DETAILED ACTION
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
This action is in reply to the communications filed on January 2, 2026. The Applicant’s Amendment and Request for Reconsideration has been received and entered.
Claims 1-20 are currently pending and have been examined. Claims 1, 9, and 17 have been amended.
Response to Arguments
Applicant’s amendments necessitated the new grounds of rejection.
The objection to the drawings has been withdrawn in view of Applicant’s amendments.
Regarding the rejection of claims 1-20 under 35 USC 101, Applicant’s arguments have been fully considered but they are not persuasive for the reasons set forth infra.
Additionally, the Examiner respectfully argues that improving the accuracy of spend classifications is not necessarily a technical improvement, but rather, improves a business’s ability to perform business functions using a computer. Indeed, per Applicant’s specification, spending classification accuracy can affect “managers [ability] to monitor trends, identify sourcing opportunities and negotiate more effectively with their supply base” (App. Spec. [0007]). Additionally, spend management activities “improve purchasing costs and lower supply base risk through the efficient use of a limited set of resources.” (App. Spec. [0002]). The claims do not improve technology merely because technology is invoked as a tool to improve business functions.
Applicant’s remaining arguments have been fully considered but they are not persuasive. Particularly, Applicant’s arguments are directed to the instantly amended claims, and are thus moot in view of the new grounds of rejection. Additionally, the Examiner respectfully notes that as per Applicant’s specification, a hallucination is when output returned “does not actually exist (i.e., hallucinated)” (App. Spec. [0045]). As Fuxman teaches “the output of the text generator, in some implementations, may exceed the output taxonomy of the image classifier, e.g., the text generator may generate phrases that are not included in the output taxonomy of the multimodal image classifier [emphasis added].” ([0015])—Fuxman therefore teaches hallucinations.
The Examiner invites Applicant’s Representative to contact Examiner for a telephonic interview to further prosecution of this application.
Claim Rejections - 35 USC § 101
35 U.S.C. 101 reads as follows:
Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title.
Claims 1-20 are rejected under 35 U.S.C. 101 because the claimed invention is directed to non-statutory subject matter.
Step 1. When considering subject matter eligibility under 35 U.S.C. 101, it must be determined whether the claim is directed to one of the four statutory categories of invention, i.e., process, machine, manufacture, or composition of matter.
Step 2A – Prong One. If the claims fall within one of the statutory categories, it must then be determined whether the claims recite an abstract idea, law of nature, or natural phenomenon.
Step 2A – Prong Two. If the claims recite an abstract idea, law of nature, or natural phenomenon, it must then be determined whether the claims recite additional elements that integrate the judicial exception into a practical application. If the claims do not recite additional elements that integrate the judicial exception into a practical application, then the claims are directed to a judicial exception.
Step 2B. If the claims are directed to a judicial exception, it must be evaluated whether the claims recite additional elements that amount to an inventive concept (i.e. “significantly more”) than the recited judicial exception.
In the instant case, claims 1-8 are directed to a process; claims 9-16 are directed to a purported manufacture; and claims 17-20 are directed to a machine.
A claim “recites” an abstract idea if there are identifiable limitations that fall within at least one of the groupings of abstract ideas enumerated in MPEP 2106. In the instant case, claim 1, and similarly claims 9 and 17, recites the steps of:
receiving a description of the product; creating a first prompt for a trained large language model (LLM), the first prompt comprising the description of the product and contextual information of the product; in response to the first prompt, using the trained LLM to generate a hallucinated product classification for the product by querying the trained LLM using the first prompt, wherein the hallucinated product classification is not one of the plurality of possible product classifications; word embedding the hallucinated product classification and the plurality of possible product classifications; and similarity matching the embedded hallucinated product classification with one of the embedded plurality of product classifications, wherein the matched one of the embedded plurality of product classifications is determined to be a predicted classification of the product, wherein the predicted classification of the product is one of the plurality of possible product classifications -- these claim limitations set forth certain methods of organizing human activity, particularly commercial interactions including advertising, marketing, and sales activities/behaviors.
Additionally, these steps set forth mental processes, particularly concepts performed in the human mind or by a human using a pen and paper, including, inter alia, the observation and evaluation of information.
Further, the limitations of the claims are not indicative of integration into a practical application. Taking the independent claim elements separately, the additional elements of: receiving a description of the product; creating a first prompt for a trained large language model (LLM), the first prompt comprising the description of the product and contextual information of the product; in response to the first prompt, using the trained LLM to generate a hallucinated product classification for the product by querying the trained LLM using the first prompt, wherein the hallucinated product classification is not one of the plurality of possible product classifications ; word embedding the hallucinated product classification and the plurality of product classifications; and similarity matching the embedded hallucinated product classification with one of the embedded plurality of possible product classifications, wherein the matched one of the embedded plurality of product classifications is determined to be a predicted classification of the product, wherein the predicted classification of the product is one of the plurality of possible product classifications -- merely implement the abstract idea on a computer environment.
Additionally, taking the dependent claim elements separately, the additional elements of performing the steps via a database, and using a second trained LLM, also merely implement the abstract idea on a computer environment. Considered in combination, the steps of Applicant’s method add nothing that is not already present when the steps are considered separately.
Thus, claims 1-20 are directed to an abstract idea.
Regarding the independent claims, the technical elements of performing the steps using the trained LLM to generate a hallucinated classification is recited at a high level of generality and thus does not amount to significantly more. Additionally, regarding the dependent claims, the technical element of performing the steps via a database merely implements the abstract idea on a computer environment, and the technical element of performing the steps using a second trained LLM is recited at a high level of generality and thus does not amount to significantly more.
When considering the elements and combinations of elements, the claim(s) as a whole, do not amount to significantly more than the abstract idea itself. This is because the claims do not amount to an improvement to another technology or technical field; the claims do not amount to an improvement to the functioning of a computer itself; the claims do not move beyond a general link of the use of an abstract idea to a particular technological environment; the claims merely amounts to the application or instructions to apply the abstract idea on a computer; or the claims amounts to nothing more than requiring a generic computer to perform generic computer functions that are well-understood, routine and conventional activities previously known to the industry.
The analysis above applies to all statutory categories of invention. Accordingly, claims 1-20 are rejected as ineligible for patenting under 35 USC 101 based upon the same rationale.
Claims 10-16 are rejected under 35 U.S.C. 101 because the claimed invention is directed to non-statutory subject matter. The claim(s) does/do not fall within at least one of the four categories of patent eligible subject matter because the recitation of “computer readable medium” encompasses transitory forms of the media such as signals per se.
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
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 of this title, 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, 2, 5, 7-10, 13, 15, and 16 are rejected under 35 U.S.C. 103 as being unpatentable over Fuxman (US PGP 2021/0264203) in view of Github (https://web.archive.org/web/20230827120821/https://github.com/DivergentAI/dreamGPT, dated Aug 27, 2023).
As per claim 1, Fuxman teaches method of classifying a product to one of a plurality of product classifications, the method comprising:
receiving a description of the product; (Fuxman: Claim 7 (for each image of a plurality of images: processing the image by a textual generator model to obtain a set of phrases that are descriptive of the content of the image, wherein each phrase is one or more terms); [0018]; Fig. 2; [0039])
creating a first prompt for a trained . . . model . . . , the first prompt comprising the description of the product and contextual information of the product; (Fuxman: Claim 7 (processing the set of phrases by a textual embedding model to obtain an embedding of predicted text for the image; and); [0039] (The process 200, for each image in a set of images, (202), obtains a textual embedding of the image (204) and obtains a pixel embedding of the image (206).); [0019]-[0030];
in response to the first prompt, using the trained . . . to generate a hallucinated product classification for the product by querying . . . using the first prompt, wherein the hallucinated product classification is not one of the plurality of possible product classifications; (Fuxman: [0015] (This patent document describes a framework that includes a text generator that generates textual labels, referred to as phrases, from image data, and a multimodal image classifier that takes textual features from the text generator and visual features from the image pixels, and produces labels according to an output taxonomy. In some implementations, the text generator is trained using web-based query/image pairs to incorporate contextual information associated with each image. The output of the text generator, in some implementations, may exceed the output taxonomy of the image classifier, e.g., the text generator may generate phrases that are not included in the output taxonomy of the multimodal image classifier.); [0030]-[0036]; Claim 7; [0039]-[0040]); [0022]-[0034] (disclosing query-image pairs))
word embedding the hallucinated product classification and the plurality of possible product classifications; and (Fuxman: [0036] (The multimodal image classifier 110 receives the textual embeddings from the textual embedding neural network 122 and the pixel embeddings from the pixel embedding neural network 124. The classifier is trained on these embeddings to produce, as output, labels of an output taxonomy to classify an image based on the image as input. Having obtained pixel and textual embeddings for arbitrary input images, the embeddings are subsequently combined to generate representations for images that incorporate both types of features. In some implementation, the process for combining these features involved concatenating the 200-dimensional textual feature vector with 1024-dimensional visual feature vector into a singular feature vector representing both concepts. The resulting vector is subsequently input to a two-stage fully connected neural network, with the output softmax layer predicting probabilities associated with classes according to a specified taxonomy.))
similarity matching the embedded hallucinated product classification with one of the embedded plurality of product classifications, wherein the matched one of the embedded plurality of product classifications is determined to be a predicted classification of the product, wherein the predicted classification of the product is one of the plurality of possible product classifications. (Fuxman: Claim 6 (wherein the textual generator model obtain a set of phrases for a given image using of a nearest-neighbor process.); [0029]; [0023] (In this example architecture, at inference time, 64-dimensional embeddings are generated for each input image, and the top N most probable queries and their associated similarity scores are extracted from the query embedding index. In some implementations, N=3, but other values of N can also be used. As queries represent contextual information, this process identifies relevant text for arbitrary images without web dependence. Furthermore, the embeddings additionally yield a way to measure image to image similarity in query space.); [0031] (The textual embedding neural network 122 is an artificial neural network that is trained to map a discrete input (e.g., a feature vector of the phrases generated by the text generator 120) to a continuously valued output (e.g., a vector or matrix). The outputs of the textual embedding neural network 122 have the property that similar inputs are mapped to outputs that are close to one another in multi-dimensional space. The output of the textual embedding neural network 122 can thus be described as a latent representation of the data that is input to the embedding neural network.); [0035]-[0036]; [0040])
Fuxman does not explicitly disclose the following known technique which is taught by Github:
. . . a trained large language model (LLM) . . . (Github: Page 1 (dreamGPT, the first GPT-based solution that uses hallucinations from LLMs))
. . . using the trained LLM to generate a hallucinated . . . (Github: Pages 1-2 (Once you run it, dreamGPT generates a random seed of concepts and will use these as a starting point for its dreaming process. Here is a screenshot of the first iteration. . . . As dreamGPT evolves the dreams, you will start to see higher scores with even better ideas. . . . Generating concept . . . ))
This known technique is applicable to the method of Fuxman as they both share characteristics and capabilities, namely, they are directed to trained models.
One of ordinary skill in the art at the time of filing would have recognized that applying the known technique of Github would have yielded predictable results and resulted in an improved method. It would have been recognized that applying the technique of Github to the teachings of Fuxman would have yielded predictable results because the level of ordinary skill in the art demonstrated by the references applied shows the ability to incorporate such trained LLM to generate hallucination features into similar methods. Further, applying the using the trained LLM to generate a hallucinated [data] to the teachings of Fuxman would have been recognized by those of ordinary skill in the art as resulting in an improved method that would allow generation of new innovative output (Github: Page 1).
As per claim 2, Fuxman/Github teach wherein the word embedding comprises converting a word into a vector. (Fuxman: [0036] (The multimodal image classifier 110 receives the textual embeddings from the textual embedding neural network 122 and the pixel embeddings from the pixel embedding neural network 124. The classifier is trained on these embeddings to produce, as output, labels of an output taxonomy to classify an image based on the image as input. Having obtained pixel and textual embeddings for arbitrary input images, the embeddings are subsequently combined to generate representations for images that incorporate both types of features. In some implementation, the process for combining these features involved concatenating the 200-dimensional textual feature vector with 1024-dimensional visual feature vector into a singular feature vector representing both concepts. The resulting vector is subsequently input to a two-stage fully connected neural network, with the output softmax layer predicting probabilities associated with classes according to a specified taxonomy.); [0031]-[0036] (The embedding model is subsequently utilized to embed the top N text results obtained from the text generator 120. In some implementations, the text embedding are obtained via a bag-of-words model of unigrams and bigrams. The embeddings for the top N query (or text predictions) are subsequently averaged, resulting in an M-dimensional textual feature vector.))
As per claim 5, Fuxman/Github teach wherein the product classifications each comprise a family, a class and a category. (Fuxman: [0036] (The classifier is trained on these embeddings to produce, as output, labels of an output taxonomy to classify an image based on the image as input.))
As per claim 7, Fuxman/Github teach wherein the word embedding is executed using a plurality of different word embedding encodings, each different word embedding encoding stored in a corresponding separate vector store. (Fuxman: [0032]-[0036] (A Siamese network is an artificial neural network that uses the same weights while working in tandem on two different input vectors to compute comparable output vectors. . . . The multimodal image classifier 110 receives the textual embeddings from the textual embedding neural network 122 and the pixel embeddings from the pixel embedding neural network 124. The classifier is trained on these embeddings to produce, as output, labels of an output taxonomy to classify an image based on the image as input. Having obtained pixel and textual embeddings for arbitrary input images, the embeddings are subsequently combined to generate representations for images that incorporate both types of features. In some implementation, the process for combining these features involved concatenating the 200-dimensional textual feature vector with 1024-dimensional visual feature vector into a singular feature vector representing both concepts.))
As per claim 8, Fuxman/Github teach generating a second prompt for using the trained LLM to generate an industry name that corresponds to a name of a company that purchased the product; or generating a third prompt for using the trained LLM to generate an industry name that corresponds to a list of products the company has purchased. (Fuxman: Claim 7 (processing the set of phrases by a textual embedding model to obtain an embedding of predicted text for the image; and); [0039] (The process 200, for each image in a set of images, (202), obtains a textual embedding of the image (204) and obtains a pixel embedding of the image (206).); [0019]-[0030])
Examiner Note: The Examiner notes that the following underlined portions of the claim are statements of intended use: generating a second prompt for using the trained LLM to generate an industry name that corresponds to a name of a company that purchased the product; or generating a third prompt for using the trained LLM to generate an industry name that corresponds to a list of products the company has purchased, as they merely state the intended purposes of generating the prompts. Intended use language does not result in a manipulative difference between the claimed invention and the prior art. See, e.g., In re Otto, 312 F.2d 937, 938, 136 USPQ 458, 459 (CCPA 1963) (The claims were directed to a core member for hair curlers and a process of making a core member for hair curlers. Court held that the intended use of hair curling was of no significance to the structure and process of making.)
As per claims 9, 10, 13, 15, and 16, these claims are substantially similar to claims 1, 2, 5, 7, and 8, respectively, and are therefore rejected in the same manner as these claims, as set forth above.
Claims 3, 4, 11, 12, 17-20 are rejected under 35 U.S.C. 103 as being unpatentable over Fuxman/Github in view of Garera (US PGP 2014/0214845).
As per claim 3, Fuxman/Github teach the invention of claim 1 as set forth above. Fuxman/Github do not explicitly disclose the following known technique which is taught by Garera:
wherein the description of the product is extracted from a product database. (Garera: [0032] (The database 204 b may store information regarding various products. In particular, information for a product may include a name, description, categorization, reviews, comments, price, past transaction data, and the like. The server 202 b may analyze this data as well as data retrieved from the database 204 a in order to perform methods as described herein.); [0036])
This known technique is applicable to the method of Fuxman/Github as they share characteristics and capabilities, namely, they are directed to receiving product description information.
One of ordinary skill in the art at the time of filing would have recognized that applying the known technique of Garera would have yielded predictable results and resulted in an improved method. It would have been recognized that applying the technique of Garera to the teachings of Fuxman/Github would have yielded predictable results because the level of ordinary skill in the art demonstrated by the references applied shows the ability to incorporate such database features into similar methods. Further, applying the description of the product is extracted from a product database to the description of Fuxman/Github would have been recognized by those of ordinary skill in the art as resulting in an improved method that would allow a more efficient and cost effective way for merchants to keep product databases up to date with new product offerings. (Garera: Para [0001]-[0002]).
As per claim 4, Fuxman/Github/Garera teach the invention of claim 3 as set forth above. Additionally, Fuxman/Github/Garera teach wherein the description of the product comprises one or more of price, enterprise details, department purchasing the product, manufacturer name or country of manufacture. (Fuxman: Claim 7 (for each image of a plurality of images: processing the image by a textual generator model to obtain a set of phrases that are descriptive of the content of the image, wherein each phrase is one or more terms); [0018] (The text generator 120 generates, for each image of images 102, phrases of one or more terms that are descriptive of content of the image.); also see Garera: [0032] (The database 204 b may store information regarding various products. In particular, information for a product may include a name, description, categorization, reviews, comments, price, past transaction data, and the like.); [0036])
Examiner Note: The Examiner notes that the particular information included in the description of the product is merely nonfunctional descriptive material and is not functionally involved in the steps recited. The receiving of the description of the product and the creating of a first prompt comprising the description would be performed the same regardless of the particular information included in it. This descriptive material will not distinguish the claimed invention from the prior art in terms of patentability, see In re Gulack, 70 F.2d 1381, 1385, 217 USPQ 401 (Fed. Cir. 1983); In re Lowry, 32 F.3d 1579, 32 USPQ2d 1031 (Fed. Cir. 1994).
As per claims 11 and 12, these claims are substantially similar to claims 3 and 4, respectively, and are therefore rejected in the same manner as these claims, as set forth above.
As per claim 17, this claim is substantially similar to the limitations of claims 1 and 3 and is therefore rejected in the same manner as these claims, as set forth above.
As per claims 18-20, these claims are substantially similar to claims 2, 4, and 5, respectively, and are therefore rejected in the same manner as these claims, as set forth above.
Claims 6 and 14 are rejected under 35 U.S.C. 103 as being unpatentable over Fuxman/Github in view of Sahu (Sahu, Gaurav, et al. Data Augmentation for Intent Classification with Off-the-shelf Large Language Models. In Proceedings of the 4th Workshop on NLP for Conversational AI, pages 47–57, Dublin, Ireland. Association for Computational Linguistics. (2022).
As per claim 6, Fuxman/Github teach the invention of claim 1 as set forth above. Fuxman/Github do not explicitly disclose the following known technique which is taught by Sahu:
further comprising: for each of the plurality of product classifications, using a second trained LLM to generate a plurality of corresponding products; wherein the word embedding further comprises word embedding each of the plurality of corresponding products. (Sahu: Page 48-51 (creating prompts from the available examples and feeding them to a large language model such as GPT-3 (Brown et al., 2020). Figure 1 illustrates the process of data generation for an intent with K available examples. . . . When GPT-3 or a baseline method is used to augment the training data we generate N − K examples per intent and refer to the resulting data as D˜ F,train. . . . We fine-tune BERT-large (Devlin et al., 2018) on the task of intent classification by adding a linear layer on top of the [CLS] token (Wolf et al., 2019).))
This known technique is applicable to the method of Fuxman/Github as they share characteristics and capabilities, namely, they are directed to classifications.
One of ordinary skill in the art at the time of filing would have recognized that applying the known technique of Sahu would have yielded predictable results and resulted in an improved method. It would have been recognized that applying the technique of Sahu to the teachings of Fuxman/Github would have yielded predictable results because the level of ordinary skill in the art demonstrated by the references applied shows the ability to incorporate such second trained LLM features into similar methods. Further, applying the for each of the plurality of product classifications, using a second trained LLM to generate a plurality of corresponding products; wherein the word embedding further comprises word embedding each of the plurality of corresponding products to the teachings of Fuxman/Github would have been recognized by those of ordinary skill in the art as resulting in an improved method that would allow classification even when available training data is very scarce (Sahu: Page 47).
As per claim 14, this claim is substantially similar to claim 6 and is therefore rejected in the same manner as this claim, as set forth above.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure:
Malhotra, Tanya. "What if LLM hallucinations were a feature and not a bug? meet dreamgpt: An open-source GPT-based solution that uses hallucinations from large language models (llms) as a feature." MarkTechPost. https://www.marktechpost.com/2023/05/20/what-if-llm-hallucinations-were-a-feature. (20 May 2023).
Jiang, Xuhui, et al. "A survey on large language model hallucination via a creativity perspective." arXiv preprint arXiv:2402.06647 (2 Feb 2024).
Lee, Jieh-Sheng, and Jieh Hsiang. "Prior art search and reranking for generated patent text." arXiv preprint arXiv:2009.09132 (July 18, 2021).
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to JENNIFER V LEE whose telephone number is (571)272-4778. The examiner can normally be reached Monday - Friday 9AM - 5PM 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, JEFFREY A. SMITH can be reached at (571)272-6763. 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.
/JENNIFER V LEE/Examiner, Art Unit 3688
/Jeffrey A. Smith/Supervisory Patent Examiner, Art Unit 3688
Prosecution Insights