Which LLM is the best at doing scientific literature searches?

[Alfred001]

I find that LLMs tend to do relatively ok with searching for scientific literature so long as the query is very basic, but as soon as you add a bit of nuance and conditions, it tends to ignore the conditions and ends up giving you only tangentially related literature.

Which free LLMs have you found to be the best at this?

[iNow]

OpenAIs Deep Research comes to mind, as does Consensus. Perplexity has one, too.

 

https://openai.com/index/introducing-deep-research/

https://consensus.app/

https://www.perplexity.ai/hub/blog/introducing-perplexity-deep-research

[swansont]

I asked all three who built the first rubidium fountain clock. They gave three different answers. Perplexity got it right, and gave a decent summary. Consensus gave a list of references but couldn’t figure out that 2004 was earlier than other dates.

[iNow]

Would one expect to find lots of content in google scholar citing who built the first rubidium clock? These tools should be better at explaining how such a clock functions and/or how it's evolved over time

[CharonY]

I have tried the tools on a topic I am currently writing on. They tend not to synthesize findings very well I found, either it is a list or just a highlight from a handful of usually recent publications. The area is somewhat cutting edge, though. They look better when asking specific questions, but the best answer is basically just citing one paper over and over.

[Alfred001]

2 minutes ago, CharonY said:

I have tried the tools on a topic I am currently writing on. They tend not to synthesize findings very well I found, either it is a list or just a highlight from a handful of usually recent publications. The area is somewhat cutting edge, though. They look better when asking specific questions, but the best answer is basically just citing one paper over and over.

I'm more looking for something that can give me a good list of references that are pertinent to the specific topic, rather than give an answer. I find that as soon as you add some complexity to the topic, the references become increasingly less related to what you're looking for.

Like, if you ask how sodium intake affects blood pressure, it will do fine, but if you say how does sodium intake affect blood pressure in people above the age of 60 and with intakes below x, it will just give you references related to the first question and ignore the qualifiers.

[CharonY]

I tried getting to provide a kind of review on the topic but it tends to be too narrow or ends up being list with with often dubious relevance.