3 ms·

LLMs are able to utilize “all the worlds” knowledge during training and give seemingly magical answers. While providing context in the query is different than t

LLMs are able to utilize “all the worlds” knowledge during training and give seemingly magical answers. While providing context in the query is different than training models, is it possible that more context will give more materials to the LLM and it will be able to pick out the relevant bits on its own? What if it was possible, with each query, to fine tune the model on the provided context, and then use that JIT fine-tuned model to answer the query?

acchow 3y ago
Are you asking what if it was possible that a "context window" ceased to exist? In a different architecture than we currently use, I guess that's hypothetically possible. As it is now, you can't fine tune on context. It would have almost no effect on the parameters. Context is like giving your friend a magazine article and asking them to respond to it. Fine tuning is like throwing that magazine article into the ocean of all content they ever came across during their lifetime.
- koliber 3y ago
  I am not an expert here so I may be mixing terms and concepts. The way I understand it, there is a base model that was trained on vast amount of general data. This sets up the weights. You can fine-tune this base model on additional data. Often this is private data that is concentrated around a certain domain. This modifies the model's weights some more. Then you have the context. This is where your query to the LLM goes. You can also add the chat history here. Also, system prompts that tell the LLM to behave a certain way go here. Finally, you can take additional information from other sources and provide it as part of the context -- this is called Retrieval Augmented Generation. All of this really goes into one bucket called the context, and the LLM needs to make sense of it. None of this modifies the weights of the model itself. Is my mental picture correct so far? My question is around RAG. It seems that providing additional selected information from your knowledge base, or using your knowledge base to fine-tune a model, seem similar. I am curious in which ways these are similar, and in which ways they cause the LLM to behave differently. Concretely, say I have a company knowledge base with a bunch of rules and guidelines. Someone asks an agent "Can I take 3 weeks off in a row?" How would these two scenarios be different: a) Agents searches the knowledge base for all pages and content related to "FTO, PTO, time off, vacations" and feeds those articles to the LLM, together with the "Can I take 3 weeks off in a row?" query b) I have an LLM that has been fine tuned on all the content in the knowledge base. I ask it "Can I take 3 weeks off in a row?"
  - acchow 3y ago
    > Is my mental picture correct so far? Yes > How would these two scenarios be different They're different in exactly the way you described above. The agent searching the knowledge base for "FTO, PTO, time off, vacations" would be the same as you pasting all the articles related to those topics into the prompt directly - in both cases, it goes into the context. In scenario a, you'll likely get the correct response. In scenario b, likely get an incorrect response. Why? Because of what you explained above. Fine tuning adjusts the weights. When you adjusts weights by feeding data, you're only making small adjustments to shift slightly along a curve - thus the exposure to this data (for the purposes of fine tuning) will have very little effect on the next context the model is exposed to.

acchow 3y ago

Are you asking what if it was possible that a "context window" ceased to exist? In a different architecture than we currently use, I guess that's hypothetically possible. As it is now, you can't fine tune on context. It would have almost no effect on the parameters. Context is like giving your friend a magazine article and asking them to respond to it. Fine tuning is like throwing that magazine article into the ocean of all content they ever came across during their lifetime.

koliber 3y ago
I am not an expert here so I may be mixing terms and concepts. The way I understand it, there is a base model that was trained on vast amount of general data. This sets up the weights. You can fine-tune this base model on additional data. Often this is private data that is concentrated around a certain domain. This modifies the model's weights some more. Then you have the context. This is where your query to the LLM goes. You can also add the chat history here. Also, system prompts that tell the LLM to behave a certain way go here. Finally, you can take additional information from other sources and provide it as part of the context -- this is called Retrieval Augmented Generation. All of this really goes into one bucket called the context, and the LLM needs to make sense of it. None of this modifies the weights of the model itself. Is my mental picture correct so far? My question is around RAG. It seems that providing additional selected information from your knowledge base, or using your knowledge base to fine-tune a model, seem similar. I am curious in which ways these are similar, and in which ways they cause the LLM to behave differently. Concretely, say I have a company knowledge base with a bunch of rules and guidelines. Someone asks an agent "Can I take 3 weeks off in a row?" How would these two scenarios be different: a) Agents searches the knowledge base for all pages and content related to "FTO, PTO, time off, vacations" and feeds those articles to the LLM, together with the "Can I take 3 weeks off in a row?" query b) I have an LLM that has been fine tuned on all the content in the knowledge base. I ask it "Can I take 3 weeks off in a row?"
- acchow 3y ago
  > Is my mental picture correct so far? Yes > How would these two scenarios be different They're different in exactly the way you described above. The agent searching the knowledge base for "FTO, PTO, time off, vacations" would be the same as you pasting all the articles related to those topics into the prompt directly - in both cases, it goes into the context. In scenario a, you'll likely get the correct response. In scenario b, likely get an incorrect response. Why? Because of what you explained above. Fine tuning adjusts the weights. When you adjusts weights by feeding data, you're only making small adjustments to shift slightly along a curve - thus the exposure to this data (for the purposes of fine tuning) will have very little effect on the next context the model is exposed to.