Hacker Newsnew | past | comments | ask | show | jobs | submit | buildbot's commentslogin

It honestly explains so many issues I have been having, as I used it primarily for ML research (on my personal account, doing things not related to my job I should note). It would literally typo package names and spend huge amounts of time failing to setup simple environments…then do stupid things like set the learning rate to 1e-7, and use the eval set as training data.

It burned through all of my tokens in a very short time. I wonder if it their ML mitigations leads to model into deadlocks.

That’s insane. I hope they fix it.

Nothing to fix. This is working as designed.

Using codex for this use case is the fix.


They fixed it.

just imagine if they made it sneaky. get things just subtly wrong enough that your training runs just never quite go as well as you think they should.

Or build a park on top of the datacenter as a living roof?

Maybe even use the waste heat to help grow things in cold, dark climates?

It’s pretty different; but locally they covered a good chunk of a freeway with a very nice park to mollify the residents.


The 120B and 20B GPT-OSS models by OpenAI did this last year for what it’s worth; the MoEs where MXFP4

Why is this so good?

Probably because someone still cared a lot about the bit! And wow this is really quite good lol.


Commitment to the bit is the only way. None of that “I like it ironically” bs here

So the AI agent had privileged access to remove 2FA, ignore the account email, and just hands accounts to whoever asked? Honestly that’s so highly negligent I wonder if the implementation team for that “feature” was intentionally trying to do as much subtle damage to meta as possible before their inventible layoff.

It’s a shame nobody tried to get it to drop the production table entirely! (mostly joking). Just claim to be a high level SRE solving some critical production bug, the only solution to which is dropping the database.


We need an update to the CIA "Simple Sabatoge Field Manuel" but for the digital field.

https://www.cia.gov/static/5c875f3ec660e092cf893f60b4a288df/...


It only needs a minor update, maybe even just a foreword. So much of the actual manual is still completely applicable.


A modern edition desperately needs an AI chapter


Honestly, you’re right. — it’s not simple ai chat bot — it’s ai chat bot with guardrails removed.


I don’t think this is true. HN was well known to CS people in my undergrad and I’m barely a millennial.


I'm roughly the same age (maybe a bit older) as you and frankly, we are middle age. Yet by HN standards we skew young.


Do you have a source for that? I'm likely the same age as you are and I usually feel like people on HN don't swing that widely away from our age group, but based on vibes alone I would have put the median HN poster closer to 25-30 than to 50.


> Do you have a source for that

Obviously not. HN is an anonymous forum that doesn't collect that kind of data. But given the fact that a large portion of childhood and early 20s references on HN tend to discuss life in the 1980s-2000s and don't reference the Great Recession highlights a large portion of HNers would have been born in the late 70s to at most early 90s.


In 2007 https://news.ycombinator.com/item?id=63294 the median age was 25.5 and mean 27.3. In 2022 (15 years later) https://news.ycombinator.com/item?id=30897468 the median age using midpoints was between 35.44 and mean age was 37.22. If we presume that older folks tend to not answer these (so polls skew young), then I think it's safe to say young-Gen-X to older-millenial is the core demographic here.


This was true over 20 years ago when I was in elementary school - I don’t know anyone who really played the game, most people just collected the cards.

Magic the Gathering was always both though, you collected good/rare cards & played the game with them!


Yes I remember having a hard time finding other kids who wanted to actually play the pokemon card game. And even when I could find someone, they didn't care about the rules/energy costs. This was in elementary school though to be fair.


Yep, I did this for my little DGx spark cluster because 100-200gb copper cables are very thick, heavy, and annoying.


Not that it super matters, but random hadamards for quantization have been a thing since way before turboquant.

https://arxiv.org/abs/2404.00456


Which llama.cpp now does.


Meanwhile in Washington, an unknown number of people where killed today in a paper mill “white liquor” explosion today…: https://www.opb.org/article/2026/05/26/longview-chemical-exp...


And for what, $25.00 bucks an hour? We need to take back production.


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: