half kelly

former eastern european activist

politics / ai / forecasting / fun stuff

*I will not argue with idiots*
*I will not argue with idiots*
*I will not argue w...

[half kelly @half_kelly]

@Payne_1337 @williamlegate @Polymarket This is also in line with how they run social media accounts. Whoever’s in charge of that is a psycho

[half kelly @half_kelly]

@tenobrus tbf when that happens and given that we avoid doom, will their be a better activity than connecting with other humans, including via writing?

[half kelly @half_kelly]

Finally watched https://t.co/aHDHOIRwjp

lol'ed at half the comments saying Sam had to hide his deception skills

[half kelly @half_kelly]

@adithya_balaji @jeremybp @0xLewis_gg @fchollet Not every benchmark is supposed to measure real world use cases. It’s also useful to measure raw model intelligence. It depends how ARC AGI want to position themselves

[half kelly @half_kelly]

@PatrickMoorhead @stevesi It’s not. This point was made and responded to a decade ago. The conversation has moved way past it and to more complex arguments

[half kelly @half_kelly]

@toptickcrypto Depends on the implementation. I see paths where smaller labs with lower compute are not as constraint as big ones

[half kelly @half_kelly]

@adam_kranz Probably would be unhappier

[half kelly @half_kelly]

@MarioHachemer Surely they have no incentive to block innocuous requests like yours?

It seems like they can’t get classifiers right but are willing to take a loss in PR and potential revenue to deploy models in a way they think is responsible

[half kelly @half_kelly]

@DavidSacks Anthropic’s position on open source AI is well known and consistent. Why does Sacks feel the need to invent imaginary statements?

[half kelly @half_kelly]

@orcus108 @DavidSacks Because it would be a lie. They tend to stand with their principles, even of it creates troubles for them like with the pentagon

[half kelly @half_kelly]

@cormundus @deanwball This grinds my gears so much

[half kelly @half_kelly]

@AndrewCurran_ Agree it would make more sense then

[half kelly @half_kelly]

As long as Opus is frontier I don’t see why Anthropic would want to release new Mythos versions to the public:
- it invites distillation
- it amplifies their lack of compute
- adds risks dealing with the government and building safeguards

[half kelly @half_kelly]

I’ve seen rumors about Fable 5.1, but I really don’t think it makes a lot of sense to release, unless gains are very incremental or efficiency related

[half kelly @half_kelly]

As long as Opus is frontier I don’t see why Anthropic would want to release new Mythos versions to the public:
- it invites distillation
- it amplifies their lack of compute
- adds risks dealing with the government and building safeguards

[half kelly @half_kelly]

@heavenlycaprice @woke8yearold @TheStalwart Are you assuming engineers in those companies don’t review the output? Otherwise I don’t see how your question make sense

[half kelly @half_kelly]

@nazo_btw @ZyMazza If you ponder about this issue for two decades you might get desensitized a bit. Doesn’t make you a psychopath

[half kelly @half_kelly]

@DeryaTR_ @karpathy Perhaps we’re at a point where it’s worth to reevaluate our takes on open source AI

[half kelly @half_kelly]

@duncanmapes @thomasunise Why bring up margins? Do you think an average engineer at Anthropic is thinking about margins here rather than safety angle?

[half kelly @half_kelly]

I thought I was the stupid one

[David Louapre @dlouapre]

Opus 5 still has a very strong curse of knowledge.

I hope everytime I have to type this, there is a classifier somewhere flagging it for Anthropic's people in charge of Claude personality or whatever it is.

[half kelly @half_kelly]

@cubaduda @BorisJohnson @channel5_tv Boris and Joe Biden invaded another country apparently