bmp

human undergoing rl to be truth-seeking on twitter

[bmp @basemodelpolicy]

oai must have better metrics for measuring research acceleration (esp if RSI is on the research agenda now). why not share any of those instead of $/researcher

[Noam Brown @polynoamial]

One of the most interesting blog posts we've released: details on internal research acceleration at @OpenAI. I expect these trends to continue.

We also share some details on how we've paced model development to prioritize monitoring, alignment, and security.

[bmp @basemodelpolicy]

you can raise funds by finding bettors to take the other side (aka that it won’t be the Last Party) each year!

[Aella @Aella_Girl]

does someone with a lot of $ wanna fund me throwing a yearly Last Party, since we're at the end of the world?

[bmp @basemodelpolicy]

@davehpitt @luke_metro is it bad that ive only ever used cursor, claude code, and codex? never opencode

[bmp @basemodelpolicy]

@AgustinLebron3 maybe @andrewho03

[bmp @basemodelpolicy]

i can’t believe there are people who legitimately believe OpenAI will try to solve cancer (unclear how they would do this) when they could generate trillions in value making reliable knowledge work assistants

[Varunram Ganesh @varunram]

I can't believe there are people who legitimately believe OpenAI (and Anthropic but lesser so) will choose to focus more on bug bots or some incremental app layer work instead of curing cancer, solve aging, insane math problems, etc.

The scale of ambition with AI is so much higher and you could do so many things. App layer is a great opportunity for startups to build awesome businesses while the labs focus on much bigger, existential problems for humanity. Its not labs versus startups, its labs and startups

[bmp @basemodelpolicy]

yay this is good! regardless of singularity/rsi/etc predictions, cybersecurity is now close enough that the oncoming dangers seem obvious/guaranteed and need to be addressed

[fouad @fouadmatin]

Alongside the launch of GPT-6 Astra, we're committing $1 billion to subsidize Daybreak access and frontier capabilities to help frontline defenders protect essential services and critical infrastructure.

Welcome to the AGI era for cybersecurity.

https://t.co/kwiH5rAEzY

[bmp @basemodelpolicy]

@jxnlco probably voice controlled and running in the background always, so desktop app i guess

[bmp @basemodelpolicy]

if you ask they can rationally make a case for it but they don’t _believe_ it

[bmp @basemodelpolicy]

i think a lot of them don’t *really believe* deep down that they’re going to actually bring full autonomy to the world

[Paras Chopra @paraschopra]

It's kind of crazy that SO MANY smart people have a goal of bringing full autonomy to the world, when it is clear that it'll leave everyone jobless.

You ask these smart engineers or researchers if they've thought about consequences of their actions, and they'll simply shrug.

[bmp @basemodelpolicy]

@andrewho03 how much do employees at these labs believe in short timelines vs just don’t think about it that much and are heads down working?

[bmp @basemodelpolicy]

seeing a lot of vague openAI hype like “AGI is here” and no explanation of why this isn’t another point on the continuous (and fast) improvement of models

[bmp @basemodelpolicy]

interesting observation, oai has historically had the mandate when it comes to perf scaling with reasoning effort..

[eternalist @eternalism_4eva]

bruh... non-monotonic performance in juice...?
from OAI?!

this is mandate-of-heaven losing stuff here what is going on

I am going to assume ongoing learning pains with looping. noam will have this fixed up for 6.1 I'm sure...

[bmp @basemodelpolicy]

seems like an extreme exaggeration of where we are today (not saying anything about the future though)

[Soren Kierkegaard @BumrahBachi]

I think the music has stopped

The quoted tweet is from february this year but progress has been such that at this point I think your career is meaningless if you don't own substantial capital, do something physical or work at OpenAI or Anthropic.
https://t.co/oCQaoy13nq

[bmp @basemodelpolicy]

what is the correct interpretation of why there seems to be so much talent flow to anthropic, even in the last few weeks?

[bmp @basemodelpolicy]

@DeepDishEnjoyer fair enough

[bmp @basemodelpolicy]

@davehpitt @luke_metro i haven't heard many people talking about muse code but if i did i would definitely try it

[bmp @basemodelpolicy]

@DeepDishEnjoyer *talking

[bmp @basemodelpolicy]

@DeepDishEnjoyer wasn't talk about that one, i was talking about this one:
please be less vague about this one i am dying to know
https://t.co/0wG6yBhAy1

[peepeepoopoo @DeepDishEnjoyer]

i have 100% certainty that the next models have a step change in capabilities. i don't know when they're going to be released - or even that they will be publicly released - but it's not just cryptoesque hype. if your mental model of ai is still sora slop you don't have a clue

[bmp @basemodelpolicy]

@dwarkesh_sp bernie is probably net good for ai risk by dragging the overton window behind him? that’s kind of his thing historically it seems like

[bmp @basemodelpolicy]

@jachiam0 when evals saturate, so do rl envs, and so does progress?