Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

The Humans Building AI Scientists

asimov.press

Post reply on HN
Print view View original

21–28 of 28 posts

Re: The Humans Building AI Scientists

#21

Post by nsagent »

light_hue_1 wrote: post #5

I'm so incredibly tired of all of the BS claims. (I'm an AI/ML researcher) > has enabled open-source LLMs “to exceed human-level performance on two more of the lab-bench tasks: doing scientific literature research and reasoning about DNA constructs” with only “modest compute budgets.” No. They did not. They just ran a crappy experiment and came up with an absurd result. As a community we need to invest much more effo…

I'm also an AI researcher and I'm not hopeful that the hype will die down anytime soon. People have been touting insignificant results through shoddy science for a long time now. They've noticed it works well enough because current ML is still pretty much alchemy (Ali Rahimi's 2017 NIPS Test of Time [1] talk still resonates today) so people rarely spend the effort to effectively refute the bogus claims.

As a result, I've opted out of the system and I'm working on trying ambitious ideas I have to try to upend the current paradigm of training on big data, which is truly insane (by age 4 the most erudite of children have probably only been exposed to 45 million words [2] and yet exhibit vastly more understanding and fluency than any language model trained on a similar amount of data).

[1]: https://www.youtube.com/watch?v=Qi1Yry33TQE

[2]: https://www.aft.org/sites/default/files/media/2014/TheEarlyC...

Re: The Humans Building AI Scientists

#22

Post by kevinventullo »

the_snooze wrote: post #15
plaidfuji wrote: post #3

> Rodriques: Many people assume we’re focused on wet lab automation. There are certainly opportunities there and we are exploring them, but the biggest opportunities are actually on the cognitive side. Wet lab automation is very difficult and capital intensive. And once you build your lab you are constraining yourself to answering questions within a certain domain for which you have the relevant sample prep and chara…

>Or to think about this another way, imagine a PhD student who was never allowed to talk to other people, attend conferences etc., and could only read papers and try things in lab. But they can read papers extremely fast. Would they be successful? For anyone who thinks this is sufficient, be aware that papers only tell you what's successful (for some definition of successful). Being part of a research community gives…

This comment made me realize something. There’s always been grumbling in the academic community about how negative or less exciting results aren’t publishable. As a result, there is quite a bit of knowledge that is essentially “lost” as no one ever bothers to write it down. Part of the reason this happens is that publishing unexciting results is unhelpful professionally, but I suspect another aspect here is that researchers know that no one would read such results; it is hard enough to keep up with the positive results in the field, much less negative results. So, in that sense they are not “contributing to the sum total of human knowledge”, which I think is a major part of many scientists’ motivations.

So then, in a world where the outcomes of all experiments could be reasonably fed into an AI model, it seems that there could be a great deal of value in having scientists publish these “low value” negative results, even in a relatively informal format (i.e. not worrying about formatting, perhaps skipping peer review, etc.). That way even if a human never reads the paper, at the very least an AI “scientist” model would pick it up.

I don’t think you can professionally incentivize this. Rather, scientists would need to do this out of the desire to contribute to the sum total of human knowledge, which could be embodied and not forgotten by these scientist AI’s.

Re: The Humans Building AI Scientists

#23

Post by ehsanu1 »

kevinventullo wrote: post #22
the_snooze wrote: post #15

Earlier quoted context omitted.

>Or to think about this another way, imagine a PhD student who was never allowed to talk to other people, attend conferences etc., and could only read papers and try things in lab. But they can read papers extremely fast. Would they be successful? For anyone who thinks this is sufficient, be aware that papers only tell you what's successful (for some definition of successful). Being part of a research community gives…

This comment made me realize something. There’s always been grumbling in the academic community about how negative or less exciting results aren’t publishable. As a result, there is quite a bit of knowledge that is essentially “lost” as no one ever bothers to write it down. Part of the reason this happens is that publishing unexciting results is unhelpful professionally, but I suspect another aspect here is that rese…

There seems to be a couple of field-specific journals of negative results for similar purposes. It seems like there should be value in citing negative results to inform current research. Perhaps if there were more journals dedicated to this, or a single one not limited to specific fields, there would still be some incentive to publish there, if the effort required was low enough (another area where AI might be applied: writing it up).

Re: The Humans Building AI Scientists

#26

Post by Xmd5a »

aithrowawaycomm wrote: post #8

I will never stop being amazed at AI folks' childish views of animal cognition: > A lot of your tools reference crows. What’s up with that? > White: When I got started in this space around October 2022, I was red-teaming with GPT4. Around the same time, a paper called “Language Models are Stochastic Parrots” was circulating, and people were debating whether these models were just regurgitating their training data or…

>This is incredibly insulting to crows

Re: The Humans Building AI Scientists

#27

Post by jaggederest »

MostlyStable wrote: post #18
jaggederest wrote: post #6

Earlier quoted context omitted.

I'm reminded of the time I saw some A/B test results that didn't make much sense, but were highly significant[1] I asked how many A/B tests they were running... hundreds. Overlapping. At least they had a holdout group (that they mostly ignored, which indicated that all the A/B tests more or less made no difference) [1] P < 0.001 with a large effect size. No, your a/b test probably didn't break the laws of economics -…

If they were running many concurrent, overlapping A/B tests, then they didn't necessarily mess up their data. You are likely to get that result, honestly and truthfully, purely by chance, if you run enough tests. Unless the "running concurrent tests and not correcting your significance level", is what you meant by messing up their data, in which case yeah.

Yes precisely, the latter. I know there are techniques to disentangle that kind of thing but neither I nor they had them readily to hand.

Re: The Humans Building AI Scientists

#28

Post by godelski »

kevinventullo wrote: post #22
the_snooze wrote: post #15

Earlier quoted context omitted.

>Or to think about this another way, imagine a PhD student who was never allowed to talk to other people, attend conferences etc., and could only read papers and try things in lab. But they can read papers extremely fast. Would they be successful? For anyone who thinks this is sufficient, be aware that papers only tell you what's successful (for some definition of successful). Being part of a research community gives…

This comment made me realize something. There’s always been grumbling in the academic community about how negative or less exciting results aren’t publishable. As a result, there is quite a bit of knowledge that is essentially “lost” as no one ever bothers to write it down. Part of the reason this happens is that publishing unexciting results is unhelpful professionally, but I suspect another aspect here is that rese…

Where the publishing of negative results is needed (and generally wanted) is in the paper of the experiments that did work. The problem is that paper is missing all the failures along the way. It is totally fine if this is all thrown in the appendix but the problem becomes that we have to write things quickly and including negative results only increases the likelihood of your work getting rejected because reviewers see their role as antagonistically, to find errors. The problem really comes down to not letting people admit mistakes. Good work can be done even while many mistakes exist. But a work is better when the mistakes are easier to find. Short term vs long term reward. Because in the long term someone needs to replicate your experiments, even in some minor way just to show that their thing is better than yours. Having information about what didn't work is helpful for learning new paths that will work.

I don't think many scientists really want to publish papers like "We did X, X had no results", but rather "We did X! X is awesome! But along the way we tried Y, Z, ... which were not so successful. We think Y because... we have no fucking clue about Z." You're still pushing the exciting stuff. You're just giving it more context.

Post reply on HN