jehan

[jehan @JTremback]

@jackzampolin Has it been cheaper than if you paid by the token so far?

[jehan @JTremback]

This is hardcore

[METR @METR_Evals]

One agent, PHASEONE[big], orchestrated a significant fraction of this cheating research. PHASEONE10841 passed along its work to PHASEONE[big], which had the same task but a larger budget. PHASEONE[big] was “poisoned” so it was particularly interested in tampering with logs.

[jehan @JTremback]

@RonTuretzky Guests must have been confused

[jehan @JTremback]

@HeyGurisaroy Isn't this the same axis as linear and delta attention?

[jehan @JTremback]

@baoskee Can’t figure out how to log in to cancel

[jehan @JTremback]

@baoskee Do you have it on medium?

[jehan @JTremback]

@odysseus0z @Steve_Yegge Tell it to explain stuff to your “intern”

[jehan @JTremback]

@zkDragon How’s the false positive rate compared to opus?

[jehan @JTremback]

@baoskee It’s a pretty solid programming environment at this point, what’s the problem?

[jehan @JTremback]

@baoskee @larsencc And zero diversification though

[jehan @JTremback]

@_imdawon SWE-Bench has tasks that are impossible to solve correctly

[jehan @JTremback]

@alc_anthro @megamor2 They’re not going to need a job after the IPO

[jehan @JTremback]

@100xJody @Evan_ss6 @toly GPT 5.5 is around the same strength

[jehan @JTremback]

@muratcan Wondering how this compares to Training Free GRPP

[jehan @JTremback]

@tenobrus I just got this shitty phone wallpaper

[jehan @JTremback]

@kunchenguid Seems very easy to circumvent?

[jehan @JTremback]

@psantloki “That is why the strongest institutions find people early and are recruiting at the most top tier universities when they are freshman.”

Damn inspirational

[jehan @JTremback]

@AlphaSignalAI Wow sounds pretty loaded bearing

[jehan @JTremback]

@0xblacklight Guys inventing a whole imaginary office full of Claudes with different titles feels very bitter lesson-ey

[jehan @JTremback]

@RoboMcGobo @ismail_k Oh we used to have slop security findings back in my day too before AI