Tyler Murphy

🏆 @duolingo won Animator of the Year

Could a fast, on-device LLM help with autocorrect? My iPhone typing only seems to be getting worse.

I wanted to share something I built over the last few weeks: isometric.nyc is a massive isometric pixel art map of NYC, built with nano banana and coding agents. I didn't write a single line of code.

How rotation lock should work

Chess is finally on Duolingo!

This is my favorite thing I've ever worked on.

<3 the ChatGPT live activity tracker on iOS lock screen This feels like the future (helpers in the background)...

@HardcoreHistory Have you considered publishing the whole catalog of shows to Audible? I have spare credits lying around that I’d definitely spend on HH

Chess on @duolingo 🤩♟️

Ayyy I got to learn chess for this.

Big news: the next subject @duolingo will teach is… chess ♟️ Strategy, logic, and a little bit of trash talk—coming soon to your app.

This lighting in dark mode is such a great detail 🔦

I'm noticing a trend. Here's to hoping AI kicks off a consumer robot renaissance.

"Move 37" is the word-of-day - it's when an AI, trained via the trial-and-error process of reinforcement learning, discovers actions that are new, surprising, and secretly brilliant even to expert humans. It is a magical, just slightly unnerving, emergent phenomenon only achievable by large-scale reinforcement learning. You can't get there by expert imitation. It's when AlphaGo played move 37 in Game 2 against Lee Sedol, a weird move that was estimated to only have 1 in 10,000 chance to be played by a human, but one that was creative and brilliant in retrospect, leading to a win in that game. We've seen Move 37 in a closed, game-like environment like Go, but with the latest crop of "thinking" LLM models (e.g. OpenAI-o1, DeepSeek-R1, Gemini 2.0 Flash Thinking), we are seeing the first very early glimmers of things like it in open world domains. The models discover, in the process of trying to solve many diverse math/code/etc. problems, strategies that resemble the internal monologue of humans, which are very hard (/impossible) to directly program into the models. I call these "cognitive strategies" - things like approaching a problem from different angles, trying out different ideas, finding analogies, backtracking, re-examining, etc. Weird as it sounds, it's plausible that LLMs can discover better ways of thinking, of solving problems, of connecting ideas across disciplines, and do so in a way we will find surprising, puzzling, but creative and brilliant in retrospect. It could get plenty weirder too - it's plausible (even likely, if it's done well) that the optimization invents its own language that is inscrutable to us, but that is more efficient or effective at problem solving. The weirdness of reinforcement learning is in principle unbounded. I don't think we've seen equivalents of Move 37 yet. I don't know what it will look like. I think we're still quite early and that there is a lot of work ahead, both engineering and research. But the technology feels on track to find them. youtube.com/watch?v=HT-UZk…

@signulll @duolingo the young lady’s illustrated primer! it’s gonna sound crazy, but neal stephenson had a pretty good vision of this in 1995

i’ve been messing around with @duolingo a lot lately—not because i had some grand ambition to become fluent in another language, but because i thought it might impress someone (spoiler: she’s gone now). in my mind the most interesting thing though is what duolingo does for languages could be scaled way beyond languages with near zero human costs. imagine a llm framework that could take any skill, woodworking, painting, music theory, astrophysics, whatever, hell you could even ingest a youtube video on demand—& break it down into gamified, interactive chunks. instead of relying on human-made content, this system would be powered by ai. an llm generates the lessons, the activities, the quizzes—even the games. the content supply chain becomes infinite, dynamically generated to suit whatever you want to learn. ai isn’t just a teacher in this system—it’s the curriculum developer, the guide, the actual coach. & it doesn’t just present the material; it shapes the experience around how humans actually learn. this is the ultimate “learn anything” app that’s actually super fun—huge opportunity for duolingo because the distribution advantage. you could build a suite of apps & even charge one subscription for it. the content platform also scales horizontally like no other.

I own an @ouraring and listen to Audible books every night to fall asleep. Is there any way to have my iPhone pause the book when I fall asleep?

ChatGPT needs group chats. Groups of friends, groups of bots, everyone in a chat figuring stuff out together.

This feature would be so good @figma

Sounds crazy but remember in 2012 when they made Apple Maps the default

“Then Apple changes Safari from making Google the default search engine to prompting users with a choice for default search…” Power move would be Apple extending their OpenAI deal and making SearchGPT the default daringfireball.net/linked/2024/08…