David Kell

1/📢 Some news! I'm leaving my role as community lead at @DuneAnalytics, and joining @epochbio as a machine learning engineer. Working at Dune was an awesome experience. I got to work with some of the best in the business. Here are my top three learnings:

Graphy's latest launch illustrates a clever UX pattern: Web clipping into structured data. I expect more of this to be combined with AI agents and workflow automation.

Magical

If you someone who tends to spend a lot of time in between browsers tabs looking at charts - I recommend you sign up to the extension early access. Especially if you are doing it for all your clients. It's going to save you ⏰⏰⏰ Great work @graphyapp team

Nice one @agaperste + @chuxin_h 🙌

7/But I think what they’ve built is unique. I’ve played with it - it’s awesome. (Don’t believe me - watch the video!) I’m super excited to see it go live 🚀

6/The answer they’ve come up with - it sounds crazy, impossible even. It goes against everything you’re “supposed” to do. And it took a lot of conviction to pursue it, to solve all the technical problems.

Let me tell you why I think this is awesome… 🧵

Notes from a talk I gave at @blockchainox / @homedao It’s about the scientific system, how it works, the bottlenecks that stop us scaling it today, and why decentralized primitives might help Cheers for the invite @jan_o_e FYI @DesciLondon davidkell.io/desci-from-fir…

23/My gut feeling is that this debate will shape the future of science for the next few decades. Alongside the evolving tradeoffs in those three constraints. Certainly an exciting time to be around!

22/But this is super relevant for AI + science. We want our learning algorithms to be accurate, and in particular in novel domains. The only way to do this is to model how physics, chemistry, etc actually work. No heuristics!

1/GPT-4 live this week! I spent this month getting back up to speed with AI, transformers, large language models, etc. Everything happening at the moment is driven by a handful of fundamental constraints. Here's how to think about modern AI from first principles.