ไม่สามารถเล่นวิดีโอนี้
ขออภัยในความไม่สะดวก
Training an LLM to effectively use information retrieval
ฝัง
- เผยแพร่เมื่อ 30 เม.ย. 2024
- This new paper presents an approach to train LLMs to effectively utilize information retrieval.
It first proposes a training approach to teach an LLM to generate a special token, RET, when it's not confident or doesn't know the answer to a question...
Paper: arxiv.org/abs/...
#ai #llms #machinelearning
Hello @Elvis, Thanks for summarizing this so nicely. I agree with your point that IR system can make it much better to use with small or large language model. I think with this we can use less complex language model and information retrieval to answer more complex questions. I believe that this is what we do ourselves as well, i.e., we use are knowledge whenever it is sufficient but when it is not and we identify those instances we access either external sources of knowledge or those compartmentalized knowledge in our brain to retrieve the information that a question requires to answer. This is a great set to actually make the large language model smaller but effective. Thanks for describing it so well. I am looking forward to the following articles.
Would love that to be a video series of interesting LLM paper summaries!