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

Deep learning gets the glory, deep fact checking gets ignored

rachel.fast.ai

Post reply on HN
Print view View original

31–40 of 174 posts

Re: Deep learning gets the glory, deep fact checking gets ignored

#31

Post by TeMPOraL »

softwaredoug wrote: post #29

We also love deep cherry picking. Working hard to find that one awesome time some ML / AI thing worked beautifully and shouting its praises to the high heavens. Nevermind the dozens of other times we tried and failed...

Even more so, we also love deep stochastic parroting. Working hard to ignore direct experience, growing amount of reports, and to avoid reasoning from first principles, in order to confidently deny the already obvious utility of LLMs, and backing that position with some tired memes.

Re: Deep learning gets the glory, deep fact checking gets ignored

#32

Post by mathgeek »

antithesizer wrote: post #28
godelski wrote: post #10

> although later investigation suggests there may have been data leakage I think this point is often forgotten. Everyone should assume data leakage until it is strongly evidenced otherwise. It is not on the reader/skeptic to prove that there is data leakage, it is the authors who have the burden of proof. It is easy to have data leakage on small datasets. Datasets where you can look at everything. Data leakage is rea…

The supposed location of the burden of proof is really not the definitive guide to what you ought to believe that people online seem to think it is.

Can you elaborate? You've made a claim, but I really think there'd be value in continuing to what you actually mean.

Re: Deep learning gets the glory, deep fact checking gets ignored

#33

Post by tbrownaw »

antithesizer wrote: post #28
godelski wrote: post #10

> although later investigation suggests there may have been data leakage I think this point is often forgotten. Everyone should assume data leakage until it is strongly evidenced otherwise. It is not on the reader/skeptic to prove that there is data leakage, it is the authors who have the burden of proof. It is easy to have data leakage on small datasets. Datasets where you can look at everything. Data leakage is rea…

The supposed location of the burden of proof is really not the definitive guide to what you ought to believe that people online seem to think it is.

What is the relevance of this generic statement to the discussion at hand?

Re: Deep learning gets the glory, deep fact checking gets ignored

#34

Post by slewis »

amelius wrote: post #2

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.

OpenAI created a benchmark for this: https://openai.com/index/paperbench/

Re: Deep learning gets the glory, deep fact checking gets ignored

#35

Post by NormLolBob »

mathgeek wrote: post #32
antithesizer wrote: post #28

Earlier quoted context omitted.

The supposed location of the burden of proof is really not the definitive guide to what you ought to believe that people online seem to think it is.

Can you elaborate? You've made a claim, but I really think there'd be value in continuing to what you actually mean.

They mean “vet your sources and don’t blindly follow the internet hive-mind.” or similar; burden of proof is not what the internet thinks.

Tacked their actual point on to the end of a copy paste of op comments context, ended up writing something barely grammatically correct.

In doing so they prove why exactly not to listen to the internet. So they have that going for them.

Re: Deep learning gets the glory, deep fact checking gets ignored

#36

Post by SamuelAdams »

godelski wrote: post #10

> although later investigation suggests there may have been data leakage I think this point is often forgotten. Everyone should assume data leakage until it is strongly evidenced otherwise. It is not on the reader/skeptic to prove that there is data leakage, it is the authors who have the burden of proof. It is easy to have data leakage on small datasets. Datasets where you can look at everything. Data leakage is rea…

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 bad? Or is it at a tolerable threshold?

[1]: https://www.cms.gov/newsroom/fact-sheets/fiscal-year-2024-im...

Re: Deep learning gets the glory, deep fact checking gets ignored

#37

Post by tbrownaw »

amelius wrote: post #2

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.

> For example, give it a paper of some deep learning technique and make it produce an implementation of that paper.

Or maybe give it a paper full of statistics about some experimental observations, and have it reproduce the raw data?

Re: Deep learning gets the glory, deep fact checking gets ignored

#38

Post by fastaguy88 »

SamuelAdams wrote: post #36
godelski wrote: post #10

> although later investigation suggests there may have been data leakage I think this point is often forgotten. Everyone should assume data leakage until it is strongly evidenced otherwise. It is not on the reader/skeptic to prove that there is data leakage, it is the authors who have the burden of proof. It is easy to have data leakage on small datasets. Datasets where you can look at everything. Data leakage is rea…

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.

"Acceptable" thresholds are problem specific. For AI to make a meaningful contribution to protein function prediction, it must do substantially better than current methods, not just better than some arbitrary threshold.

Re: Deep learning gets the glory, deep fact checking gets ignored

#39

Post by bee_rider »

tbrownaw wrote: post #37
amelius wrote: post #2

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.

> For example, give it a paper of some deep learning technique and make it produce an implementation of that paper. Or maybe give it a paper full of statistics about some experimental observations, and have it reproduce the raw data?

Like, have the AI do the experiment? That could be interesting. Although I guess it would be limited to experiments that could be done on a computer.

Re: Deep learning gets the glory, deep fact checking gets ignored

#40

Post by bee_rider »

ojosilva wrote: post #17
amelius wrote: post #2

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.

I thought you were going to say "give AI the first part of a paper (prompt) and let it finish it (completion)" as a validation AI can produce science at par with research results. Before it can do that, I have no hope that it can produce novel ideas.

I guess it would also need the experimental data. It would, I guess, also need some ability to do little experiments and write off those ideas as not worth following up on…
Post reply on HN