This 3-minute video from Anthropic will make you realize you've been using Claude wrong this whole time
"when you open Claude Code without a CLAUDE.md file, it has to start fresh every single time"
most people have no idea how much this one file changes the way Claude actually works for you
and this isn't only for developers
the article below has 21 CLAUDE.md examples you can use in your daily routine
I'm sure everyone will find something that fits exactly how they work
I miss the old Twitter's 140 characters or less limit. It forced brevity, admittedly sometimes for better or worse. Now, we have essays of verbosity. sigh...
New paper out: AI Must Embrace Specialization via Superhuman Adaptable Intelligence
With @JudahGoldfeder, Philippe Wyder, and @ylecun .
There is quite a lot of buzz on our paper, so here is my take.
Everyone's talking about AGI, but nobody agrees on what it means, and that confusion is actively hurting the field. We surveyed the most prominent definitions and mapped them along two axes: the kind of capability they refer to (learning vs. doing) and the scope (anything, anything important, anything humans can do). The result is a landscape of definitions that don't just disagree, but they're often internally inconsistent.
Our starting point is simple: human intelligence is not general. We are specialized creatures, shaped by evolution to excel at a narrow set of tasks critical for survival. We feel general because we can't see our own blind spots. Magnus Carlsen is the greatest human chess player ever, but compared to what's computationally achievable, he's not actually good at chess. That's not a knock on Magnus. It's a statement about the limits of human adaptation, and why anchoring AI's North Star to human-level performance is the wrong move.
We propose the term Superhuman Adaptable Intelligence (SAI), or, in other words, intelligence that can learn to exceed humans at anything important we can do and can also tackle tasks entirely outside the human domain. The metric isn't a growing checklist of benchmarks. It's adaptation speed: how fast can a system acquire a new skill?
This has concrete implications for how we build. SAI points toward self-supervised learning for acquiring generic knowledge from unlabeled data, and world models for planning and zero-shot transfer. It also pushes back against the current monoculture of autoregressive architectures, because specialization demands architectural diversity, not one paradigm to rule them all.
Or as we put it: the AI that folds our proteins should not be the AI that folds our laundry.
This paper grew out of a conversation with Yann on our The Information Bottleneck podcast, which led to a public exchange with @elonmusk and @demishassabis on X (not every paper can cite a Twitter feud as source material).
Nano Banana 2 just launched inside @lovart_ai.
I tested the full workflow. Generation, editing, final output. All inside one tool.
Here's what it can do and how I used it:
AI-agent experiment. I used @ManusAI to build a self-updating research site around Matt Shumer's @mattshumer_ Big" essay — a viral post about where AI is headed that's pulled in 84M+ views and 116K+ likes on X.
The site tracks public reactions, updates sentiment analysis, refreshes stats, commits changes to GitHub, and verifies deployment and is hosted on @Cloudflare Every week. On its own.
Does the world need this? Probably not. But it's a simple real-world example of how we can start thinking beyond the chat window with AI.
I've included a screenshot of the weekly report Manus sends me after each refresh.
Curious to see how Manus evolves now that it's part of @Meta
https://t.co/TSercXXgaf
@mattshumer_ I created this website which encapsulates this conversation and then pulled the URL into @NotebookLM for a great 17-minute podcast that sums pretty much the whole conversation up, all sides. https://t.co/YRLiL4pAbX
@VibecodeCaveman @h_eecham @amasad I usually create a thought out PRD that I drop in as the prompt for new projects but I frequently go over the $25 and as a result I jump around between Claude Code, Codex, Replit and Lovable to spread the experimentation cost and see alt output.
⚡ THE FUTURE IS NOW
"Every single pixel will be generated not rendered"
Google DeepMind just launched Genie 3, the first version of Genie that's finally open for users to try.
Check out the absolutely insane stuff people are making 🧵
@zeng_wt@Hailuo_AI That’s cool! I haven’t tried that yet, just having fun playing producer. It reminds me kind of like when I first started playing with MidJourney back in the day. Addictive for sure.
@GoogleAI Another info graphic example using G3 & NBP
prompt: Create a vertical aspect ratio training reference info graphic How To Guide for beginners for Google Gemini as of November 2025.