one underrated thing i’ve been doing w/ opus 4.5 in cursor: i ask it to sketch the first design as terminal charts. super fast way to see structure + flows before i touch real ui. feels like thinking in ascii rn.

It’s an extra step to turn autocaps off and we all know it. Not so nonchalant after all!

I think one reason why many of you are unhappy with this website is that you’re spending too much time on here. 45 minutes to an hour per day is the sweet spot. The experience degrades after about an hour and a half, and by two hours it’s actively dysphoric.

x.com/i/article/2014…

Very happy (and unsurprised) to see Railway succeeding. I have a couple side projects on it and it’s an absolute delight to use 😌

Twitter ads seem to have gone from complete trash to not bad overnight. Suddenly the only ads I get are for various synthesizers and MIDI controllers.

Today we're sharing Signals: our closed-loop system for recursive self-improvement in software development agents. Droid learns from interactions with hundreds of thousands of users - detecting improvement opportunities and autonomously implementing them in it’s own codebase.

Love this! Generating toy apps like this was already one of my favorite uses for CC. Great for testing out things like bulk LLM-based info extraction, since you can easily check the results’ distribution, tweak your prompt, etc

this is what coding was like just 3 years ago

sarah I kind of don't trust anyone who isn't a fan of at least one low brow thing. If you're 100% high brow culture you're trying too hard.

i'm still on the $20 chatgpt plab for codex i have been since the start, it's probably the best price quality for any AI subscription

John Collison told a London audience last year that Stripe averaged 8,015 pull requests per week across ~3,400 engineers. That’s 2.3 PRs per engineer per week, actually below the industry average of 3.5. Now 1,300 of those weekly PRs are fully AI-generated. Zero human-written

claude code but its hooked up to an exercise bike and you have to earn your tokens

Computer use models shouldn't learn from screenshots. We built a new foundation model that learns from video like humans do. FDM-1 can construct a gear in Blender, find software bugs, and even drive a real car through San Francisco using arrow keys.

What can half of GPT-1 do? We trained a 42M transformer called SONIC to control the body of a humanoid robot. It takes a remarkable amount of subconscious processing for us humans to squat, turn, crawl, sprint. SONIC captures this "System 1" - the fast, reactive whole-body

I love making Sonnet 4.5 use nonexistent tools