LLM research on Hacker News is drying up

https://dylancastillo.co/til/llm-research-on-hacker-news-is-dying.html

by dcastm • 3 months ago

30 11 3 months ago

simonw2• 3 months ago

Hacker News isn't a great place to discuss papers generally.

Having a productive discussion around a paper requires at least reading and understanding the abstract, and the most successful content on HN (sadly) is content where people can jump in with an opinion purely from reading the headline.

Anyone know of any forums that are good for discussing papers?

irishcoffee• 3 months ago

I just email the authors with questions. Surprisingly high response rate.

tptacek• 3 months ago

This is true across all research subject areas (I'm not especially tuned into LLM research but am to cryptography, which also happens to be a field that gets a lot of play on HN). I think it's just a function of how many people conversant in the field are available to talk about it at any one time.

gessha2• 3 months ago

There are/were isolated communities on Discord around fast.ai, MLC, MLOps that talk papers more in depth but it’s hard to organize a community without commercial or academic incentive.

fc417fc802• 3 months ago

The difficulty is perhaps unsurprising given the time sink that is reading a given paper to any reasonably complete degree of understanding.

gcr• 3 months ago

/R/MachineLearning is not bad

But the gold standard is a small signal or discord community of like-minded, fairly tight knit friends. You may have to organize this yourself

gavinray• 3 months ago

Unironically, very niche subreddits.

simonw2• 3 months ago

... and this thread over here seems to be proving me wrong already: https://news.ycombinator.com/item?id=47893779

gessha2• 3 months ago

With big commercial labs clamming up about training details, hardware requirements going up, and overall tired sentiment about AI in general, that’s not really surprising.

ML research shows up on the front page if it shows splashy claims or results. Mundane cockroach papers that advance the field one nudge at a time aren’t that interesting for the average reader.

Cool to see the sentiment visualized though.

latexr• 3 months ago

So I asked Claude (…)
I asked Claude (…)
So I asked him (…)
(…) so I asked Claude (…)
(…) so I asked Claude (…)
Thanks, Claude.

There doesn’t seem to be one bit of original research in the post, and no explanation of the data or conclusions. For example, what exactly does it mean that the papers “held up”, and how exactly did Claude reach that conclusion? If you don’t know, we can’t trust the data. If you do know, it should be in the post.

As it is, the post is almost devoid of information. Everything was “I asked Claude”. There’s no value here (aside from some saved tokens) above just crafting a prompt and saying “here, ask your favourite model this question”.

dormento• 3 months ago

And it has a citation block at the bottom! Like anyone would want to cite this in a serious context.

This must be a troll.

LLM research on Hacker News is drying up

Related Stories