41–50 of 174 posts
Before making AI do research, perhaps we should first let it __reproduce__ research. For example, give it a paper of some deep learning technique and make it produce an implementation of that paper. Before it can do that, I have no hope that it can produce novel ideas.
[flagged]
Human "reality checker" systems are analogous to a discriminator in a generative adversarial network, but strongly informed by emotion. Psychology tells us that regardless of how "emotional" we are, our sense of truth/falsehood goes first through an emotional circuit, which is informed by underlying beliefs. If someone states something you strongly disagree with, your first internal response will be emotional; then y…
Earlier quoted context omitted.
Every system has problems. The better question is: what is the acceptable threshold? For an example Medicare and Medicade had a fraud rate of 7.66%. Yes, that is a lot of billions, and there is room for improvement, but that doesn’t mean the entire system is failing: 93% of cases are being covered as intended. The same could be said with these models. If the spoilage rate is 10%, does that mean the whole system is ba…
In the protein annotation world, which is largely driven by inferring common ancestry between a protein of unknown function and one of known function, common error thresholds range from FDR of 0.001 to 10^-6. Even a 1% error rate would be considered abysmal. This is in part because it is trivial to get 95% accuracy in prediction; the challenging problem is to get some large fraction of the non-trivial 5% correct. "Ac…
Before making AI do research, perhaps we should first let it __reproduce__ research. For example, give it a paper of some deep learning technique and make it produce an implementation of that paper. Before it can do that, I have no hope that it can produce novel ideas.
Producing novel ideas is the most famous trait of current LLMs, the thing people are spending all their time trying to prevent.
"And for most deep learning papers I read, domain experts have not gone through the results with a fine-tooth comb inspecting the quality of the output. How many other seemingly-impressive papers would not stand up to scrutiny?" Is this really not the case? I've read some of the AI papers in my field, and I know many other domain experts have as well. That said I do think that CS/software based work is generally easi…
Validation of biological labels easily takes years - in the OP's example it was a 'lucky' (huge!) coincidence that somebody already had spent years on one of the predicted proteins' labels. Nobody is going to stake 3-5 years of their career on validating some random model's predictions.
It’s like fake news is taking in science now. Saying any stupid thing will attract much more view and « likes » than those debunking them. Except that we can’t compare twitter to nature journal. Science is supposed to be immune to these kind of bullshit thanks to reputed journals and pair reviewing, blocking a publication before it does any harm. Was that a failure of nature ?
Have you seen the statistics about high impact journals having higher retraction/unverified rates on papers? The root causes can be argued...but keep that in mind. No single paper is proof. Bodies of work across many labs, independent verification, etc is the actual gold standard.
[1] - https://search.brave.com/search?q=site%3Anytimes.com+Journal...
[2] - https://en.wikipedia.org/wiki/Replication_crisis#In_psycholo...
Anyone here still doing verification or reproduction work? Feels like it’s becoming rare, but I find it super valuable.
Let's suppose you read a paper that does "X with Y", but you are interested in "Z", so the brilliant idea is to do "Z with Y" and publish the new combination, and citing it.
Sometimes you cross your fingers and just try "Z with Y", but if the initial attempt fails or you are too cautious you try "X with Y" to ensure you understand the details of the original paper.
If the reproduction of "X with Y" is a success, you now try "Z with Y" and if it works you publish it.
If the reproduction of "X with Y" is a failure, you may email the authors of just drop the original paper in the recycle bin. Publishing a failure of a reproduction is too difficult. This is a bad incentive, but it's also too easy to make horrible mistakes and fail.