Your curated collection of saved posts and media
@BowsersaurusRex @peterwildeford I think we're maybe talking past each other. Let me make a specific mechanistic claim: I claim that when the model writes a poem, there are activations on the tokens at the end of the line which represent candidate targets for the end of the next line. https://t.co/fyxkKMJIEh https://t.co/RmFIf1jcFA

@BowsersaurusRex @peterwildeford We know this is true, because we can modify those activations and change the next line correspondingly. https://t.co/sDoFDk9CA5
We did a very careful study of 10 optimizers with no horse in the race. Despite all the excitement about Muon, Mars, Kron, Soap, etc., at the end of the day, if you tune the hyperparameters rigorously and scale up, the speedup over AdamW diminishes to only 10% :-( Experiments are made possible by Marin (https://t.co/UgEjGM0HPY); anyone developing new optimizers: please come try your method on this benchmark!
(1/n) Check out our new paper: "Fantastic Pretraining Optimizers and Where to Find Them"! >4000 models to find the fastest optimizer! 2Γ speedups over AdamW? Unlikely. Beware under-tuned baseline or limited scale! E.g. Muon: ~40% speedups <0.5B & only 10% at 1.2B (8Γ Ch
Excited to share what friends and I have been working on at @Standard_Kernel We've raised from General Catalyst (@generalcatalyst), Felicis (@felicis), and a group of exceptional angels. We have some great H100 BF16 kernels in pure CUDA+PTX, featuring: - Matmul 102%-105% perf of cuBLAs in 100 lines of code - Attention 104% perf of FlashAttention3 in 500 lines - Fused Llama3 FFN 120% perf of PyTorch (gpt-fast) Reach out if you want to work on AI kernel gen with us!
How does ChatGPT not know that Donald Trump is the current US President? I asked ChatGPT: "Can you give me the top news stories of the day?" The exchange below is pretty bonkers -- even when I pushed back, it still did not acknowledge he is the current US President https://t.co/OkfKKG1uGP
A must read indictment of generative AI. Like seriously, stop what youβre doing and read this essay. It doesnβt nibble around the edges. https://t.co/75R9eZG31u
Taking ππ§ ππ―πΊπ°π―π¦ ππΆπͺππ₯π΄ ππ΅, ππ·π¦π³πΊπ°π―π¦ ππͺπ¦π΄ to the streetsβ¦ https://t.co/EVsNnzJyHC
Is If Anyone Builds It, Everybody Dies by @ESYudkowsky and @So8res βthe most important book of the decadeβ or just βScientologyβ? Donβt just blow the book off; take these potentially catastrophic issues seriously. But donβt believe every word, either. @TheTLS I dissect what they got right, and what they got wrong. Link below.
Dissecting rather than dismissing If Anyone Builds It, Everybody Dies @TheTLS : https://t.co/oKZF5Gvl8Q
This is the clip so the audience can see it in full. https://t.co/jzSvBJg0PG
Y'all fuck with ilya merch? https://t.co/rISV4ABGit

Brand exploration for OpenAI with Sam Altman, February 2023 Two logo concepts β Circle and Monogram β alongside broader exploration for ChatGPT across brand and product. More details and link below. https://t.co/y0gWzAqlPV

AI challenges t he dominance of Google search https://t.co/Y7jJ7YY58c @sbearne @bbcnews
AI challenges t he dominance of Google search https://t.co/Y7jJ7YY58c @sbearne @bbcnews
βI have to do itβ: Why one of the worldβs most brilliant AI scientists left the US for China https://t.co/qcVOQH4fQJ
βI have to do itβ: Why one of the worldβs most brilliant AI scientists left the US for China https://t.co/qcVOQH4fQJ
Meta struggles to decouple from Chinese supplier of AI smart glasses https://t.co/aZFcI2JjXv @MsHannahMurphy @EleanorOlcott @ft
Meta struggles to decouple from Chinese supplier of AI smart glasses https://t.co/aZFcI2JjXv @MsHannahMurphy @EleanorOlcott @ft
AI companies want copyright exemptions for NZ creatives, the market is their best protection https://t.co/G4ZZPYTWQH @ConversationEDU
Google owner reveals Β£5bn AI investment in UK ahead of Trump visit https://t.co/pBbUnenDOY @google @faisalislam @bbcnews
Google owner reveals Β£5bn AI investment in UK ahead of Trump visit https://t.co/pBbUnenDOY @google @faisalislam @bbcnews
Vibe coding has turned senior devs into βAI babysitters,β but they say itβs worth it https://t.co/HrFWgM9wmn @DominicMadori @techcrunch
Vibe coding has turned senior devs into βAI babysitters,β but they say itβs worth it https://t.co/HrFWgM9wmn @DominicMadori @techcrunch
βI love you too!β My familyβs creepy, unsettling week with an AI toy https://t.co/FCT72XebCL
βI love you too!β My familyβs creepy, unsettling week with an AI toy https://t.co/FCT72XebCL
Facebook owner Meta unveils new AI-powered smart glasses https://t.co/N9uROyQ1dd @lilyjamali @bbcnews
Jensen Huang 'disappointed' by reported China Nvidia chip ban https://t.co/ymEJunHXDj @faisalislam @bbcnews
AI CEO says technology βmoving very quickly,β could soon replace more jobs https://t.co/2BRDBik9OL
US tech groups answer Starmerβs call for AI infrastructure spending https://t.co/4NRwwSOjkg @tim @RachelMillardAS @AnnaSophieGross @ft
AI Chatbots Are Not Therapists: Reducing Harm Requires Regulation https://t.co/zKk76Z8KAA @techpolicypress
AI can forecast your future health just like the weather https://t.co/XTm41WyFCl @JamesTGallagher @bbcnews
What is Microsoft's new AI supercomputer in Loughton, Essex? https://t.co/UePEDMFdCu @bbcnews