@0xSero Supremely impressive engineering, but the 8k context for a single 5090 might be a severe limitation for many apps. Multiple GPUs opens things up.
@DudespostingWs Good idea, but it blocks airflow to the radiator, which is likely to cause overheating and could affect engine lifetime. He just needs to revise the design to let more air through.
@levelsio Generally speaking, these metrics look in the mirror only & do not consider future (possibly explosive) demand. Some like Q ratio assume commodity dynamics so no sustainable competitive advantages like network effect, patents, etc.
Is AI hot? Yes.
Justified? These don't answer.
@evan_thayer There is an interesting senior management challenge in balancing "Think Different" creative culture versus an obsessive OCPD-like focus on eliminating any deviance.
@mathemetica Common confusion - AI tensor math (like PyTorch) is not the same as physics tensor math.
In physics, typically all dims are involved in a calc.
With many PyTorch ops, only the last 2 dims are involved, so its really 1D & 2D math, with higher order dims just along for the ride.
@StockSavvyShay Usage demand for AI is exploding due to change from chat to AI agents running hours, now for days, soon weeks. Hence NVDA GPU demand is explosive. Further, memory (HBM) per GPU has grown from 40 GB (2022) to 284 GB (2026) so that is 7x NVDA's growth. Spectacular MU upside.
@RealMattCouch Cop seemed fine, gave him a break by approaching him in a parking lot. Motorcyclist is lucky he wasn't stopped when he was on the road, assuming plate would be still hidden. (Why hide a plate only when you buy gas.)
@sudoingX Also highly innovative exploration of architectural landscape resulting in innovative models that make max. use of compute. Also way more STEM grads. Hard to make a case that they won't lead soon in AI, just as they lead US now in manufacturing stuff & infrastructure.
@petergyang@NousResearch@karan4d * View of, and any plans for enhancing (other than Telegram, etc.), coordination between multiple Hermes instances.
* How he sees local LLMs (or more generally local models) playing out and how that affects Hermes architectural direction and monetization plans.
@petergyang@NousResearch@karan4d Hermes differentiation vs. OpenClaw and others, plus where he thinks that carries Hermes longer term vs. the others, and why that destination is a better place to be.
@davepl1968@FrankBr05713205 2/ Another way to see it: fully charged battery presents backwards opposing voltage so normally zero current flow. When spike hits battery back voltage is exceeded, causing it to carry current & that reduces the spike. Spikes are modest energy so limited by moderate current flow.
@davepl1968@FrankBr05713205 OP is right; don't disconnect it if engine on. When it sees a larger voltage than fully charged voltage, battery acts like a capacitor, small resistor & voltage clamp due to its electrochemical reactions, electrolyte conductivity at these higher voltages, & other characteristics.
@AncientEpoch Congrats, but where is the English translation in 1 document? Specific URL please. I looked on the web site 2 days ago but couldn't find it. I saw the paper and lots more on technique and prizes, but not the translation. I am obviously curious about what it said.
@smdcapital@PodcastAlphaX@brent_e_trader@GavinSBaker It will cost you $50 billion, it will take you 2-4 years to get all the equipment running & then 2-4 more years to get acceptable yield. So 4 to 8 years out, at best.
If you were the richest person in the world and can throw 200+ billion on the bringup, cut that to 3 to 6 years.