@Jobvo It can run Kimi or MiniMax models at around 60t/s from what I've seen... so it could output about 5 million tokens of inference per day... Maybe useful if you run it 24/7 working on a project, but too slow to do much real work... certainly not enough to recoup $30k in hardware
@MetalJesusRocks@analogue There's a zero percent chance they do this... Think of the customer support nightmare of an official product offering that requires non-tech users to open their device up with a screwdriver and transplant all of the components into a new shell... think of all that could go wrong.
@retrotink2@WobblingP damn, dude. do you ever sleep?? congrats on making the best retro gaming product even better with each firmware release! happy new year to you and @MrsRetrotink
@pr4m0d I remember electric playground. I think his cohost is actually still running it to this day.
Tommy was the strangest reviewer. He'd pick some arbitrary gripe about a game and be like "they used too much orange, I give it a 5 out of 10"... totally bonkers stuff, but entertaining.
@MAlshaik__ > you can have any stance you want in private. who cares?
Idk man, the position is still reprehensible, even if he kept it to himself. Maybe we can be charitable and assume he's a complete kool-aid drinking ignoramus with regard to Israel/Palestine, but that's not much better.
@rileybrown_ai "better"? no... faster? yes. and in many agentic coding sessions that tradeoff is a good one.
I really want a way to use GPT-5 for planning and Sonnet 4.5 for the work.
@Willi_ammmm @EricPhy@claudeai I mean, it's no surprise that the LLM itself doesn't know what it is, but they probably ought to put something in the chat's system prompt to tell Claude what it is... makes these types of introspective queries less awkward.