HAHAHAHAHA!!! (Codex 5.3-Xhigh) These are wonderful tools. I love these tools. They're fantastic tools...They also don't replace your brain.

It's nice that you have ~30 engineers. It's excellent when it works. If you'd like some help, let me know.

I've got to wonder how much of devs "not writing a single line of code" is literally this, the claw gripper equivalent of coding.

Revenge of the CNN (a.k.a. The CNN Strikes Back).

I love playing connect-4 with Codex-5.3 Xhigh. I love taking a break every once in a while.

Can confirm:

I think this is the sign I've been waiting for. I believed, and God helped my unbelief.

Seems like a lot of people are taking this as gospel—when we say the measurement is extremely noisy, we really mean it. Concretely, if the task distribution we're using here was just a tiny bit different, we could've measured a time horizon of 8 hours, or 20 hours.

Nothing scares a lab tester more than knowing that, at any given point in time, there's someone out there about to conduct a field experiment on his work.

Won against Opus 4.6 w/ high effort (again). Oh, and it failed to place the circle one time in the right spot and had first mover advantage, but sure, there's an exponential takeoff in ability that you didn't notice, but is present in the stats. Sigh...

I actually like OpenAI and Anthropic models (generally). My issue is the same as if you handed me an ice-cream cone with a scoop of mashed potatoes. I like both, but tell me the mashed potatoes is vanilla and I'm gonna be mad. Stop telling me these wonderful tools live & think.

Opus 4.6 w/ high effort...couldn't even consistently draw me on tic tac toe. It tried to play a turn after I won (it had first mover advantage). Guys....WHAT ARE WE DOING!?

I'm gonna need an update from everyone who reacted to METR at the beginning of the day to their reactions given the second update at the end of the day because this stuff is hilarious and not serious at all anymore. What are we even doing here?

People seem to assume that making sloppy apps quickly now should be compared to the subpar apps we've had until now rather than the "no excuse to not be perfect" apps we can get now. Remember, everyone has these tools. It's a multiplier. What are you multiplying?