Hallucination detection and remediation in text generation interface systems
Enumerated source text passages may be determined based on one or more source text documents. The enumerated source text passages may include source text passage identifiers uniquely identifying the passages. A novel text passage including novel text portions may be determined based on a query and the enumerated source text passages. One or more of the novel text portions may be verified by a large language model to produce text verification information. A novel text generation message including novel text generated by the large language model may be determined based on the text verification information and sent to a client machine.
1 . A method comprising:
determining a plurality of enumerated source text passages based on one or more source text documents, each of the plurality of enumerated source text passages being included in one or more of the one or more source text documents, each enumerated source text passage of the plurality of enumerated source text passages including a respective source text passage identifier that uniquely identifies the enumerated source text passage;
determining a novel text passage based on a query and the plurality of enumerated source text passages, the novel text passage including a plurality of novel text portions, one or more text portions of the plurality of novel text portions including a respective one or more cited source text passage identifiers;
transmitting a text verification prompt to a large language model, the text verification prompt including a designated novel text portion of the plurality of novel text portions, the text verification prompt including a designated one or more enumerated source text passages identified in the designated novel text portion, the text verification prompt including a natural language instruction to evaluate whether the designated novel text portion is supported by the designated one or more enumerated source text passages;
receiving from the large language model a text verification response message including text verification information indicating that the designated novel text portion is insufficiently supported by the designated one or more enumerated source text passages; and
transmitting to a client machine a novel text generation message determined based on the text verification information and including novel text generated by the large language model.
2 . The method recited in claim 1 , the method further comprising:
selecting a subset of the plurality of enumerated source text passages based on their relevance to the query, wherein the novel text passage is determined based on the selected subset of the plurality of enumerated source text passages.
3 . The method recited in claim 2 , wherein the subset of the plurality of enumerated source text passages is selected based on a completed relevance prompt received from a large language model, the completed relevance prompt being determined based on an input relevance prompt that includes one or more natural language instructions executed by the large language model to select the subset of the plurality of enumerated source text passages.
4 . The method recited in claim 1 , wherein the novel text passage is generated by the large language model based on a text generation prompt sent to the large language model, the text generation prompt including one or more of the plurality of enumerated source text passages and one or more natural language instructions to generate the novel text passage based on the one or more of the plurality of enumerated source text passages.
5 . The method recited in claim 1 , wherein the query includes a request to summarize the one or more source text documents.
6 . The method recited in claim 1 , wherein the query includes a request to answer one or more questions based on the one or more source text documents.
7 . The method recited in claim 1 , wherein the query includes a request to generate an argument based on the one or more source text documents.
8 . One or more non-transitory computer readable media having instructions stored thereon for performing a method, the method comprising:
determining a plurality of enumerated source text passages based on one or more source text documents, each of the plurality of enumerated source text passages being included in one or more of the one or more source text documents, each enumerated source text passage of the plurality of enumerated source text passages including a respective source text passage identifier that uniquely identifies the enumerated source text passage;
determining a novel text passage based on a query and the plurality of enumerated source text passages, the novel text passage including a plurality of novel text portions, one or more text portions of the plurality of novel text portions including a respective one or more cited source text passage identifiers;
transmitting a text verification prompt to a large language model, the text verification prompt including a designated novel text portion of the plurality of novel text portions, the text verification prompt including a designated one or more enumerated source text passages identified in the designated novel text portion, the text verification prompt including a natural language instruction to evaluate whether the designated novel text portion is supported by the designated one or more enumerated source text passages;
receiving from the large language model a text verification response message including text verification information indicating that the designated novel text portion is insufficiently supported by the designated one or more enumerated source text passages; and
transmitting to a client machine a novel text generation message determined based on the text verification information and including novel text generated by the large language model.
9 . The one or more non-transitory computer readable media recited in claim 8 , the method further comprising:
selecting a subset of the plurality of enumerated source text passages based on their relevance to the query, wherein the novel text passage is determined based on the selected subset of the plurality of enumerated source text passages.
10 . The one or more non-transitory computer readable media recited in claim 9 , wherein the subset of the plurality of enumerated source text passages is selected based on a completed relevance prompt received from the large language model, the completed relevance prompt being determined based on an input relevance prompt that includes one or more natural language instructions executed by the large language model to select the subset of the plurality of enumerated source text passages.
11 . The one or more non-transitory computer readable media recited in claim 8 , wherein the novel text passage is generated by the large language model based on a text generation prompt sent to the large language model, the text generation prompt including one or more of the plurality of enumerated source text passages and one or more natural language instructions to generate the novel text passage based on the one or more of the plurality of enumerated source text passages.
12 . The one or more non-transitory computer readable media recited in claim 8 , wherein the query includes a request to summarize the one or more source text documents.
13 . The one or more non-transitory computer readable media recited in claim 8 , wherein the query includes a request to answer one or more questions based on the one or more source text documents.
14 . The one or more non-transitory computer readable media recited in claim 8 , wherein the query includes a request to generate an argument based on the one or more source text documents.
15 . A system comprising:
one or more processors configured to perform a method, the method comprising:
determining a plurality of enumerated source text passages based on one or more source text documents, each of the plurality of enumerated source text passages being included in one or more of the one or more source text documents, each enumerated source text passage of the plurality of enumerated source text passages including a respective source text passage identifier that uniquely identifies the enumerated source text passage;
determining a novel text passage based on a query and the plurality of enumerated source text passages, the novel text passage including a plurality of novel text portions, one or more text portions of the plurality of novel text portions including a respective one or more cited source text passage identifiers;
transmitting a text verification prompt to a large language model, the text verification prompt including a designated novel text portion of the plurality of novel text portions, the text verification prompt including a designated one or more enumerated source text passages identified in the designated novel text portion, the text verification prompt including a natural language instruction to evaluate whether the designated novel text portion is supported by the designated one or more enumerated source text passages;
receiving from the large language model a text verification response message including text verification information indicating that the designated novel text portion is insufficiently supported by the designated one or more enumerated source text passages; and
transmitting to a client machine a novel text generation message determined based on the text verification information and including novel text generated by the large language model.
16 . The system recited in claim 15 , the method further comprising:
selecting a subset of the plurality of enumerated source text passages based on their relevance to the query, wherein the novel text passage is determined based on the selected subset of the plurality of enumerated source text passages.
17 . The system recited in claim 16 , wherein the subset of the plurality of enumerated source text passages is selected based on a completed relevance prompt received from the large language model, the completed relevance prompt being determined based on an input relevance prompt that includes one or more natural language instructions executed by the large language model to select the subset of the plurality of enumerated source text passages.
18 . The system recited in claim 15 , wherein the novel text passage is generated by the large language model based on a text generation prompt sent to the large language model, the text generation prompt including one or more of the plurality of enumerated source text passages and one or more natural language instructions to generate the novel text passage based on the one or more of the plurality of enumerated source text passages.
19 . The system recited in claim 15 , wherein the query includes a request to summarize the one or more source text documents.
20 . The system recited in claim 15 , wherein the query includes a request to answer one or more questions based on the one or more source text documents.