Burned almost 200 million tokens for 5$ with Deepseek V4 Pro, which is incredible. But asking Claude Opus the same thing, still feels like night and day, for reasoning, thinking and coding. So 150k tokens Claude > 200m tokens Deepseek.
Good to see Kevin Edengren back in action this week, and what a display so far in Kutaisi! First a tough win against Catry and then a superb performance against Vandeermersch in yesterdays QF. His return game has really been sharper than expected against this level.
@damiankust The decline of Withrow (and also Lammons for that part) this year is nearly mindblowing I think. From top ATP to almost not winning anything anymore, wonder what happened there.
@octonion A bit curious, how would you evolve your current glmer models in the repo to factor in these kind of things? Market info seems to be the easiest? But aside from that, would you move to something else if you had the option (beyond glmer)?
@inpredict Hi Mike, I have a question about one of your articles on the inpredictable site. If you would like to help, is there any way to contact you (without paying for verification on X)? Is the form on the site still in use? Thanks.
@Tennisform@tennismasterr In what way? I can say that all players are limited if so, but with his past results since turning pro, stats and outpowering this week. I just think it is an odd statement from a group I thought more of.
@Gingfacekillah Thanks. Have you done any ”heavier” tests for stan and hardware? Like multilevel hierarchial models? I have tried a lot of cloud computers but haven’t really seen significance in cheap vs expensive solution. Guessing the Ryzen should be better here.