answer to the search query );
in response to determining that the user has the access rights to the data, decode the encoded data of the LLM, and performing an LLM inference using the decode the data( 0013/0014 A semantic search can be performed to identify content that are relevant to the submitted queries. The queries and the relevant content items are sent to the LLM to formulate an answer to the queries based on the content items. The output returned in response to the query can therefore include a portion of text, i.e. LLM inference, generated by the LLM (such as a bulleted list of key takeaways from a product overview deck), in addition to, or instead of, a list of semantically matched content items ); and
in response to determining that the user does not have the access rights to the data, performing the LLM inference with the preset parameters without decode the encode data(0043 the search engine honors access rights of content in the content repository when executing a search. For example, if a particular user does not have access to a particular content item, the search engine does not return the content item to the user as a search result. Similarly, a content item to which a user does not have access may be excluded from a list of content sent to the LLM for generating an answer to the user's queries. However, in other cases, depending on access rights associated with each content item, a content item may be used as the knowledge base for the LLM to generate an answer even if the access rights of the content item do not permit the content item itself to be accessed by the user).
Carley and Berglund are both considered to be analogous to the claimed invention because they are in the same field of machine learning for dataset.
Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified Carley to incorporate the teachings of Berglund and provide access rights to obtain data from the LLM.
Doing so would provide protection for the content of the LLM, thereby increasing the protection of the content from the large Language model.
As per claim 15. Carley and Berglund disclose the system of claim 14,Carley discloses wherein the at least one hardware processor is configured to encrypt the changed parameters by encrypting at least one layer comprising the changed parameters(([0029] As shown in FIG. 1, the system also includes a secure machine learning platform 120 that receives encrypted labeled data 114 from the data store 152, and enables training devices 106a-n to access the labeled data 114, as described below. The machine learning platform 120 may include a data module 122, a source code repository 124, and an integration module 126).
As per claim 16. Carley and Berglund disclose the system of claim 14, Carley discloses wherein the at least one hardware processor is configured to encrypt the changed parameters by encrypting a difference between a first state of the changed parameters prior to the training and a second state of the changed parameters after the training (0042 the training device 106a-n, can modify the machine learning model being trained to include a new or modified algorithm, new or modified hyperparameter(s), i.e. parameters changed, fine-tune a model, perform bug fixes, revise various parameters, i.e. parameters changed ).
As per claim 17. Carley and Berglund disclose the system of claim 16, Carley discloses wherein the at least one hardware processor is configured to perform the LLM inference using the decrypted changed parameters by decrypting the difference and applying the decrypted difference to the first state of the changed parameters to determine the second state of the changed parameters ( 0030] The data module 122 may receive encrypted labeled data i.e. changed and decrypt , i.e. changed, it using a suitable protocol (e.g., using a decryption key). The data module 122 may also divide the received training data into subsets such as, without limitation, training data, test data, and validation data for use in training a machine learning model, testing a trained model, and validating a trained model).
As per claim 18. Carley and Berglund disclose the system of claim 16, Carley discloses wherein the first state of the changed parameters is the preset parameters (0042 the training device 106a-n, can modify the machine learning model being trained to include a new or modified algorithm, new or modified hyperparameter(s), i.e. parameters changed, fine-tune a model, perform bug fixes, revise various parameters, i.e. parameters changed , 0007 encrypting the labeled data units; encryption mechanisms to be used for encrypting the labeled data units and [0044] In certain embodiments, the integration module 126 may also include a model encryptor 126(c) configured to encrypt trained (fully or partially trained) machine learning models before storage in a model repository 140. )).
As per claim 19. Carley and Berglund disclose The system of claim 16, Carley discloses wherein the at least one hardware processor is configured to: determine if a value of the difference between the changed parameters and prior parameters is less than a threshold amount, and in response to determining that the value of the difference is less than the threshold amount, revert the changed parameters such that the changed parameters return to a state prior to the training with the training dataset, and not encrypt the changed parameters (0042] The evaluation metrics generated by the evaluator may be transmitted back to a training device 106a-n or developer, via a metrics module 126(b) and may be used to determine whether to adjust the training process and/or to stop the training process. For example, the evaluation metrics can indicate that the machine learning model is performing poorly (e.g., has an error rate above a threshold value, has a statistical distribution that is not an expected or desired distribution (e.g., not a binomial distribution, a Poisson distribution, a geometric distribution, a normal distribution, Gaussian distribution, etc.), has an execution latency above a threshold value, has a confidence level below a threshold value)) and/or is performing progressively worse (e.g., the quality metric continues to worsen over time). In response, in some embodiments, the user, via the training device 106a-n, can modify the machine learning model being trained to include a new or modified algorithm, new or modified hyperparameter(s), fine-tune a model, perform bug fixes, revise various parameters, or the like. In certain embodiments, the metrics module 126(b) may only transmit a subset of the evaluation metrics to a training device in order to improve confidentiality of the training process and/or a trained model. For example, the metrics module 126(b) may only transmit evaluation metrics for a version that was trained by a particular user and/or using a particular training device. Optionally, the metrics module 126(b) may only transmit evaluation metrics for a portions of a model do not meet certain threshold requirements and that portion needs to be updated or retrained).
As per claim 20. Carley discloses A non-transitory computer readable medium storing thereon computer executable instructions for secure deployment of a Large Language Model (LLM)(0005 methods and systems for secure machine learning are disclosed. The system includes a processor and a non-transitory computer readable medium for storing instructions that when executed by the processor ), including instructions for: receiving a training dataset for the LLM that includes confidential data (fig.2, 0007 the labeled dataset, i.e. a training dataset, may be created by: receiving a corpus of unlabeled data, ,0020 (1) labeled data is kept confidential and parameters and artifacts of a machine learning models are kept confidential wherein the labeled data is the confidential data [0051] At 206, the system may receive from a user of a labeling device, labeled data units including labels suitably associated with the assigned data units), wherein the LLM comprises preset parameters ( 0008 training data from the labeled dataset based on one or more of random sampling, i.e. parameters);
training the LLM using the training dataset (0008 training data from the labeled dataset based on one or more of random sampling), including:
identifying one or more parameters changed during the training ( 0042 the training device 106a-n, can modify the machine learning model being trained to include a new or modified algorithm, new or modified hyperparameter(s), i.e. parameters changed, fine-tune a model, perform bug fixes, revise various parameters, i.e. parameters changed,), and encrypting the changed parameters( 0007 encrypting the labeled data units; encryption mechanisms to be used for encrypting the labeled data units and [0044] In certain embodiments, the integration module 126 may also include a model encryptor 126(c) configured to encrypt trained (fully or partially trained) machine learning models before storage in a model repository 140. );
receiving an input query for an LLM form a user (par 0018 a user search query and 0064 allows for receipt of data from input devices 45 and 0061 the secure machine learning pipeline discussed above prevents malicious attacks that may tamper with the integrity of data used for training the model and/or the model because they only allow limited access to data such as labeled data and evaluation metrics as well provide encrypted model repository with checkpoints);
determining the user has access rights to the confidential data(0007 determining an access level associated with a labeling device and 0008 types of data in the labeled dataset and 0044 the system encapsulates the trained encrypted model progress (e.g., parameters, etc.) for providing to a user requiring access for, for example, further training or updating of a stored model to prevent such users from having direct access to a trained model and 0054 [0054] At 304, the system may receive encrypted labeled data and decrypt the data. At 306, the system may identify a plurality of clusters (i.e., subsets) of the decrypted data for use in training of the machine learning model).
Carley does not disclose
receiving an input query for an LLM form a user;
determining if the user has access rights to the data;
in response to determining that the user has the access rights to the data, decode the encoded data of the LLM, and performing an LLM inference using the decode the data; and
in response to determining that the user does not have the access rights to the data, performing the LLM inference with the preset parameters without decode the encode data.
Berglund discloses
receiving an input query for an LLM form a user (0013 When a user submits a search query);
determining if the user has access rights to the data (0013 the search platform identifies relevant content and leverages the relevant content as a knowledge base or context for an LLM to generate an answer to the search query );
in response to determining that the user has the access rights to the data, decode the encoded data of the LLM, and performing an LLM inference using the decode the data( 0013/0014 A semantic search can be performed to identify content that are relevant to the submitted queries. The queries and the relevant content items are sent to the LLM to formulate an answer to the queries based on the content items. The output returned in response to the query can therefore include a portion of text generated by the LLM (such as a bulleted list of key takeaways from a product overview deck), in addition to, or instead of, a list of semantically matched content items ); and
in response to determining that the user does not have the access rights to the data, performing the LLM inference with the preset parameters without decode the encode data(0043 the search engine honors access rights of content in the content repository when executing a search. For example, if a particular user does not have access to a particular content item, the search engine does not return the content item to the user as a search result. Similarly, a content item to which a user does not have access may be excluded from a list of content sent to the LLM for generating an answer to the user's queries. However, in other cases, depending on access rights associated with each content item, a content item may be used as the knowledge base for the LLM to generate an answer even if the access rights of the content item do not permit the content item itself to be accessed by the user).
Carley and Berglund are both considered to be analogous to the claimed invention because they are in the same field of machine learning for dataset.
Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified Carley to incorporate the teachings of Berglund and provide access rights to obtain data from the LLM.
Doing so would provide protection for the content of the LLM, thereby increasing the protection of the content from the large Language model.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to ABU S SHOLEMAN whose telephone number is (571)270-7314. The examiner can normally be reached EST: 9am-5pm.
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, JORGE ORTIZ CRIADO can be reached at 571-272-7624. 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.
/ABU S SHOLEMAN/Primary Examiner, Art Unit 2496
Prosecution Insights