Deep learning gets the glory, deep fact checking gets ignored • Hacker News Board

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

#171

Post by lesostep » Sun, Jun 08, 2025, 6:17 PM UTC

davidclark wrote: post #140

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.…

I’m not sure anyone I know could make an em dash with their keyboard off the top of their head. [meta] Here’s where I wish I could personally flag HN accounts.

Alt+0151 — Alt+0150 –

I used to write documents and it just felt wrong to knowingly use the wrong dash. The habit stuck.

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

#172

Post by socksy » Tue, Jun 10, 2025, 6:03 AM UTC

reaperducer wrote: post #159

davidclark wrote: post #140

Earlier quoted context omitted.

I’m not sure anyone I know could make an em dash with their keyboard off the top of their head. [meta] Here’s where I wish I could personally flag HN accounts.

I’m not sure anyone I know could make an em dash with their keyboard off the top of their head. I have endash bound to ⇧⌥⌘0, and emdash bound to ⇧⌥⌘=.

If you're using macOS anyway, surely that's more complicated than the built in ⌥- and ⇧⌥-?

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

#173

Post by Breza » Tue, Jun 10, 2025, 5:03 PM UTC

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.…

I have a tangential question: how useful do you find SVM in production? I've generally found it to be too slow compared to other algorithms to be worth it.

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

#174

Post by naasking » Fri, Jun 13, 2025, 1:08 PM UTC

> However, if you do the painstaking work of combing through someone else’s published work, and discovering that they are riddled with serious errors, including hundreds of incorrect predictions, you can post a pre-print to bioRxiv that will not receive even a fraction of the citations or views of the original.

This is not a unique phenomenon to AI, a study a few years before AI exploded showed that sensational but wrong papers had a higher chance of being published in top journals, and were cited more than papers that showed they were incorrect. This is an outcome of the bad incentives in science publishing these days.