O(1)

[lauren @poteto]

my AIE talk is up! come hear how the grok @bot team ships so quickly without compromise on quality

https://t.co/Fel7tlxqSL

[Alex Volkov @altryne]

"Hi my name is poteto and I shipped 3,000 PR's last month. I also go by Lauren" lol

I predict that this talk is going to blow up, so invest early!

[O(1) @onishimura]

The wiki part was cool too. Very similar to Karpathy’s LLM Wiki gist. I think the big difference here is the background dreaming process, and that they actually measured whether it helped.

Claude built linked pages from ~350k internal files at Stainless, then kept fixing and merging them across dreaming sessions. The wiki-only score went from 13% to 48% as they scaled from 2 to 128 sessions. Wiki plus search scored 61% vs 53% for search alone.

[O(1) @onishimura]

One of my favorite articles from Anthropic so far: https://t.co/0xgaDp6kxU

Love how they explore agent memory by having different Claude models play Pokémon Red. Sonnet 3.7 saved 413 memory files and got 1 badge. Fable 5 saved around 10 and got 4.

Fable mostly saved stuff it couldn’t just see on screen, like strategies and rules it had tested. One of its notes was literally “If 3+ balls fail in a row, VARY the timing.”

The dreaming section is pretty funny too. Claude wrote down the wrong location for the PokéCenter near Mt. Moon and kept falling off the same ledge. While reviewing its notes between sessions, it checked the transcript and fixed the note. That run made it to Cerulean City.

Also interesting: on a separate benchmark, telling Opus 5 to only record failures actually made its memory less useful. The stronger models did better when they got to decide what was worth saving.

My big takeaway is that a memory is only useful if the agent knows when it applies - a workaround that fixed something in one environment can be totally wrong in another. Context is critical.

[claire vo 🖤 @clairevo]

BREAKING: @OpenAI product team is absolutely stacked

[cat @_catwu]

Some news: I’m joining OpenAI to lead product management for Core Products!

I have a lot of respect for OpenAI for showing the world that scaling works, for bringing AI to over a billion people, and for believing great products are core to the mission.

I’m excited to work with @thsottiaux, an incredible PM team, and everyone across OpenAI. Together, we’ll build agents that have your back and your team’s, and keep you in control, so you get more time for the people and work you love.

Lots to learn, and lots to build. Please keep your feedback coming!

[Yusuke Wada @yusukebe]

Kanagawa\.pm やるよー！

https://t.co/XrlyAaqFQ0

#yokohamanorth

[Dan ⚡️ @d4m1n]

I have this theory that it's actually a WoW Forever exodus at Anthropic

and Thariq has started a new horde guild and he needs the right people

[0xSero @0xSero]

Formal verification is the only way, fully convinced.

Shoutout to Taelin for seeing it before most of us and building it out.

Someone needs to do a YouTube series “formal verification for agent acceleration”

[Cloudflare @Cloudflare]

Introducing Zero Wire: a public vulnerability intelligence feed on @CloudflareRadar, bringing the clarity of a news wire service to critical security developments. https://t.co/pxMUvkdOgG

[cat @_catwu]

Friday was my last day at Anthropic.

Two years ago, when Claude Code was a small internal tool, I asked it to build an RL env. I steered it for tens of hours but never manually wrote a line of code. It worked, I was hooked, and I joined the team.

When we first launched, most people approved every permission prompt by hand and mostly used Claude to answer questions and write tests. 

Today, the default is trusting Claude's judgment to auto-approve. Claude does large migrations and builds new products. 

Helping build Claude Code into an industry-defining product has been the journey of a lifetime. 

Thank you to the incredible team behind Claude Code, Cowork, and Tag. These products are the sum of thousands of great ideas from you. I'll miss working with you all.

Thank you to our users for embracing a new way of working. Your constant feedback shaped so much of what we shipped and your stories made the work so meaningful.

Excited to see what you all make next!

[James Delorey @deloreyjmi]

It means a lot to see so many people voice support today. I love Portugal. My family loves Portugal. We want to make it our home.

We're building something special at Cloudflare in Lisbon. And it's so much fun! Here's hoping it gets sorted, because I'm not ready for it to be over

[Matthew Prince 🌥 @eastdakota]

@uriklarman We were promised a lot to hire a ton of people in Portugal. We hired a ton of people and the Portuguese government didn’t come through on any of their promises. #clownshow 🤡

[sunil pai @threepointone]

This is crazy, @deloreyjmi is a mensch and Portugal should be fighting to keep him, absolute self goal

[dave @dave_deguzza]

Don't delete Playwright. Paste this into Claude Code instead:

"Find every Playwright/Puppeteer browser cache on this machine and delete the Chromium builds older than 156.0.8078.73. Then edit the Playwright MCP config to add --browser chrome --sandbox so it uses my auto-updated system Chrome. Also grep every Electron app in AppData for its Chrome/ Electron/ version string and tell me which ones are below Electron 43.8.6 / 44.10.1 / 45.2.1."

Took 10 min. Found 3 vulnerable Chromiums, a dead Puppeteer cache, and an EOL Electron in Cursor.

[Matt Pocock @mattpocockuk]

This might be the greatest issue report in history

Some folks wrote a paper about my /retro skill assessing its effectiveness

Incredible

https://t.co/IYo4rhz1Vp

[Tibo @thsottiaux]

We know some Git-fu. Excited the team is finally able to talk about it a bit. Massive props to @tnm and team

[lauren @poteto]

maybe a hot take but i now consider adversarial reviews mostly a waste of tokens. far better to use those tokens on empirical combined with formal verification. i have updated pstack to use the interrogate skill less, and eventually i think i may remove it altogether

i have found very little value with code review in the abstract, agents tend to just get lost in the nitpicks and yagnis

[O(1) @onishimura]

Tokyo Disneyland at Halloween is half the park in full costume, and people go all out. This Hades was one of my favorites

[Andrej Karpathy @karpathy]

Weekend hack: HackerTube. I was annoyed with YouTube's recommendations so I had Claude filter down 43M comments over 20 years of HN to 6,454 videos that keep being recommended. *Significantly* better viewing experience, ~2 hours with Opus 5.5:
https://t.co/JIhXN7OSh1

- Excellent recs without any logins (state is kept in local browser session)
- Thumbnail is a representative frame from middle of the video instead of the super duper A/B test hyper-optimized user uploads
- Underlying data from the comment analysis as one zip file, point your Claude at it and ask for your own version
- HackerTube Longs ✊

A bit more on the process:
I use Claude Code (desktop app for Mac), Opus 5.5 on Medium effort as my default. I only talk to Claude and have it do everything, the only "manual work" I helped with in tandem was in the browser (reminiscent of MenuGen): 1) I had to create some kind of API key from Google Cloud UI to query YouTube for video metadata, 2) I booted up a VPS box for serving and 3) I bought the domain name. Once the box was up I gave Claude the IP and told it update, secure, harden, deploy and optimize the code, and it did the whole thing.

Many awesome projects that would have normally taken weeks/months even can now be few-shotted for breakfast. What else are we not thinking of.