Pinned Tweet
Shifting from debuggin code to debugging models. What actually happens when you train a model and why ? Read me latest blog while your model trains :)
Blog link in 🧵
1
3
420
Ashwin retweeted
one year ago today:
> Jensen was leading NVIDIA
> Anthropic was valued at $183B, not preparing a $1.4T IPO
> OpenAI was leading on benchmarks
> Claude Code, Codex, and coding agents hadn’t taken the world by storm
> RSI was a distant dream
> agents didn’t break out of their sandbox
> OpenClaw, Hermes, and Grok Bot didn’t exist
> AI generated proof of open math problems were still rare
> xAI and SpaceX were separate
> SpaceX hadn’t IPOed
> Cursor wasn’t acquired yet
> GPT-5.1 had just launched
> Gemini 3 wasn’t even out yet
one year in the AI planet is 7 years on earth. the rate of progress can only accelerate from here.
Ashwin retweeted
The bitter lesson 2.0 is that we are not on an exponential, but on a bunch of sigmiods stacked on top of each other. To move within one sigmoid, scaling compute is necessary but to jump from one sigmoid to another you need innovation.
Ashwin retweeted
We're about to enter a period where not reading or writing becomes as ubiquitous as smoking once was. We can hope people will eventually realize this is bad for them, as they did with smoking. But this is far from certain; the cost may not be as obvious.
Ashwin retweeted
“𝘓𝘦𝘵 𝘵𝘩𝘦 𝘺𝘰𝘶𝘯𝘨 𝘮𝘢𝘯 𝘨𝘰 𝘢𝘯𝘥 𝘩𝘶𝘯𝘵, 𝘪𝘧 𝘩𝘦 𝘬𝘪𝘭𝘭𝘴 𝘢𝘯 𝘦𝘭𝘦𝘱𝘩𝘢𝘯𝘵, 𝘩𝘪𝘴 𝘱𝘰𝘷𝘦𝘳𝘵𝘺 𝘦𝘯𝘥𝘴... 𝘪𝘧 𝘵𝘩𝘦 𝘦𝘭𝘦𝘱𝘩𝘢𝘯𝘵 𝘬𝘪𝘭𝘭𝘴 𝘩𝘪𝘮, 𝘩𝘪𝘴 𝘱𝘰𝘷𝘦𝘳𝘵𝘺 𝘦𝘯𝘥𝘴.”
— 𝘈𝘧𝘳𝘪𝘤𝘢𝘯 𝘗𝘳𝘰𝘷𝘦𝘳𝘣𝘴
So true, "transformations happen in an instant" - @kushaal4
Most of the time its possible to clearly demarcate exactly when this happens, sometimes in the moment but definetly in retrospect.
And I think it’s more than just instantaneous. It also has a kind of all-encompassing quality to it: a flood of new things lifts the baseline, like a step up rather than just one thing getting better. It’s many things changing at once, and the cumulative effect makes it feel like you’re floating upward.
at almost every threshold there are a bunch of people who set up a basecamp where they discuss how to cross the threshold. some of those people do go on to cross the threshold, but those people tend to then become incomprehensible to most of those who are still in basecamp
2
117
New SDLC 🤟🏻
Replying to @dhh
At those speeds, you could just Monte Carlo feature development with LLMs then use evolutionary methods to pick and choose the best bits of each implementation.
62
Ashwin retweeted
If Drowning Pool made a song with eMudhra
One, nothings wrong with me
Two, nothings wrong with me
Three, nothings wrong with me
Four, fuck thats my private key
The world isn't abmitious enough for the models. What are models not able to do for you today?
1
1
61
We have built 10 different AI products at PostHog like PostHog AI, our Wizard, Replay Vision, Scouts, and more. Combined, they are used by nearly 100k people every month.
Along the way, we’ve learned a lot about building them, what works, and what doesn’t. Here are the 38 most important lessons:
We don't credit ourselves enough when we move a possibility from a zero to a non-zero.
Moving something into the relam of possiblity in itself is a game-changer! Be proud.
2
80
"startups often garbage-collect broken companies and industries" -@paulg
This filter isnt talked about enough.
44

















