Michael Royzen

@0interestrates @StandardSignal Thank you Rahul! 🐐

@MichaelRoyzen @StandardSignal congrats michael!

I've been working on something new, and today I'm excited to announce @StandardSignal! Standard Signal is the first hedge fund that researches and executes trades purely with AI. We train models discover and trade on new fundamental truths about the world before humans can. We are also incredibly excited to be participating in the @ycombinator P26 spring batch. If you'd like to participate in our portfolio, please don't hesitate to reach out about investing as an LP.

Unsloth is doing great work

@arlanr Saint Franks in Rincon Hill is my favorite

sohn is the best coffee shop in the entire city of san francisco.

We did a full-parameter finetune of Llama 3 405B back in August 2024 with DeepSpeed ZeRO 3 + Accelerate + MS-AMP, back when MS-AMP was still maintained. @TheZachMueller Thank you for your work on this. Do you plan to add ao support to Accelerate? Is there anything I can do to help? And @StasBekman I’d like to personally thank you for basically single handedly integrating DeepSpeed into the HF Trainer back in 2021. That saved me so much time and helped make my startup possible.

@TheZachMueller That's super-useful, Zach. In fact I didn't know of MS-AMP's existence until I checked HF Accelerate docs on fp8 training ;) Yeah, benchmarks of te vs ao would be great!

I'm compiling a list of good fp8 training libraries: - transformer_engine github.com/NVIDIA/Transfo… - torchao github.com/pytorch/ao/blo… - MS-AMP github.com/Azure/MS-AMP/t… Which one(s) do you prefer - ease of use, stability, support? Which ones am I missing? Thank you!

@ElizabethTurner @Valkyrie_vc Congrats, Beth!

Today I’m launching @Valkyrie_vc - a $45M early-stage fund investing in AI, infrastructure, and critical industries. Grateful to the founders, partners, and friends I’ve worked with along the way. You’ve truly shaped how I invest and why I built Valkyrie. To the founders: Valkyrie was built for you. To partner with you. To build what matters. This is our new frontier. Let’s build it. Thank you to @_pheebini at @Forbes who captured the story → forbes.com/sites/phoebeli…

@seanwssmith @arlanr @paulg Thanks, Sean — hope you are well! It’s been a minute.

@MichaelRoyzen @arlanr @paulg this you? congrats

3 months ago, @paulg introduced me to a founder. He said we are very similar because we both can easily write 10,000 lines of code a day and we are both solo founders who automate a lot of our work with agents. I assumed he would start hiring soon and that being solo was only temporary. Today he announced his $10.4M round led by @BessemerVP. He is still solo. He also thinks the same way I do: frontier models can be smarter than actual humans. And I agree. A lot of human engineers can be dumb as hell. This is a new era where a 1 or 2 person team can outplay entire organizations.

Congrats to @phindsearch on their $10.4M round! Walls of text are so 2022. On Phind, every answer is now an interactive mini-app. It’s a step towards their vision of creating a personal internet just for you. axios.com/pro/enterprise…

Walls of text are so 2022. Introducing Phind 3, where every answer is now an interactive mini-app. It’s a step towards our vision of creating on-demand software and a personal internet just for you.

@rauchg @vercel Congrats! We are happy customers :)

Announcing @vercel's Series F to build the AI Cloud vercel.com/blog/series-f

@yacineMTB This will age well in the context of the next version of Phind… stay tuned :)

LLMs are going to get so fast that the bottleneck will be compiling the code they write

@AnjneyMidha Yes, but on a slightly smaller scale (with 512 H100s and our custom training stack doing a native FP8 Llama-405B post-train 😁)

Yes sex is great but have you ever had a training run on 10K+ GB200s converge successfully If so could you pls dm me thx

@RazRazcle @AnthropicAI Congrats, Raza!

So pleased to announce the Humanloop team is joining @AnthropicAI! I couldn't imagine a better home for the team. Everyone I've interacted with at Anthropic has been incredibly talented, high-trust and conscious of the stakes at play. Enormous gratitude to our customers and investors who supported us to get here!

@prodmarkllc @phindsearch Interesting, what API use cases do you have?

@phindsearch please make an API this time, or even better become a partner with fal.ai Phind was first to have coding answers powered by references, but no API meant no one cared

@ivanhivanov @phindsearch Yep, you can upload all of your brand imagery!

@phindsearch that looks sick, can you define brand guidelines so it creates assets suitable for my product/company

Good advice when reading a math book

Very impressed with this new search kid on the block. It produces answers with creatively generated diagrams and images and this makes such a huge difference in experience! Multimodal outputs have been promised for long but never delivered. Phind shows this is true gamechanger.

Excited to see the new Phind.com come to life, featuring multi-step reasoning and Generative UI answers: phind.com/blog/phind-2

We're excited to launch Phind 2 today! The new Phind is able to go beyond text to present answers visually with inline images, diagrams, cards, and other widgets to make answers much more delightful. Phind is also now able to seek out information on its own. If it needs more information, it can do multiple rounds of additional searches to get you the most comprehensive answer it can. Check out our blog post: phind.com/blog/phind-2