darren

･ *ﾟ 　 ･ ﾟ* ･｡ *･｡ *.｡ ｡･ °*. RuntimeError: CUDA out of memory. ｡｡ ･ ｡ ･ﾟ ｡°*. ｡*･｡･ *ﾟ 　 ･ ﾟ*

how small a thought it takes to fill a whole life

Can you pull in Leviathan with a fishhook     or tie down its tongue with a rope? Can you put a cord through its nose     or pierce its jaw with a hook? Will it keep begging you for mercy?     Will it speak to you with gentle words? Will it make an agreement with you     for you to take it as your slave for life? Can you make a pet of it like a bird     or put it on a leash for the young women in your house? Will traders barter for it?     Will they divide it up among the merchants? Can you fill its hide with harpoons     or its head with fishing spears? If you lay a hand on it,     you will remember the struggle and never do it again! Any hope of subduing it is false;     the mere sight of it is overpowering. No one is fierce enough to rouse it.     Who then is able to stand against me? Who has a claim against me that I must pay?     Everything under heaven belongs to me.

@raveeshbhalla sol is right a lot, turns out

@darrenangle Sol is right though

had sol xhigh train 6 writing style loras in the night. woke up to a microsite to blind eval data recipes before we scale one up

@tszzl @ZoharAtkins

Lightcone Infrastructure

@willccbb a 24oz white monster energy drink goes well with these xeets

doing OPSD on terminal bench with "you are an expert principal software engineer at MIT, don't make mistakes" as the privileged info chart looks good here

@tszzl me and my leviathan

@shiraeis no rollback

so openai’s model personality team might be going through something

@Crazymoments01 i like when the cage door vanishes into thin air

When the litter of six came in to Greenfield Animal Care in Sacramento last June, one puppy, the smallest, wasn't responding. Vet tech supervisor Lena Park said they'd done what they could. The mother, Daisy, was kenneled separately. Then at 2:14 a.m., the security camera caught it. Daisy broke out and covered him. By morning he was nursing. Dr. Yusuf Okonkwo called the puppy's survival "genuinely difficult to explain without accounting for her." The puppy, named Dime, is now fully healthy. Daisy knew something the monitors didn't.

The prompt engineering here is super impressive! Such a great example of agent prompting:

there will be a cambrian explosion of models after all. sol will train a lot of them

o lord. getting every request blocked on benign style transfer lora work in codex. tiring.

@julien_c @PrimeIntellect

Who will be the “deepseek of the west”?

We believe enlightenment humanism, scientific progress, open discourse, AI that benefits humanity = we give you a custom model We are 2 months from RSI doom loop = use our API

to arrive ahead of its time is the fate of the fish washed up on shore do you know that he just won't feel right till he's swallowed up by the tide

“Tools refer to one another to finally refer to our care for existing. In turning on a bathroom switch, we open up the entire ontological problem“ - Emmanuel Levinas

At humans&, we train models from the long-term impacts of their interactions with people. This requires prioritizing long-horizon multi-agent RL. We've developed and are excited to share an open-source, hardware-native 4-bit RL recipe, significantly accelerating training

war and battle-related metaphors trigger fable and 5.6’s security classifiers, asking for ‘hostile’ code review on otherwise harmless networking tasks etc. side effect of safety guardrails is softening language and possibly the speaker

We started Thinking Machines a year and a half ago with a couple of instincts: that people should have much more ability to customize models and do research on them, and that even as AI becomes more autonomous, there's a lot more to build to make humans and AIs work well together. A lot has happened since then, especially the massive progress in agents, so we wanted to revisit those instincts in light of everything we've learned, argue about them, and write down what we actually believe now. This is where we landed after a lot of debate. I'm happy with it!