Tessera
Folgen

Anthony Panozzo

@panozzaj.com

Here for AI and more positive vibes

996 Posts346 Follower629 Folgt
Liz Fong-Jones (方禮真)@lizthegrey.com · 3d ago

I've been busy putting Claude Fable to work saving the environment from wasted cycles, rather than generating excess cycles, namely by adding arm64 NEON support to everything that benefits from it in the Go standard libraries and runtime and popular libraries that no one had gotten around to.

Anthony Panozzo@panozzaj.com · 2w ago

Opus 5 has achieved true senior engineer status: > Committed. One honest course-correction on the checklist before I continue: > A2 as I specced it is redundant, and I'd rather not build it.

Pekka Lund@pekka.bsky.social · 2w ago

Meta AI (FAIR) researcher says it's fun to watch how LLMs have their minds blown when they are given the polynomial only and they realize it's a counterexample to the Jacobian Conjecture. If only raw reasoning was available from more models. Summary of mind blown isn't as mind-blowing.

In love with just giving this polynomial to an LLM with no context and innocently asking it "study" then watching its mind get blown in real time
whet moser@whetmoser.com · 2w ago

from watching people who are using AI at a very high level instead of coding at a very high level, they still seem to be doing something that requires a lot of thought and sophistication. it seems like you're still going to have to know ball but ball will be different

Kyle Orland@kyleor.land · 3w ago

A lot of people probably heard about a study last year that found coders were 19% less efficient when using AI tools. A lot fewer people heard about the February update from the same researchers showing 2026-era AI coding tools *increase* productivity (I just found out about it today, personally)

We are Changing our Developer Productivity Experiment Design

Our second developer productivity study faces selection effects from wider AI adoption, prompting us to redesign our approach.

metr.org

Steve Klabnik@steveklabnik.com · 3w ago

things that used to suck in the past: "oh no, my website is two major versions out of date on its build tooling. upgrading that will take forever and be a pain" now: "claude did that for me while i did other actual work and validated that there were no regressions"

Axe Ghost. On Steam!@axeghostgame.bsky.social · 4w ago

"[llms] make it harder to construct a detailed mental theory of the software, but they allow you to build a partial theory quickly and they can help you leverage that partial theory more effectively"

Post nicht verfügbar.
Ethan Mollick@emollick.bsky.social · 4w ago

This was one of those impressive AI thresholds for me. I gave GPT-5.6 Sol in Codex control over my computer, and asked it to win the daily challenge for the game Slay the Spire 2 (randomized factors, so can't cheat). Its a complex game. It worked for 5 hours, making complex game choices... and won.

BildBildBildBild
Ben Lamb@bennyfactor.bsky.social · 4w ago

well the robot looked over some code I wrote ten years ago for importing 20k-50k address lines of atm locations and geocoding them and took the processing down from about 2 hours to 7 minutes (plus time waiting for openstreetmap / esri / google to respond for new addresses) so... AI not all bad

Arseny Khakhalin@khakhalin.bsky.social · 4w ago

A nice clash with pragmatics (when "Could you pass the sugar?" is not a question, and answering "Yeah I could" is a bad option) Fable understands "What do you think?" as an invitation to provide an opinion, but "What do you say?" as an invititation to implement the idea, UNLESS it disagrees!

Anthony Panozzo@panozzaj.com · 4w ago

With better translation tech, for the first time I am following several primarily non-English-writing accounts and getting some value. Could be nice to bake that in to the product or some extension. Seems like a new capability!