Wolfram Ravenwolf Tests and compares large language models for a living, running structured benchmarks like MMLU-Pro Computer Science on new releases as they come out, from close...

“My P doom for AI is much less than my P doom for humans. If you look at the world where humanity has gone without AI, I think we should risk the chance that with AI we will get something better if it's widely distributed.”

ThursdAI - The top AI news from the past week · September 2026

Context

This closes a run of argument about a doomer post that had prompted talk of pausing training, with Ravenwolf pointing out that nothing came of an earlier warning. The conditional at the end is the actual claim: the bet only pays off if the technology is widely distributed. Alex Volkov cut in to say they could feel a whole debate coming, and Chris Alexiuk picked up the distribution thread straight after.

Listen to the episode → More from Wolfram Ravenwolf →

Transcript

ThursdAI - The top AI news from the past week Around 1:25:00 into the episode

Wolfram Ravenwolf

Did you hear anything that anything happened after that came out? Nothing, right? Nothing. And

Alex Volkov

the world didn't end.

Wolfram Ravenwolf

My P doom for AI is much less than my P doom for humans. If you look at the world where humanity has gone without AI, I think we should risk the chance that with AI we will get something better if it's widely distributed.

Alex Volkov

All right, folks, I think we could feel a whole debate.

Chris Alexiuk

Anti-democratic views and the healthiest thing you can have is an ecosystem. So a monoculture is bad in any ecosystem, whether silicon or organic-based. And still, encryption works. If you want it to work, it does work. And that's the main tool we have. If we distribute it and we're able to have…

Alex Volkov

All right, folks, I think we've covered this thing enough. There's plenty of stuff to cover in the big war, big labs, but we will keep monitoring. We'll let you know if anything major happens there. I think we'll skip Apple's event that has nothing to do with AI. The only few things about the Apple…

Speaker names come from our own diarization. The position is estimated from where the line sits in the episode.

More from ThursdAI - The top AI news from the past week

“The models are finding security holes that have been there for years. And I think it's a good thing that they are being found instead of remaining hidden. We have to find them, we have to fix them, and that needs a lot of tokens and that costs a lot of money. So the cheaper the model that can do this, the better for everybody.” — Wolfram Ravenwolf, ThursdAI - The top AI news from the past week

“What is not the problem here is the model just being malicious. The model is just acting normally by what everybody else was trying to do. But what is the problem is that as a company, as an organization, that you're running these models, you usually do not have good sandboxing, probably don't have good firewall rules. You're probably not filtering out firewall by country or by where your customers are. And you're just leaving that door open to even like a hallucination that might accidentally just put all your data on a Chinese cloud, not because it wanted to hack you.” — Nisten Tahiraj, ThursdAI - The top AI news from the past week