🔥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

America is the greatest country in the world for entrepreneurs ❤️🇺🇸🚀 and it’s going to get even better.

💯

Most people who attack or compete with you will be opportunists. So if the thing you're working on is your life's work, all you have to do is keep going, and eventually they'll drop away. Once several generations of them have dropped away you'll even stop caring about new ones.

16 years ago, an older friend sat me down and walked me through the massive ups and downs of American history and explained how it is a country of extremes. Extremes that lead to extreme outcomes. They explained that America is best when it seems down and needs a comeback. The

Deepseek V3 and R1 discourse boils down to this. Shifting the curve means you build more and scale more dummies

Many people commenting on DeepSeek are hugely underestimating how much more compute it will take to get to AGI

I’ve been training neural nets for 30 years. There was never a day I didn’t wish I had more compute. Finding solutions to problems just means we can now solve bigger problems and that is very exciting! I am finding the hype followed by panic this month to be rather

As always Andrej Karpathy explains it best! "2. is significantly significantly more powerful." <- this is why we started poolside

From our day 0 document (April, 2023) poolside.ai/vision

Why can’t the world just accept that the DeepSeek team has worked hard for years, and on their 3rd model generation released great work that has benefited from compounding research and efforts to improve compute efficiency (training & inference) and data. A focused team that

A sure thing we can take away from DeepSeek is that it’s still possible for a company no one knows to work in the shadows for years and emerge as a threat to OpenAI overnight. Exciting!

If people would know how efficient inference and training was at character.ai under Noam Shazeer they wouldn’t be so surprised by DeepSeek. AFAIK character.ai never went MoE but had the most efficient dense models of anyone, and could have easily gone MoE

Some of my thoughts on China and AI from October 2024. This wasn’t surprising to any of us in the industry.

I kid you not, entering into 🇺🇸, the border patrol officer just asked me: “What do you think about DeepSeek?” 🤯🤣

.poolside is rethinking how AI learns to code. Here’s how co-founder and CTO Eiso Kant explains why their model stands apart. Catch the full interview with Felicis founder and Managing Partner Aydin Senkut on YouTube: youtube.com/watch?v=BAIyTE…

i'm really excited to have my friend Eiso Kant from poolside join us to speak at our annual RAAIS conference on 13 june in london request ticket at www(dot)raais(dot)co it was only a few months ago that everyone freaked out about there being "a wall" to ai capabilities as

🇸🇬 Shoot me a DM if you’re at ICLR

We've been doing this poolside with currently our Code Execution environment being a Top 10 user of regular CPU compute for AWS EKS, and counting >800,000 real world repositories (and 10s of millions of revisions)