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

121–130 of 174 posts

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

#121

Post by catlifeonmars »

somenameforme wrote: post #112
godelski wrote: post #71

Earlier quoted context omitted.

> Almost nobody is "anti-science". Last I checked: - 15% of Americans don't believe in Climate Change[0] - 37% believe God created man in our current form within the last ~10k years (i.e. don't believe in evolution)[1] I don't think these are just rounding errors. They're large enough numbers that you should know multiple people who hold these beliefs unless you're in a strong bubble. I'm obviously with you in news a…

I think one of the most important 'social values' for science to thrive is a culture with a freedom to disagree on essentially anything. In most of every era where there was rapid scientific progress from the Greeks to the Islamic Golden Age to the Renaissance and beyond, there was also rich, and often times rather virulent, disagreements over even the most sacred of things. Some of those disagreements were well foun…

I don’t think that’s quite right. Disagreement for the sake of disagreement is not particularly meaningful. The basis for science is iteration on the scientific method. Which is to say: observe -> hypothesize -> falsify.

Anti science means to make claims that have no basis in that process or to categorically reject the body of work that was based on that process.

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

#122

Post by kmacdough »

boxed wrote: post #63

Oh look, just what I've been predicting: https://news.ycombinator.com/context?id=44041114 https://news.ycombinator.com/context?id=41786908 It's the same as "AI can code". It gets caught with failing spectacularly when the problem isn't in the training set over and over again, and people are surprised every time.

With "AI can code", though, we can get pretty far by working around the problem. Use it to augment the workflow of a real SWE and supply it with guardrails like linters, tests, etc. It doesn't do the hard bits like architecture, design and review, but it can take huge amounts of the repetitive "solved" bits that dominate most SWEs time. Very possible to 2-5x productivity without quality loss (because the human does all work to guarantee quality).

But yes, unmanaged and unchecked it absolutely cannot to the full job of really any human. It's not close.

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

#123

Post by kgilpin »

I feel the same way about code generation vs code review. Everyone knows there are deep problems with LLM generated code (primarily, lack of repo understanding, and proper use of library functions).

Deep, accurate, real-time code review could be of huge assistance in improving quality of both human- and AI-generated code. But all the hype is focused on LLMs spewing out more and more code.

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

#124

Post by weatherlite »

JackC wrote: post #119
imiric wrote: post #76

It's interesting to see this article in juxtaposition to the one shared recently[1], where AI skeptics were labeled as "nuts", and hallucinations were "(more or less) a solved problem". This seems to be exactly the kind of results we would expect from a system that hallucinates, has no semantic understanding of the content, and is little more than a probabilistic text generator. This doesn't mean that it can't be use…

The difference in fields is key here: AI models are going to have a very different impact in fields where ground truth is available instantly (does the generated code have the expected output?) or takes years of manual verification. (Not a binary -- ground truth is available enough for AI to be useful to lots of programmers.)

> does the generated code have the expected output?

That's many times not easy to verify at all ...

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

#125

Post by ActivePattern »

b0a04gl wrote: post #66

Man, I’ve been there. Tried throwing BERT at enzyme data once—looked fine in eval, totally flopped in the wild. Classic overfit-on-vibes scenario. Honestly, for straight-up classification? I’d pick SVM or logistic any day. Transformers are cool, but unless your data’s super clean, they just hallucinate confidently. Like giving GPT a multiple-choice test on gibberish—it will pick something, and say it with its chest.…

Ironically, this comment reads like it was generated from a Transformer (ChatGPT to be specific)

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

#126

Post by dvfjsdhgfv »

disgruntledphd2 wrote: post #80
ErigmolCt wrote: post #69

Earlier quoted context omitted.

Transformers will ace your test set, then faceplant the second they meet reality. I've also done the "wow, 92% accuracy!" dance only to realize later I just built a very confident pattern-matcher for my dataset quirks.

Honestly, if your accuracy/performance metrics are too good, that's almost a sure sign that something has gone wrong. Source: bitter, bitter experience. I once predicted the placebo effect perfectly using a random forest (just got lucky with the train/test split). Although I'd left academia at that point, I often wonder if I'd have dug in deeper if I'd needed a high impact paper to keep my job.

I believe it's very common. At some point I thought about publishing a paper analyzing some studies with good results (published in journals) and showing where the problem with each lies but at some point I just gave up. I thought I will only make the original authors unhappy, everybody else will not care.

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

#127

Post by tough »

ActivePattern wrote: post #125
b0a04gl wrote: post #66

Man, I’ve been there. Tried throwing BERT at enzyme data once—looked fine in eval, totally flopped in the wild. Classic overfit-on-vibes scenario. Honestly, for straight-up classification? I’d pick SVM or logistic any day. Transformers are cool, but unless your data’s super clean, they just hallucinate confidently. Like giving GPT a multiple-choice test on gibberish—it will pick something, and say it with its chest.…

Ironically, this comment reads like it was generated from a Transformer (ChatGPT to be specific)

its the em dashes?

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

#128

Post by hbartab »

AIs can write code in seconds, but you may have years of regret _if_ you believe whatever it spits out without verification. The cold-war maxim "trust, but verify" is truer than ever.

The danger behind usage of LLMs is that managers do not see the diligent work needed to ensure whatever they come up with is correct. They just see a slab of text that is a mixture of reality and confabulation, though mostly the latter, and it looks reasonable enough, so they think it is magic.

Executives who peddle this nonsense don't realize that the proper usage requires a huge amount of patience and careful checking. Not glamorous work, as the author states, but absolutely essential to get good results. Without it, you are just trusting a bullshit artist with whatever that person comes up with.

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

#129

Post by tough »

dvfjsdhgfv wrote: post #126
disgruntledphd2 wrote: post #80

Earlier quoted context omitted.

Honestly, if your accuracy/performance metrics are too good, that's almost a sure sign that something has gone wrong. Source: bitter, bitter experience. I once predicted the placebo effect perfectly using a random forest (just got lucky with the train/test split). Although I'd left academia at that point, I often wonder if I'd have dug in deeper if I'd needed a high impact paper to keep my job.

I believe it's very common. At some point I thought about publishing a paper analyzing some studies with good results (published in journals) and showing where the problem with each lies but at some point I just gave up. I thought I will only make the original authors unhappy, everybody else will not care.

Peer Review is a thankless job

but that’s how science advances

there should be an arxiv for rebuttals maybe

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

#130

Post by stevenae »

TeMPOraL wrote: post #113
stevenae wrote: post #110

Earlier quoted context omitted.

> Lately, I just steal embeddings from big models and slap a dumb classifier on top. Works better, runs faster, less drama. You may know this but many don't -- this is broadly known as "transfer learning".

Is it, even when applied to trivial classifiers (possibly "classical" ones)? I feel that we're wrong to be focusing so much on the conversational/inference aspect of LLMs. The way I see it, the true "magic" hides in the model itself. It's effectively a computational representation of understanding . I feel there's a lot of unrealized value hidden in the structure of the latent space itself. We need to spend more time…

For this and sibling -- yes. Essentially, using the output of any model as an input to another model is transfer learning.
Post reply on HN