We’ll pay you up to $100K for every company you refer that licenses its data to @Mercor. We’ve partnered with > 500 companies so far. If you know a company with 100+ employees and data in code, gaming, manufacturing, healthcare or professional services, refer them at data.mercor.com/refer.

Some observations on 'autoresearch', in our experience: 1) 'make this thing faster': work wells. 2) 'improve this frontier training run' deeply unimpressive. 3) 'i want to catch up to near frontier' works fine. (2) is obviously the dream, basically = RSI. We don't have that yet, clearly. (1) can at times be (2), although generally its not been for us, so it works (for code not at frontier already). Plus, 'code' easier than 'research', at least easier to posttrain for, so, I think it follows. I do hate drawing the distinction with sharpness though. (3) is pretty useful - straight labor substitution, you could do it yourself since its basically running a playbook, but don't have to. Agents can be helpful for (2) type work too of course, experiment manager on steriods. I guess the quadrillion dollar question is when we cross the line for (2). I think soon enough, although one can make an argument that RLVR is not a great paradigm to get there (but I wouldn't make a big bet against it being enough).

Introducing Sudo - A delightful music player for the songs you love Reserve yours today: sudo.music

Cars and buses go from Oxford Station down Botley Road for the first time in three years! The project took longer than the construction of The Titanic and saw costs balloon to £237m. Looks good though.