←
Eric Zelikman1,346 posts
Eric Zelikman avatar

Eric Zelikman

@ericzelikman

cofounder & ceo @humansand - building ai for humans // was lgtm-ing @xAI, phd-ing @stanford

◷ Joined April 2010▧ 171 Photos and videos
Jensen Huang avatar
⟳ Eric Zelikman retweeted
Jensen Huang@JensenHuang·Nov 17

Over the past week, I’ve been overwhelmed by the love and support of friends, industry colleagues, and people I've never met. Thank you to everyone who has shown kindness to my family and me. It means the world to me. I came to the United States 53 years ago, and becoming an American citizen is one of the greatest gifts and proudest moments of my life. NVIDIA has been my life’s work. Building it into the company it is today alongside my colleagues has been a privilege and an honor. Nowhere but in America could a company like NVIDIA be possible. I love our country, and I love NVIDIA. I did not live up to the standards expected of everyone at NVIDIA, and I have made serious mistakes in my personal life. I created problems for the Board that I deeply regret. I own my mistakes completely, and I am committed to cooperating with the Board’s independent investigation. Since my departure, I’ve seen an outpouring of calls from NVIDIA’s employees, investors, business partners, and customers for me to return as CEO. Many continue to reach out to me directly as the company navigates complex technology, market, competitive, and geopolitical dynamics. NVIDIA is a once-in-a-generation company. If the Board of Directors decides that my return is in the best interest of our shareholders, our employees, the company, and America, I would be honored to answer these calls and return as CEO. Thank you again to everyone who has shared their good wishes. I am profoundly humbled by your support. To my family: I love you, and I will spend the rest of my life being worthy of yours. Jensen

humans& avatar
⟳ Eric Zelikman retweeted
humans&@humansand·Nov 17

Try Ampersand for free for a bit longer! We love what folks have shared with Ampersand, so we’re extending our promo for you to build some more

Post attachment

Ampersand, our collaborator model, is now on Vercel AI Gateway - use it free for the next week! Ampersand matches or beats all frontier models (e.g. Fable and Astra) in multi-user long-horizon settings on our evals, and most on external cooperation and multi-agent coding evals, at 744b params (40b active)

Ziang Li avatar
⟳ Eric Zelikman retweeted
Ziang Li@zianglih·Nov 13

and production low-precision RL (NVFP4, MXFP8) at 2:20pm!

Post attachment

Looking forward to talking about the many interesting RL and infra challenges with building AI for humans tomorrow at 10am!

Eric Zelikman avatar
Eric Zelikman@ericzelikman·Nov 13

Looking forward to talking about the many interesting RL and infra challenges with building AI for humans tomorrow at 10am!

humans from @humansand will be attending and speaking at the SGLang summit tomorrow, come say hi!

Jensen Huang avatar
⟳ Eric Zelikman retweeted
Jensen Huang@JensenHuang·Jul 24

For my first post, I’m sharing a letter @NVIDIA signed on why open models matter. AI will transform every industry, power every company, and be built by every country. Open models strengthen safety and cybersecurity, accelerate innovation and diffusion, and enable sovereignty. The world needs both frontier closed models and frontier open models. images.nvidia.com/pdf/Open-W…

Post attachmentPost attachmentPost attachment
humans& avatar
⟳ Eric Zelikman retweeted
humans&@humansand·Nov 10

Want to try a top model for free without giving up your data for training? Check out Ampersand in @vercel

Ampersand, our collaborator model, is now on Vercel AI Gateway - use it free for the next week! Ampersand matches or beats all frontier models (e.g. Fable and Astra) in multi-user long-horizon settings on our evals, and most on external cooperation and multi-agent coding evals, at 744b params (40b active)

Yuchen He avatar
⟳ Eric Zelikman retweeted
Yuchen He@YuchenHe07·Nov 8

a good model - RL’ed to collaborate with us and care for us

Without any targeted training, Ampersand gets frontier-level performance on DeepSWE v1.1 (77.9% in multi-agent mode, matching / beating Astra and Fable) and we found it topped EQ-Bench (ELO 1380.5) - thanks @YuchenHe07 and @alexisjross for the findings To learn more about Ampersand, try it or see our blog: ampersand.humansand.ai/blog/

Eric Zelikman avatar
Eric Zelikman@ericzelikman·Nov 8

try ampersand free on @vercel for a week!

Ampersand, our collaborator model, is now on Vercel AI Gateway - use it free for the next week! Ampersand matches or beats all frontier models (e.g. Fable and Astra) in multi-user long-horizon settings on our evals, and most on external cooperation and multi-agent coding evals, at 744b params (40b active)

Yuchen He avatar
⟳ Eric Zelikman retweeted
Yuchen He@YuchenHe07·Nov 5

We cared about how model collaborates w humans& ai, and fun to figure out it also matches frontier coding in collaborative way even though we did not train coding explicitly

Read more details in ampersand.humansand.ai/blog

Eric Zelikman avatar
Eric Zelikman@ericzelikman·Nov 6

it is a bit remarkable a model of only 40b active params (744b total) goes head-to-head with some of the biggest and best models including astra and fable, with no coding tasks in our training mix thanks to @YuchenHe07 for the discovery

Co-Coding: frontier coding by working together We trained Ampersand to work better with people - and found that multi-agent Ampersand matches frontier coding models while being faster and cheaper, despite no extra coding training - 92.2% on SWE-bench Pro v2 HARD-51, matching Fable 5.1 (while 70% cheaper) and Gemini 4 Argon and outperforming Astra, Muse Wonder, and Kimi K3 - 77.9% on DeepSWE v1.1, on par with Gemini 4 Argon and ahead of Astra, Grok-4.7, Kimi K3, and Deepseek-v4-pro Try co(llaborative)-coding with Ampersand’s multi-agent mode, available exclusively through our Agent API Have fun building together - for the next week, get a $250 API credit to try it out!

RadixArk avatar
⟳ Eric Zelikman retweeted
RadixArk@radixark·Nov 2

Really nice work from @humansand. Built on the delta weight sync in Miles, taken much further, and contributed back to Miles so everyone on the stack gets it. This is exactly the kind of collaboration we hoped Miles would enable. Thanks to the humans& team!

Large-scale RL has a problem: weight updates from trainers to samplers are slow and expensive We’re sharing weight teleportation: faster weight updates than existing intra-cluster techniques with internet speeds, enabling samplers anywhere (maybe even space!) Weight teleportation is so efficient (<0.23s sampler pause time, ~3s non-blocking compression + prepare), the bottleneck ends up being the GPU hardware’s decompression ASIC We unlock these gains by using GPUs for massively parallel compression of model weight changes, engineering a heterogeneous decompression and patching pipeline, along with other optimizations and overlaps. These techniques are co-designed with hardware-native features in NVIDIA’s Blackwell generation of chips We’ve open-sourced weight teleportation, so everyone can use it today

Ziang Li avatar
⟳ Eric Zelikman retweeted
Ziang Li@zianglih·Nov 2

Forget about p2p rdma, we now count weight sync pause time in milliseconds

Large-scale RL has a problem: weight updates from trainers to samplers are slow and expensive We’re sharing weight teleportation: faster weight updates than existing intra-cluster techniques with internet speeds, enabling samplers anywhere (maybe even space!) Weight teleportation is so efficient (<0.23s sampler pause time, ~3s non-blocking compression + prepare), the bottleneck ends up being the GPU hardware’s decompression ASIC We unlock these gains by using GPUs for massively parallel compression of model weight changes, engineering a heterogeneous decompression and patching pipeline, along with other optimizations and overlaps. These techniques are co-designed with hardware-native features in NVIDIA’s Blackwell generation of chips We’ve open-sourced weight teleportation, so everyone can use it today

Eric Zelikman avatar
Eric Zelikman@ericzelikman·Nov 2

Datacenters and RL algos are usually built assuming trainers and samplers must be in one place as weight updates send tons of data Weight teleportation lets us rethink this, making crosscluster updates often faster than intracluster, removing decentralized RL's core drawback 🛰️

Large-scale RL has a problem: weight updates from trainers to samplers are slow and expensive We’re sharing weight teleportation: faster weight updates than existing intra-cluster techniques with internet speeds, enabling samplers anywhere (maybe even space!) Weight teleportation is so efficient (<0.23s sampler pause time, ~3s non-blocking compression + prepare), the bottleneck ends up being the GPU hardware’s decompression ASIC We unlock these gains by using GPUs for massively parallel compression of model weight changes, engineering a heterogeneous decompression and patching pipeline, along with other optimizations and overlaps. These techniques are co-designed with hardware-native features in NVIDIA’s Blackwell generation of chips We’ve open-sourced weight teleportation, so everyone can use it today