Garrett Robinson

He/him. @AnthropicAI. Formerly  SEAR, @brave, @freedomofpress CTO, @SecureDrop lead developer, @mozilla.

[rohit @krishnanrohit]

We're all looking for the guy who did this

[Garrett Robinson @garrettr_]

Seeing our leadership stand firm on principle makes me proud to work @AnthropicAI

[Anthropic @AnthropicAI]

A statement on the comments from Secretary of War Pete Hegseth.

https://t.co/Gg7Zb09IMR

[Lydia Hallie ✨ @lydiahallie]

Excited to announce Claude for Open Source ❤️

We're giving 6 months of free Claude Max 20x to open source maintainers and core contributors.

If you maintain a popular project or contribute across open source, please apply!

https://t.co/inuh0hxREA

[Alex Albert @alexalbert__]

Introducing Claude Opus 4 and Claude Sonnet 4.

Our best models yet. The Claude 4 family is here.

[Catherine Olsson @catherineols]

Claude Code is very useful, but it can still get confused.

A few quick tips from my experience coding with it at Anthropic 👉

1) Work from a clean commit so it's easy to reset all the changes. Often I want to back up and explain it from scratch a different way.

[Garrett Robinson @garrettr_]

@camfassett such fog this morning, very ethereal vibes

[Deena Mousa @deenamousa]

Inspired by this viral meme about DOGE: Research the US government has supported that can be made to sound silly, but that has contributed to human progress.

Valuable work can often be framed as absurd out of context. That doesn’t make funding research any less important.🧵

[Garrett Robinson @garrettr_]

@JeffLadish https://t.co/AEOdHrXifR by @GreatDismal gave me nightmares as a kid. It’s great!

[Gene Kim @RealGeneKim]

Fleming describes the fantastic Anthropic Claude Artifacts interface, and how much he learned from the 3,000 word "system prompt" that, among other things, guides Claude on when and when not to use Artifacts.

"Anthropic has this web interface like ChatGPT, but you can say, 'Build me a website.' It will build the website and open it up in a side panel and an iframe, and it works. I mean, it's just running right there, you can interact with it."

"What surprised me when I investigated this was the main secret source to this is actually just a prompt. It's just something that they've tacked onto the system prompt to make it understand and be able to use artifacts. That prompt is really really big."

"This was my first experience of seeing, I guess, a production prompt that is required when you're going to be putting a tool based on an LM in front of tons of people. Some of whom will be actively trying to subvert what you're doing, and some of whom will just be dumb, and some of whom will be trying things that never occurred to you.

"You need the LLM to stay on track in all those kinds of different cases, and that prompt is massive. It's like 3,000 words; it's about 10 pages of code. When I skipped through it, I was just like, 'Wow, this is incredible.'"

You can see the full prompt here: https://t.co/ug5etTgZQJ

[yan @bcrypt]

reminder that the bcrypt hash function ignores input above a certain length! so if you do bcrypt(username || password) for some reason, a sufficiently long username will make it accept any password. to fix this you can sha256 the input first.

[Anthropic @AnthropicAI]

Claude can now view images within a PDF, in addition to text.

This helps Claude 3.5 Sonnet more accurately understand complex documents, such as those laden with charts or graphics.

Enable the feature preview: https://t.co/bJ8BjBT6zG.

[Garrett Robinson @garrettr_]

@josephbonneau Congratulations!

[SecureDrop @SecureDrop]

SecureDrop just wrapped up its 6th security audit! Only 3 vulnerabilities (2 low-severity, 1 medium) were found and fixed in the 2.10.0 release. Big thanks to @7ASecurity and @OpenTechFund for supporting our work to protect journalists and whistleblowers.

https://t.co/AZervTKUJc

[Anthropic @AnthropicAI]

Over the past few months, our Interpretability team has put out a number of smaller research updates. Here’s a thread of some of the things we've been up to:

[Garrett Robinson @garrettr_]

@ManishEarth Big ups to @kqedbaycurious for their episodes on this and the rest of the state props, they really helped me understand where there was more than meets the eye on this and some other props on this year’s ballot.

[Garrett Robinson @garrettr_]

@ManishEarth This one really ticked me off because it’s so sneaky, you would never know it’s targeted at the AIDS healthcare foundation for their rent control activism just by reading the description on the ballot or even in the official voter guide (!!)

[Feross @feross]

🚀 Big news! Socket now officially supports Java! 🎉

💻 We’re bringing our developer-friendly security platform to the most widely used language in the enterprise ✨

🤩 Whether you’re building large-scale apps or shipping Android apps, @SocketSecurity has got your back 💪

1/6

[yan @bcrypt]

there is a scam going around where someone says they are interested in hiring you, but insist on using "brave talk" / "wechat" / etc. to do a call. when you click to join the call, it says you have to download an installer which is actually malware.

[Dario Amodei @DarioAmodei]

Machines of Loving Grace: my essay on how AI could transform the world for the better

https://t.co/CFGd4Sbq86

[Alex Albert @alexalbert__]

"The [model] that I would say is net best is 3.5 Sonnet, I think that's a consensus opinion"

Appreciate the Claude shoutout @amanrsanger, the full episode was a great listen!