Keep on to blur preview images; turn off to show them clearly
another major reason, almost all RL papers use veRL. veRL didn't have fp16 support lmao
RL and efficient distributed pretraining • eXperiments lab • memes and training lores
> representing an annual market size of roughly $50 billion In the limit, China can absorb so much more than $50B of GPUs that, soybeans (lol), and energy = most of the deficit can be canceled we'll see what Trump's priorities are
We're in a race. It's not USA vs China but humans and AGIs vs ape power centralization. @deepseek_ai stan #1, 2023–Deep Time «C’est la guerre.» ®1
RT @arvidkahl: This is quickly turning out to be my most popular podcast episode in months. Who knew that founders like to hear why they s…
Building https://t.co/od97B0HVrk and https://t.co/666FnyVVE0 in Public. Raising all the boats with kindness. 🎙️ https://t.co/6w69DZmi8H · ✍️ https://t.co/lpnor5rsTW
Engine is one of the less legit robotic companies they might have good hardware (or just push it harder) but they don't have good robust policy and, what's worse, they oversell what they have with a few scripted demos. At least Unitree is honest that they sell research platforms
We're in a race. It's not USA vs China but humans and AGIs vs ape power centralization. @deepseek_ai stan #1, 2023–Deep Time «C’est la guerre.» ®1
完整详细内容访问：https://t.co/ZRJVf2gqyv
学AI找小互，找小互，上 https://t.co/4PVaHEr5r3 ...
!! I really cannot believe this works. Large splat scene in VR on a Quest 3 using sparkjs. LoD budget is 500k splats. Runs about 30 fps. Full scene is ~6million splats. It's janky (sorry about the controls) But still, just insane this works at all ... (Metaverse is coming!)
GP @ a16z ... questionable heuristics in a grossly underdetermined world
