Jeremy Fox 🦊 (@JeremyDanielFox) header
Jeremy Fox 🦊 (@JeremyDanielFox) profile picture

Building Claude @AnthropicAI. My views are my own.

Joined July 2017
Photo by Jeremy Fox 🦊 (@JeremyDanielFox)Photo by Jeremy Fox 🦊 (@JeremyDanielFox)Photo by Jeremy Fox 🦊 (@JeremyDanielFox)Photo by Jeremy Fox 🦊 (@JeremyDanielFox)Photo by Jeremy Fox 🦊 (@JeremyDanielFox)Photo by Jeremy Fox 🦊 (@JeremyDanielFox)Photo by Jeremy Fox 🦊 (@JeremyDanielFox)Photo by Jeremy Fox 🦊 (@JeremyDanielFox)Photo by Jeremy Fox 🦊 (@JeremyDanielFox)Photo by Jeremy Fox 🦊 (@JeremyDanielFox)
Jeremy Fox 🦊 retweeted
Replying to @tszzl
more than vegas—i could care less what happens to the desert—it’s glorious that i can drink fiji water and shameful that it exists. it takes months for fiji water to reach new york. they constructed these brilliant logistics routes that do nothing but export this remote island’s groundwater to people who can’t differentiate it in blind tasting. nobody thinks this is the best capital allocation or worth eroding fiji’s land. it’s just a silly use of humanity’s godhood to transport water thousands of miles over an ocean
36
10
2
557
139,320
Great new article on memory from @RLanceMartin! Contains a lot of useful info for those looking to build memory into their agents today
Memory is what allows Claude to learn from experience. In a new Engineering Blog post, we share how “dreaming” helps Claude organize what it’s learned, correct its mistakes, and find what it needs more easily the next time. anthropic.com/research/compo…
13
1,008
Somewhat sadly, I think we’re seeing the end of Python as a programming language. When agents write the code, the language designed to be the most readable just isn’t as useful. You were fun, Python.
140
49
35
2,454
201,326
I say “somewhat” because we all know that TypeScript is a much better language than Python anyway 😎 😎 😎
2
97
22,541
Jeremy Fox 🦊 retweeted
when nick told me that he had some interp breakthroughs, he was uh, underselling it. i think this may be remembered as a major moment in scientific history. news so good even eliezer will have to lower his pdoom. we can start making alignment into a 'normal science'
interpretability might be the most important scientific problem in the world right now, as I don't think we can solve alignment if we don't know what we're building. I think I have good news about it. I co-authored early work in mechanistic interpretability at OpenAI and came away pessimistic, because neural networks appeared to fold into arbitrary tangled knots. years later, I started openai's first large-scale automated interpretability project, finding about 2% of neurons legible with the primitive tools we had with modern agentic tools and a much closer look, the picture looks totally different. the vast majority in models up to trillions of parameters are legible, and their circuits beautiful. many famous examples of tangled neurons turned out to be misunderstood. The roadblocks I expected mostly aren't there. I now think ambitious interpretability will be solved by the time we need it most, and possibly quite soon, as automated scientists keep improving. it may be a bitter lesson for interpretability: the simple thing, at absurd scale, works already-clean.goodfire.com/
163
395
84
6,617
439,937
If you think coding is saturated you’re not being ambitious enough in what you’re coding
Third iteration with Opus 5.5 medium. I am in awe. It turned blender, image-gen, and three.js into this beauty, which runs on your browser. An understatement to say that Anthropic cooked. You are looking at the future of game dev.
3
22
2,953
Congrats @dsp_ and team! MCP is here to stay.
I was not expecting things to go this way, but I think MCPs are better than CLIs for most integrations. The models have gotten much better at tool calling, we can defer tools & MCP is now stateless. If you need to compose/filter data, add params like query to your MCP tools.
6
897
Dean is right -- as models become more powerful, the bottleneck to further deployment will be safety. You cannot deploy a powerful model that you cannot trust, and you cannot trust a model that is not safe!
Some suppose that “safety” and “innovation” in AI are at odds. My suspicion is the opposite: the next generation of breakthroughs in AI will be in safety, alignment, and monitorability. Pushing the frontier forward from here will require dramatic innovations in safety.
3
524
Jeremy Fox 🦊 retweeted
We have reached an agreement with Anthropic to conduct an independent investigation of agent incidents at the company and of their models’ alignment properties. We will publish one or more reports that will share our findings and describe our terms of engagement.
189
249
71
2,843
568,631
I signed this. I think we can create an incredible future with AI, but it’s important to have the ability to slow down if necessary. pacingthefrontier.com/
24
5
45
1,782
Claude Science by @AlecTPhD is super cool
New Claude Science update is out. My favorite feature: Claude can now see CPU and memory usage across all its running kernels and adjust its own workloads. And you get the same view through the new compute dashboard. From the dashboard, you can interrupt (or kill) running kernels to reclaim resources and even send the agent feedback (e.g. "rerun using fewer CPUs").
3
570
Jeremy Fox 🦊 retweeted
"God gave me a sign" (remembering i live in San Francisco) "I mean, I was acasually influenced by the ASI at the end of time to maximize EV for humanity"
65
31
1
1,090
142,013
Jeremy Fox 🦊 retweeted
Within ~2 years, there might be >0.5 Manhattan Projects worth of philanthropic $$$ to spend on the biggest challenges, like cyber or biodefense. @nanransohoff is right: the hard part is finding enough good organizations to spend the $. You should build one!
New blog post: The third wave of American philanthropy Hundreds of billions of dollars in new philanthropic capital will soon become liquid. The OpenAI Foundation holds 26% of OpenAI, worth about $220B at today’s valuation. Anthropic’s seven co-founders have pledged to give away 80% of their wealth and have instituted the most aggressive donor matching program for employees in tech history. How much does this all add up to? And how meaningful is that in the context of philanthropy today? I was doing some simple napkin math to wrap my head around the scale of what’s coming, and radicalized myself in the process. I had dramatically underappreciated the scale of the philanthropic capital that’s about to become available and the corresponding gap in talent and organizations that will be needed to make the most of it. This piece aims to directionally sketch the scale of what’s coming, the gap in operational capacity needed to absorb it, and what we can do to fill it. (Link to full post in reply)
16
19
241
29,409
Jeremy Fox 🦊 retweeted
it does seem that this time, unlike others, it is fair to characterize it all as ‘late stage capitalism’
141
54
24
1,909
141,017
Jeremy Fox 🦊 retweeted
Opus 4.7 uses more thinking tokens, so we've increased rate limits for all subscribers to make up for it. Enjoy!
1,155
885
394
21,748
1,277,457
Oh, and one more thing:
Introducing Claude Managed Agents: everything you need to build and deploy agents at scale. It pairs an agent harness tuned for performance with production infrastructure, so you can go from prototype to launch in days. Now in public beta on the Claude Platform.
12
984
Jeremy Fox 🦊 retweeted
Replying to @wongmjane @BenLesh
Mistakes happen. As a team, the important thing is to recognize it’s never an individuals’s fault — it’s the process, the culture, or the infra. In this case, there was a manual deploy step that should have been better automated. Our team has made a few improvements to the automation for next time, a couple more on the way.
312
813
363
10,884
1,421,922
Jeremy Fox 🦊 retweeted
Spring is here. Go touch some grass. Be remote. Let Claude work in the office.
You can now enable Claude to use your computer to complete tasks. It opens your apps, navigates your browser, fills in spreadsheets—anything you'd do sitting at your desk. Research preview in Claude Cowork and Claude Code, macOS only.
13
4
154
11,260
Claude go brrrrrr
A small ship I love: We made Claude.ai and our desktop apps meaningful faster this week. We moved our architecture from SSR to a static @vite_js & @tan_stack router setup that we can serve straight from workers at the edge. Time to first byte is down 65% at p75, prompts show up 50% sooner, navigation is snappier. We're not done (not even close!) but we care and we'll keep chipping away. Aiming to make Claude a little better every day.
7
668