GDP's hot take: Compressed Sparse Attention was not a rejection of MLA (DSA ver.) but its dialectical rebirth. Similarities go pretty deep. MLA-NSA(stillborn)-DSA-CSA is one evolution…
"the runs" (can we pick another term btw? how about "training jobs"?) cost on the order of tens of millions of dollars. They *are* cheap. Meituan's…
Tbh I also look forward more to K3, because I know that to expect from V4.But on fundamentals, sorry guys, I still think this is Whale's…
who knew that the race to AGI could be so petty?
RT Andrew KerrNEW: @FreeBeacon has the receipts on Ro Khanna’s obscene, oligarchic wealth:—his 2 kids (under 10yrs old) own 3 private golf courses in Ohio (not…
Good pointDSpark is ironically a more obvious win for the class of models DeepSeek does not work onantirez: My feeling is that DeepSeek DSpark, like any…
RT Ross TaylorSpeaking at @aiDotEngineer today about long-horizon RL and the different tradeoffs involved. 2:25pm.Also here in SF for a week if anyone wants to catchup…
RT Nicholas KristofThe discussion about @elonmusk's comments on humanitarian aid has mostly focused on our values, which seems right. But note that USAID served our interests…
RT Cohere LabsWe're excited to be sponsoring and co-organizing IOL-AI 2026, a new open challenge with the International Linguistics Olympiad @IOLing_official.🌎💬🌐 Open-science🗓️ One-month competition🎯 Targeting AI…
RT clem 🤗A study from @Stanford showed that 71.3% of chatgpt queries could be accurately answered by a local model. I suspect a major part of…
RT Ara KharazianWe can finally say AI isn't killing jobs.A new paper from me, @tryramp, and @RevelioLabs uses firm-level spend and workforce data across 21K U.S.…
Common challenge that will come up in the near future is capturing the gains of greater AI intelligence in organizations. High human capital firms need to…
RT Doctor JackMy grandfather, communism, and the tremor that never left. A story I rarely share.He was the principal of a private school; respected, steady, a…
> The US and China are running very different races with different conditions.That's true but that's mostly the error and dare I say moral failing of…
RT Sara DragutinovicIs Muon as good as they say? We looked beyond training speed and found a hidden cost: Muon loses the simplicity bias of older…
RT NVIDIA GTCGet the best price for #NVIDIAGTC Berlin conference passes today:✅ GTC alumni get the biggest savings✅ Early bird pricing is open to everyone✅ All…
Elite Human Capital can eat you alive anonHighly Skilled Immigration is not good enoughNathan 🔎: The misleading Alzheimer’s data stuff makes me so angry. Billions of…
RT DogeDesignerBREAKING: Philippines becomes the first country in Southeast Asia to commercially launch Starlink Direct-to-Cell through a mobile network. Globe has secured regulatory approval to offer…
RT Daniel van StrienYou can now use 100s of tools with @huggingface Buckets, thanks to the new S3 API! Usually just one or two lines to…
RT mr-r0b0t"Trained on only 15B tokens, HydraHead improves long-context performance by 69%+ on NIAH at 512K context, approaching the performance of Qwen3.5."NIAH = Needle In A…
RT Tiezhen WANGDeepSeek is the forever goat on ML infras.If they haven't open source all the god-like AI infra projects like DeepSpec / DeepEP / DeepGEMM,…
RT Konstantin KisinWe tax cigarettes to reduce smoking.We tax alcohol to reduce drinking.We tax fuel to reduce driving.What do you think happens when you tax employing…
I have very little idea of what K3 might deliverwhat hints we have are pretty intimidating.Elaina: Kimi K3 in July.
RT Quentin Lhoest 🤗S3-clients + Hugging Face Buckets = 💥You can now query and write directly to HF Storage Buckets via the S3-compatible APIJust one secret.…