AI safety @ Anthropic | AI investing @ Pioneer Fund

San Francisco, CA
Joined July 2009
Francesco Mosconi retweeted
As someone who usually disagrees with @DarioAmodei, I gotta give him *major* props for this one. “If you believe you have the right to force me—use your guns openly. I will not help you to disguise the nature of your action." This is what Hank Rearden, the steel magnate from Ayn Rand’s Atlas Shrugged, tells the government after disobeying its command to run his business in a manner contrary to his conscience. Amodei channeled Rearden’s spirit in a letter refusing to remove safeguards from Claude, as requested by the War Department.
23
192
6
1,858
168,809
Today feels like a turning point in motivation for why I do what I do. Deciding to drop everything else and focus my efforts on AI safety felt right then and feels even more right today as the stakes get real. Proud to work here!
1
17
365
Francesco Mosconi retweeted
I've updated towards a higher chance of much faster than trend AI progress in 2026. I also generally updated towards (somewhat) faster doubling times in the next 2 years. I was expecting ~170 day doubling times (5.7 months) and now I expect ~150 day doubling times (5 months).
We estimate that, on our tasks, Claude Opus 4.5 has a 50%-time horizon of around 4 hrs 49 mins (95% confidence interval of 1 hr 49 mins to 20 hrs 25 mins). While we're still working through evaluations for other recent models, this is our highest published time horizon to date.
23
38
4
549
55,500
Today we released Claude Opus 4.1! Our most powerful model yet.
Today we're releasing Claude Opus 4.1, an upgrade to Claude Opus 4 on agentic tasks, real-world coding, and reasoning.
1
3
454
We conducted voluntary automated testing to track capability progression and validate our safety assumptions.
1
1
102
Francesco Mosconi retweeted
So many things to love about Claude 4! My favorite is that the model is so strong that we had to turn on additional safety mitigations according to Anthropic's responsible scaling policy
83
43
73
1,032
318,971
Francesco Mosconi retweeted
Proud to announce that for much of this year Anthropic has been working with the National Nuclear Security Administration (NNSA, part of DOE) to test out whether LLMs like Claude know about dangerous things relating to nuclear weapons. axios.com/2024/11/14/anthrop…
4
36
5
235
38,768
Proud to keep working on making the most advanced AI safe!
Introducing Claude 3.5 Sonnet—our most intelligent model yet. This is the first release in our 3.5 model family. Sonnet now outperforms competitor models on key evaluations, at twice the speed of Claude 3 Opus and one-fifth the cost. Try it for free: claude.ai
2
334
Late to share, but absolutely proud of this work towards making AI safer and more interpretable!
New Anthropic research paper: Scaling Monosemanticity. The first ever detailed look inside a leading large language model. Read the blog post here: anthropic.com/research/mappi…
2
266
Yes. In hindsight it was obvious.
The AI race: Microsoft + OpenAI Google + DeepMind Tesla + xAI Meta + Meta AI and now Amazon + Anthropic History is unfolding in front of our eyes. The AI boom will be bigger than the internet.
1
400
If you’re surprised by the speed of AI development you’re not paying attention.
ChatGPT can now see, hear, and speak. Rolling out over next two weeks, Plus users will be able to have voice conversations with ChatGPT (iOS & Android) and to include images in conversations (all platforms). openai.com/blog/chatgpt-can-…
469
You want AI copilots to do what you don’t want to do to free you up to do what you are uniquely good at… … until they are good at everything you could possibly do 🤔
1
138
This image is all around twitter, and it took me a while to understand why. Step away from the screen, and step back until you notice it, after that you can't unsee it.
1
192
Generative AI makes creating these kind of images extremely easy. How will we protect ourselves from embedded subliminal messages everywhere?
159
At You.com we launched YouPro today! A subscription plan with access to all the newest AI tools including GPT-4 and Stable Diffusion XL! Get a month free with the code LIKEAPRO. Great work team!
3
12
1,327
Francesco Mosconi retweeted
Search the web, write, code, and create with AI like a⚡Pro. Introducing You⚡Pro, an AI chatbot with citations and real-time data, GPT-4 and Stable Diffusion XL. A 🧵
5
14
4
161
31,131
Really neat summary of AI political spectrum.
1/9 After my Saturday chat w/ @RokoMijic / @romanyam, I did a lot of thinking. The matrix below, as far as I can tell, will more or less DEFINE global politics from now until human extinction. From the motives of OpenAI to global government, let's unpack this: 🧵
1
1
6
521