“The Assistants API means that you can construct AI assistants inside your personal purposes. An Assistant has directions and might leverage fashions, instruments, and data to reply to consumer queries”, OpenAI.

Sounds nice, so we’re going to take a look at how we will use the brand new API to do information evaluation on native information.

The Assistants API represents an strategy that’s an alternative choice to at the least some makes use of of Retrieval Augmented Era (RAG). So has RAG simply been a stopgap measure, a brief resolution to the drawbacks of the present era of LLMs? In any case, LlamaIndex’s Jerry Liu has stated that RAG is only a hack (albeit a strong one).

Listed below are three particular issues inherent to LLMs that RAG at the moment addresses and that the Assistants API may also deal with:

• LLMs are old-fashioned. It takes loads of money and time (to not point out power) to coach a mannequin, so the data that they have been educated on may very well be a few years previous.
• LLMs don’t find out about your information. It’s fairly unlikely that your information have been a part of the coaching set for an LLM.
• LLMs hallucinate. Typically they’ll give completely believable responses which might be completely false.

By offering the LLM with information that’s related to your software you’ll be able to cut back these issues.

For instance, if you need the LLM to provide Streamlit code, you could possibly give it information from the newest documentation to allow it to make use of new options of the framework. Or, if you wish to do some evaluation on some particular information, then clearly giving it that information is important. And, lastly, by offering related information to the LLM, you improve the possibility of it offering an appropriate response and thus cut back the potential for it simply making issues up.

Whereas RAG has been used to mitigate these points, they’re now additionally addressed by the brand new Assistants API. The RAG strategy makes…