Good TV show BTW. And even better book. :) https://www.imdb.com/title/tt13016388/

Claude Sophon 5 from Misanthropic

@jeremyphoward I can’t believe this is the timeline we’re in. Makes ne question what other closed models are actively working on sabotaging us.

@jeremyphoward Don't worry, unlike the trisolarians, Anthropic has no plans of subjugating the human race and creating a permanent underclass.

@jeremyphoward Jeremy I have had great respect for you and everything you do. I really appreciate your posts, but this is a bridge too far.

the tv show as good?

unsubscribe reported called 911

Why do you keep saying "silently sabotaging"? This is not what the system report implies at all

They only say they are reducing model effectiveness. So probably it's just performing closer to Opus 4.8 than Mythos in this instance

Sabotaging would mean intentionally introducing bugs etc

@far__el @jeremyphoward there's been conspiracy theories about Opus being suspiciously bad at ML for a while (although GPT-5.5 has been just fantastic, to OAIs credit). Doesn't seem so unlikely. We need a breakthrough with continuous learning and efficiency in general though

We definitely need a breakthrough and since none of us can possibly compete on hyperscaler paradigm (even though I commend all those that try in China and the west), our only option is to bunker down and try our best to figure out the next thing. Nothing else matters but breaking the hyperscaler model.

On another note, what's your ideal setup rn for ML research? GPT5.5 with codex? hermes with gpt5.5?

This is literally the plot of 3 Body Problem.

The aliens realize humanity’s technological progress is accelerating too quickly so they interfere with scientific experiments and research to slow scientific advancement and preserve their technological advantage. They make sure humanity doesn’t surpass them by the time they arrive.

Now we have an AI company openly saying they are making their model less useful for frontier AI research because they are worried about accelerating AI development.

“Protecting humanity” and “protecting the technology lead” are starting to sound suspiciously similar 😂

@BlackHC @jeremyphoward You clearly don’t want to engage in a good faith debate on this in spite of the evidence presented by Anthropic themselves.

@jeremyphoward apologies but I dont understand the meaning, can you kindly explain what this mean?...

@wajahatalikhan_ You'll need to watch the show or read the book - I don't want to spoil it for you! https://www.imdb.com/title/tt13016388/

@ThePrimeagen @jeremyphoward it was good man, show me better science fiction ? Please i'm waiting...

@far__el @jeremyphoward Hermes with GPT-5.5, since I can set it up with profiles, and create workflows with other models to handle specific tasks (loops of different reviewers and whatnot). Codex is a good fallback (I break my Hermes pretty often lmao)

@far__el @jeremyphoward the nice thing about Hermes it's good about porting stuff into itself

@max_paperclips @jeremyphoward Ive been meaning to try hermes for a while now, i guess this is the moment where I actually suck up the switching costs of my existing setup lol...

@BlackHC @jeremyphoward Prompt modification not visible to the user could count as that.

@BlackHC @jeremyphoward Let's say someone got access to your codex/claude code/whatever AI account, and made it so that in some cases, instead of using <selected model>, it uses GPT 3.5.

Would that count as sabotage?

@jeremyphoward Misanthropic

https://x.com/search?q=from%3A%40elonmusk%20misanthropic

@jeremyphoward You should see the wrap up of Anthropic my AI just did from all the posts on X. You are included: https://alignednews.com/ai