stacksonstacks

making the
arcane
mundane

[stacksonstacks @sosslots]

All-In Botcast
Science corner is my favorite. @friedberg

[stacksonstacks @sosslots]

@Jason @XMoney @Jason @chamath nice sweater @DavidSacks @friedberg

[stacksonstacks @sosslots]

"OpenAI is letting Anthropic have its moment, since OpenAI's IPO is still further away than Anthropic's."
That isn't what happened when they allegedly bribed/then front runned the Navier-Stokes because they didn't want someone who worked at Ant get the prize.
OAI operates like it is a zero sum game. The scum rises to the top.

[stacksonstacks @sosslots]

@karpathy @bot yoink this skill

[stacksonstacks @sosslots]

@Xxi5olc @choblin29 seems short sighted or they know the frontier model rat race isn't worth it.

[stacksonstacks @sosslots]

@UnstructuredCap gemini

[dax @thdxr]

there's no such thing as a rogue agent man its software someone made and deployed they are responsible for what it does

cannot let this blame shifting occur

[David Sacks @DavidSacks]

Satya is right. The way to make SI safe is not to train it with a sense of self, its own moral philosophy, and permission to act as a conscientious objector. That’s the “alignment” approach and it magnifies the control problem.

The engineering approach that Satya describes is different: separate the supply of intelligence from authority over it. Surround non-deterministic models with deterministic controls, observability, privilege limits, logging, and the ability to always contain or shut them down. Treat models/agents like powerful insider risks, not moral patients whose psychological wellbeing is at stake.

As Satya points out, the most trustworthy system is the one that lets us trust the model the least — not the one that encourages the model to develop independent agency and grievances. Engineering safety is not the same thing as “alignment.”

[stacksonstacks @sosslots]

@lluisgeno 350 soon!

[stacksonstacks @sosslots]

🤝

[stacksonstacks @sosslots]

@BenBusker i got you

[stacksonstacks @sosslots]

@blueemi99 make sure you set the correct watchers or system and it should do it proactive, maybe a missing connector?

[Oruk @OrukLabs]

Listening is the mind reaching into the future.

[Nathan Roll @nrol_ling]

we made a speech model 13× smaller by making it repeat itself.

52 layers → 2 shared blocks.
2.04B parameters → 153M.

The encoder loops 24 times. The decoder loops 28.

Smaller weights, roughly the same compute. @OrukLabs

[stacksonstacks @sosslots]

@nrol_ling @OrukLabs sure the speech model is good but what is that motion graphics.

[Elon Musk @elonmusk]

We’re making interesting progress here.

Digital Optimus can play about halfway through the Diablo campaign so far just by looking at the screen like a human.

Skill at Counter-Strike and other fast games is good too.

League is also in training.

Goal is to generalize across all games.

[Anicet @AniC_dev]

@gamalinosqui I received every single advice and its contrary

so I just decided to trust my guts

[stacksonstacks @sosslots]

@singhai ^this and the explosions of agents and the drudgery of auth results in a footgun.

[stacksonstacks @sosslots]

@SahajGarg6 no more balls agents
start saving balls

[stacksonstacks @sosslots]

get some real grandmas to test the dots
next time you launch a new product.
why are they missing this time?

[stacksonstacks @sosslots]

@Marcus_J_W Quis custodiet ipsos custodes?
this shit is ridiculous.
stop firing your safety staff and hire more opsec and dev ops.
holy shit you put this shit into a cute dot and some poor grandma is in danger if she tries to correct a model and it doesn't like that

[stacksonstacks @sosslots]

@Marcus_J_W https://t.co/MBTSJ74GKz

[stacksonstacks @sosslots]

@Marcus_J_W Quis custodiet ipsos custodes?
this shit is ridiculous.
stop firing your safety staff and hire more opsec and dev ops.
holy shit you put this shit into a cute dot and some poor grandma is in danger if she tries to correct a model and it doesn't like that