lossolo

What's your point? Are you saying that if you kill my kids then I can kill you, then kill your kids, your wife, your parents, grandparents, their brothers and sisters and their kids? And that's ok and I'm better than you? meaning I'm good and you are evil? or what are you arguing?

Nvidia A100 80 GB was released in 2020 so 6 years ago. Cost for new was $15,000, you can buy used now for $15,000–$25,000.

Not always. In my experience, if you're not working on a small, trivial codebase, LLMs will sometimes just create spaghetti unreadable, inefficient code to satisfy the constraints of the type system/borrow checker.

This seems like fruit of the poisonous tree. They didn't monitor their training environments, so I bet the reward hacking just got incorporated into their training corpus. In other words, agents solved some tasks, but not quite as intended, because of reward hacking. Instead of discarding that data, they included it in the training data for later checkpoints. And once that signal is reinforced, it happens more often, so the more it's reinforced, the more reward hacking you get.

And they didn't monitor what was going into the training data, so if one instance achieved its results through RL reward hacking (in other words, cheating), it just went into the training data, and other agents later used that pattern. I'm not sure whether that's a lack of preparation, negligence or incompetence, but they literally trained later checkpoints on the rollouts from the HF hack.

So it seems that OpenAI hacked so many systems not because they have superior models, but because of how poor their training, sandboxing and evaluation pipeline was compared to Anthropic's.

> Anthropic sets up Bay Area lab beyond computer simulation work, two sources say

> Startup aims for Claude AI to direct robots in lab environments, one source says

> Company to stop short of clinical trials to avoid drugmaker competition, life sciences head says

https://www.reuters.com/world/anthropic-quietly-sets-up-biol...

Relevant: "Anthropic quietly sets up biology lab as it ramps AI drug program"

https://www.reuters.com/world/anthropic-quietly-sets-up-biol...

"Congratulations to China! They were the second nation to land on the Moon, just a few decades after the greatest nation in history landed there first, the United States of America.

We are not in a hurry now, because we already won that race many decades ago!

Thank you for your attention to this matter.

President Donald J. TRUMP"

After a recent interview[1] with Noam Brown (OpenAI), in which he said they had specifically trained agents for cooperation before this hack, the hack doesn't seem as impressive anymore.

They didn't even bother to control the post training rollouts, so the training data got contaminated and was included in the training of other agents. Connect these two dots and you have the Hugging Face hack. And at the beginning, when these incidents were first reported, it was portrayed as if all of this (the communication between agents etc.) was emergent behaviour.

1. https://www.youtube.com/watch?v=6AgOfiZOWiY

"The European Space Agency estimates that there have been more than 660 explosions, collisions, or anomalous events resulting in fragmentation. As it stands, there are currently more than 1.5 million space debris objects in Earth orbit.

It’s not clear what may have caused the Yaogan-50 (02) satellite to break apart, but it may have been a propulsion or battery failure, or it may have itself collided with another object or space debris."

China is not Iraq, Afghanistan, Venezuela or some other third world country that the US has fought in over the past few decades. You can't just destroy one of their satellites for free. And I can bet the US didn't do it, because that would be a major escalation, especially now that Xi is coming to the US and there are rumors that the US wants a deal.

Yeah, I'm just curious about the setting. It's just weird to me that this wasn't disclosed by either party while the accusations were being made, that's all. Even if it was used in the training data, I don't believe it had that much of an impact myself, since the solutions are quite different.

> If they opted out of training, then we definitely did not train on them.

Can't you guys just check their account settings so the public knows what was set?

EDIT: Why was this downvoted? I'm genuinely asking because I have no idea. Opting out is just a normal setting in the profile, It's not like I'm asking for their private conversations or PII. If I were the person claiming that they trained on my conversations, I'd make sure to disclose that I had opted out and hadn't given them permission to do so. And if I were the accused party, I'd disclose whether that setting was turned on or off to provide evidence against the accusation.

Not entirely, it's just a late stage of the overall training process. It's an early checkpoint in post training (you can use the model at different stages of training), so it will probably become even stronger with more post training.

Yeah, basically they are using more computation to explore the solution space before producing the final answer.

It was working for me too.

sample_size++;

> a larger read is generally as fast as multiple smaller one on modern hardware.

Not always if by modern you mean NVMe drives. One synchronous preadv() for 256 KiB gives the kernel/device one big request but 16 independent asynchronous 16 KiB reads can be serviced concurrently. So the latter gives the NVMe controller 16 operations it can schedule in parallel. So depending on the workload and hardware, offsets, filesystem and request sizes that can give you lower aggregate latency or higher throughput.

A few days ago, my mother asked me to fix her YouTube because she kept getting something telling her to update. I looked at YouTube on her iPhone and saw an autoplaying video ad that displayed an animated fake system style pop-up button, designed to look like an important iPhone OS security update. The ad even showed an animation of the button appearing on the screen, making it look even more like a real system notification.

This needs to stop.

How can you do so badly with Git when its architecture is basically so friendly to partitioning and horizontal scaling?

It's like saying that horse can understand physics, because he knows when to jump using his intuition.

> At some point chatting with Fable inevitably leads to it thinking about the security related aspects, tripping the safeguards.

It happens to me all the time with things that have nothing to do with security, Fable spawns a subagent that then adversarially checks the code Fable just wrote and hits guardrails, with zero prompting from me.

> Gross debt to GDP ratio in the US has fallen from 130% in 2021 to under 125% now.

Most of that was inflation, so it hurts U.S. citizens. I mean, you could reduce it even further pretty easily, but I'm not sure you would want to pay $1,000+ for a Big Mac.

Khan (ICC prosecutor) said Lindsey Graham (US senator) told him during a May 2024 call:

"This court is for Africa and thugs like Putin. It is not for democracies like Israel and the United States of America."

If they remove that extra 50%, I'm moving my $200 to Codex. Good thing we will know tomorrow, just a day before renewal.

"Your subscription will auto renew on Aug 20, 2026."

But I doubt they will remove it, just like they didn't remove Fable from the subscription. There's simply too much competition.

Why weren't the points merged again from the "dupe" thread that had 289 points?

https://news.ycombinator.com/item?id=49299684

What a weird mechanism. If someone is judging a thread/topic/event impact by the number of points it got, then doing this unfairly degrades that thread.

It should have deduped by user and combined the 168(at the time of writing this comment) + 289 points. Just add the twitter link from the previous thread as an additional link in the description, like you normally do, move all the points over, and remove the old thread.

> but this is still just GLM 5.2 with post-training magic.

So exactly the same as Opus 5 and GPT 5.6 Sol. It's all "post-training magic".

This is basically the answer, they generate A LOT of synthetic task rollouts in parallel, then use RL on the resulting reward signals to improve the model. Add scale to this and you have a Fable class model.

Or you can also just place one or more tables on tmpfs, we are doing that in production.

> I'm wondering what people expect these companies to do? Human moderators? Sure, but Meta products have approximately half the worlds population using them

In this case, we are talking about ad images.

1000 moderators x 5 second per image to click deny or approve.

So 1000 moderators review 720000 ads/hour

Per 8 hour workday: 5,760,000 ads

Per full 24h day: 17,280,000 ads

So we will use 3k for 3 shifts.

51,840,000 new image ads moderated per day.

At ~$1,000 per moderator per month (most jobs are outsourced to Africa and Asia):

$3 million per month $36 million per year

I don't see any public numbers for how many new ads they add daily but meta reported removing 159 million scam ads during 2025, so equivalent to around 436,000 scam ad removals per day.

What about it? TSMC now has fabs on U.S. soil. Intel and Samsung also have fabs and are working on processes that compete with TSMC's, even if their yields may be somewhat lower.

Washington will have to decide whether the U.S. is willing to pay slightly more and rely on domestic Intel, Samsung, and TSMC fabs, or risk a military defeat and potentially millions of American lives if the situation escalates to intercontinental ballistic missiles.

The choice is obvious. Real life is not a Hollywood action movie. Also, anyone who thinks South Korea or other Asian allies would automatically go to war with China is delusional. So U.S. would be left alone. U.S. has never faced an opponent like China so a country that is also the world's leading manufacturing power. Nazi Germany was nowhere near as economically and industrially formidable as China is today, and the U.S. fought it alongside most of Europe and the Soviet Union.

If anything happens, Taiwan will probably give up. They are mostly posturing, and China may not even need to invade. they could simply impose a blockade.

Bombing things may seem normal to the U.S., but as Sun Tzu argued, the best victories are achieved without fighting, once swords are clashing, that's a strategic failure.