Elicit
elicit.org
elicit.org
You can also teach Elicit to do custom tasks by giving input-output examples of intended behavior, e.g. for brainstorming counterarguments or decomposing research questions.
Our mailing list at list.elicit.org has more about what we've recently been adding.
Not related to the service itself, but I strongly dislike those random (seemingly personal but are actually not) automated emails that start coming in 15 minutes after signing up. This is spam. Please don't spam.
From playing around a little with it, just a few points of feedback.
The first issue I noticed was with LaTeX syntax being presented in search results, e.g. "$${\text {O}}\bigl (f\bigr )$$O(f) rounds or $${\text {O}}\bigl (f\delta \bigr )$$O(fδ) time to achieve consensus". This also appeared in "Most Relevant Sentence" on my favorites page for the same paper (DOI: 10.1007/s00236-019-00334-w).
I also tried abstract summarization with the whitepaper for Cassandra's "CEP-15: General Purpose Transactions", and it came up with completely made-up facts such as describing the ACCORD algorithm (not an acronym) as Atomic Consensus by Convex Ordering of Replicated Data, something like that. I don't know what convex means in this context, but I would suggest tuning this particular feature to be a bit less creative :-)
There seem to be many more features that I haven't tried yet – or don't really have a need for – and it would be good to see those explained more, with examples/tutorials and generally more documentation.
I also second beshrkayali's earlier complaint that the follow-up email is spam. I unsubscribed and reported it as such. I never signed up for this and was never warned I would sign up or given the option not to. Do better.
Could we have a desktop version to data-mine local collections of knowledge material (i.e. all kinds of media)?
I would really like an AI desktop search engine that can answer natural language questions.
For now we've chosen to build Elicit around aggregating and synthesizing research - what is the evidence in each paper, what are the arguments, how do they come together to inform questions people care about?
We think research is a good starting point for figuring out how to use language models to do high-quality reasoning because (a) researchers care a lot about what's true and (b) research is already more structured than most other sources of knowledge.
Actually it would be interesting to mix in some world-knowledge base like Cyc into such tool to make querying more powerful, and maybe even enable deduction of additional information out of the recognized facts and structures.
But it should still run locally of course. I'm refusing to pay for any SaaS offers.
Especially as it comes to research with commercial background it's also a matter of confidentiality. So I guess I'm not the only one who would prefer to have such a thing strictly on-prem.
What? You can run software on a computer you own? How does it work if it doesn't have cloud in it?
"Don't stab me with that knife" --> "There are lots of ways that person can hurt me other than by stabbing me with a knife"