Computer-generated content based on text classification, semantic relevance, and activation of deep learning large language models
The disclosure relates to systems and methods of automatically generating unique content including natural language text based on a corpus of previously generated response documents and discrete requirements defined in a requirements specification. The system may use generative stitching that includes multi-layer processes that execute to influence the generation of unique content including natural language text through an artificial intelligence (AI) language transformer model trained to output the content based on previously written material that is semantically relevant to the discrete requirements and is weighted against labeled attributes. The labeled attributes may determine the influence asserted against the language transformer, thereby generating unique on-target content that may be combined to create a computer-generated response document.
1. A system of automated prompt engineering for large language models (LLMs), comprising:
a processor programmed to:
receive a requirement for which an LLM is to automatically generate a computer-generated response, the LLM being pretrained with natural language content to automatically generate text;
generate one or more requirement vectors based on the requirement;
identify one or more candidate response portions that are semantically similar to the requirement based on the one or more requirement vectors and one or more candidate response vectors corresponding to the one or more candidate response portions, the one or more candidate response portions having been previously generated in response to one or more previous requirements;
generate an LLM prompt based on the identified one or more candidate response portions and the requirement, the LLM prompt providing, to the LLM, input that includes one or more concepts from the requirement; and
provide the LLM prompt to the LLM, which is to generate the computer-generated response responsive to the LLM prompt.
2. The system of claim 1 , wherein to identify the one or more candidate response portions, the processor is further programmed to:
compare the one or more requirement vectors and the one or more candidate response vectors to identify a delta between the one or more requirement vectors and the one or more candidate response vectors;
determine a semantic similarity between the requirement and the one or more candidate response portions based on the delta; and
generate the LLM prompt based on the semantic similarity.
3. The system of claim 2 , wherein to compare the one or more requirement vectors and the one or more candidate response vectors, the processor is further programmed to:
determine a dot product score between the one or more requirement vectors and the one or more candidate response vectors.
4. The system of claim 2 , wherein to generate the LLM prompt, the processor is further programmed to:
generate a series of LLM prompts for input to the LLM, the series of LLM prompts being provided to the LLM in a prompt chaining sequence that specifies an order in which to provide the series of LLM prompts.
5. The system of claim 4 , wherein the processor is further programmed to:
modify the prompt chaining sequence of the LLM prompts based on the semantic similarity.
6. The system of claim 4 , wherein to generate the LLM prompt, the processor is further programmed to:
modify one or more of the prompts in the prompt chaining sequence based on the semantic similarity.
7. The system of claim 2 , wherein the processor is further programmed to:
generate, based on the semantic similarity, an instruction to add new concepts that match both the requirement and the one or more candidate response portions; and
include the new concepts in the LLM prompt.
8. A method of automated prompt engineering for large language models (LLMs), comprising:
receiving, by a processor, a requirement for which an LLM is to automatically generate a computer-generated response, the LLM being pretrained with natural language content to automatically generate text;
generating, by the processor, one or more requirement vectors based on the requirement;
identifying, by the processor, one or more candidate response portions that are semantically similar to the requirement based on the one or more requirement vectors and one or more candidate response vectors corresponding to the one or more candidate response portions, the one or more candidate response portions having been previously generated in response to one or more previous requirements;
generating, by the processor, an LLM prompt based on the identified one or more candidate response portions and the requirement, the LLM prompt providing, to the LLM, input that includes one or more concepts from the requirement; and
providing, by the processor, the LLM prompt to the LLM, which is to generate the computer-generated response responsive to the LLM prompt.
9. The method of claim 8 , wherein identifying the one or more candidate response portions comprises:
comparing the one or more requirement vectors and the one or more candidate response vectors to identify a delta between the one or more requirement vectors and the one or more candidate response vectors;
determining a semantic similarity between the requirement and the one or more candidate response portions based on the delta; and
generating the LLM prompt based on the semantic similarity.
10. The method of claim 9 , wherein comparing the one or more requirement vectors and the one or more candidate response vectors comprises:
determining a dot product score between the one or more requirement vectors and the one or more candidate response vectors.
11. The method of claim 8 , wherein generating the LLM prompt comprises:
generating a series of LLM prompts for input to the LLM, the series of LLM prompts being provided to the LLM in a prompt chaining sequence that specifies an order in which to provide thee series of LLM prompts.
12. The method of claim 11 , further comprising:
modifying the prompt chaining sequence of the LLM prompts based on the semantic similarity.
13. The method of claim 11 , wherein generating the LLM prompt comprises:
modifying one or more of the prompts in the prompt chaining sequence based on the semantic similarity.
14. The method of claim 9 , further comprising:
generating, based on the semantic similarity, an instruction to add new concepts that match both the requirement and the one or more candidate response portions; and
including the new concepts in the LLM prompt.
15. A non-transitory computer readable medium storing instructions for automated prompt engineering for large language models (LLMs), the instructions when executed by a processor, programs the processor to: :
receive a requirement for which an LLM is to automatically generate a computer- generated response, the LLM being pretrained with natural language content to automatically generate text;
generate one or more requirement vectors based on the requirement;
identify one or more candidate response portions that are semantically similar to the requirement based on the one or more requirement vectors and one or more candidate response vectors corresponding to the one or more candidate response portions, the one or more candidate response portions having been previously generated in response to one or more previous requirements;
generate an LLM prompt based on the identified one or more candidate response portions and the requirement, the LLM prompt providing, to the LLM, input that includes one or more concepts from the requirement; and
provide the LLM prompt to the LLM, which is to generate the computer-generated response responsive to the LLM prompt.
16. The non-transitory computer readable medium of claim 15 , wherein to identify the one or more candidate response portions, the processor is further programmed to:
compare the one or more requirement vectors and the one or more candidate response vectors to identify a delta between the one or more requirement vectors and the one or more candidate response vectors;
determine a semantic similarity between the requirement and the one or more candidate response portions based on the delta; and
generate the LLM prompt based on the semantic similarity.
17. The non-transitory computer readable medium of claim 16 , wherein to compare the one or more requirement vectors and the one or more candidate response vectors, the processor is further programmed to:
determine a dot product score between the one or more requirement vectors and the one or more candidate response vectors.
18. The non-transitory computer readable medium of claim 16 , wherein to generate the LLM prompt, the processor is further programmed to:
generate a series of LLM prompts for input to the LLM, the series of LLM prompts being provided to the LLM in a prompt chaining sequence that specifies an order in which to provide thee series of LLM prompts.
19. The non-transitory computer readable medium of claim 18 , wherein the processor is further programmed to:
modify the prompt chaining sequence of the LLM prompts based on the semantic similarity.
20. The non-transitory computer readable medium of claim 18 , wherein to generate the LLM prompt, the processor is further programmed to:
modify one or more of the prompts in the prompt chaining sequence based on the semantic similarity.