🔥The 9th Round of Easy Loan, Earn $40 Reward is in progress❗️ ⏰ Promotion Period: January 15th - Feburary 15th, 2025 👉 Register now and check more details at gate.io/campaigns/358

caching can have huge impact on throughput! great working with Paul Merolla and team MK1 Technologies

DeepSeek is clearly a cracked engineering team that made smart design choices across the stack—from kernel optimizations all the way to model architecture. But… what if they’re also part of a psyop to tank the US tech sector? Sounds wild—wait, hedge fund vibes... oh no. 🫠📉

Needed an excuse to try Wan2.1 text-to-video model--so decided to make my own Trump Zelensky videos. From initial HF download to making a video took a few min and a nice Sunday distraction 😂

Excited to be speaking alongside Paul Merolla tomorrow at lu.ma/beyondcuda25

Scaling LLMs across 2 & 4 GPUs on AMD MI300X was limited by xGMI bandwidth—so we built an optimized ROCm All-Reduce library. Now open-sourced, it achieves up to 2.25x speedups! Excited to see more DeepSeek-style innovations from the AMD community 🚀 github.com/mk1-project/qu…

Congrats! We love working with Susa. DM me if you are considering Susa, and want reasons to join the family.

Curious how much revenue OpenAI generates simply from startups benchmarking against them. We've spent thousands—collectively, it's gotta be tens of millions. Leadership perception is lucrative.

🌎 The Neuralink Patient Registry is now open globally! If you have quadriplegia and are interested in exploring groundbreaking ways to control a computer using your thoughts, you can now sign up from anywhere in the world. neuralink.com/patient-regist…

Super happy to see the love for CHAI from our OG user! 🙌 And you're 100% correct—we started with GPT-J. Today, our AI team has built our own in-house AI system. It’s not just a single LLM but a powerful fusion of ~10 LLMs, a reward model, and an advanced orchestration system

LLMs like GPT4o often struggle with complex queries on large documents. But why? Our new benchmark spotlights a key deficiency—comprehension—that worsens as context grows. The good news: We show how our custom LLM CAMeL can augment a frontier model, vastly improving

Zyphra is releasing our first reasoning model, ZR1-1.5B. This small but powerful reasoning model excels at both math and code, making it one of the best models in these categories for its size. It also uses 60% less reasoning tokens than comparable models. 🆓Apache 2.0 license.

I am the 3rd person in the world to receive the @Neuralink brain implant. 1st with ALS. 1st Nonverbal. I am typing this with my brain. It is my primary communication. Ask me anything! I will answer at least all verified users! Thank you Elon Musk!

Congrats on the progress ❤️

Congratulations, @Neuralink team!

Neuralink really listens to patients—their electrodes sit right next to neurons. Watch Dave walk through a key part of Neuralink’s incredible tech stack, and how they’re pushing the boundaries to make electrodes invisible to the brain.

100% this. Focus on fundamentals and be the best at the underlying technologies, and amazing things are possible 🧠🚀

MK1 on the "token usage" leaderboard for Grok on open router! This is a drop in the bucket compared to the traffic that is coming 🚀

I love that there is an organic community forming with the first patient group. Watching them interact and play streaming games using "Telepathy" is surreal — like glimpsing the future of human connection.

There are still people who think LLMs are just fancy pattern matchers — incapable of reasoning on unseen problems. Here’s Grok-4 solving a problem I’m pretty sure no one has ever posed before. To get it right, it needs: • a mental model of a physical keyboard • rule-following