Hacker News Reader

Comment by neltnerb | Hacker News Reader

Is there an article explaining how it is like HN? I thought the algorithm here was upvotes plus some time constants and flagging mostly, what does "Let AI simplify articles, clarify terms, and help you do research" mean?

I like the idea of trying to produce decent composite review-level datasets of scientific papers if that's what you mean. I have way less objection to this than most AI because the results are supposed to be public and accessible in exactly this way.

Is it something revolutionary and amazingly useful like that, or is this more just deciding what to show you in a feed based on past history?

View on HN

> what does "Let AI simplify articles, clarify terms, and help you do research" mean?

If you select a text from a paper and click on "Simplify" in a popup, it will try to tell you what it means. The "help you do research" is a more broad part thought huh

> Is it something revolutionary and amazingly useful like that

We (right now a team of two) plan to develop an AI-assisted peer review. What I mean is that we can use LLMs to check for some "bad things" in paper (like small sample size or some weird hypothesis etc) + analyze the comments and the users behind these comments.

> just deciding what to show you in a feed based on past history?

Surely not, but this is what we already have in the product right now

Sounds reasonably handy, do you rely on users to tell you if the LLM generated simplifications are misleading? It is clearly doing more than just popping up definitions of confusing words, kudos for that.

This makes me want a tool that lets the scientists (who obviously are fine writers but don't specialize in writing) write at whatever level of detail they want, and then shorten or lengthen it to different audiences. EXCEPT the actual authors would verify and publish specific alternates.

I think it's great to have a 3 page short paper for super experts in a short journal article (though I don't see why journal articles have page limits in 2023). I think it's better if that three page paper can be turned into a ten page paper that's way easier to read and which the authors actually vetted as accurately representing what they did.

> Sounds reasonably handy, do you rely on users to tell you if the LLM generated simplifications are misleading?

Not yet, but this is in our roadmap for sure

> EXCEPT the actual authors would verify and publish specific alternates.

This is also a good one, except that there's already 100s of millions of research papers that need to be summarized. But probably we can ask the audience / researchers themselves to verify summaries once we produce a summary for a particular paper

Reply on news.ycombinator.com