Today, we share a breakthrough on the planar unit distance problem, a famous open question first posed by Paul Erdős in 1946.
For nearly 80 years, mathematicians believed the best possible solutions looked roughly like square grids.
An OpenAI model has now disproved that belief, discovering an entirely new family of constructions that performs better.
This marks the first time AI has autonomously solved a prominent open problem central to a field of mathematics.
Personal update: I've joined Anthropic. I think the next few years at the frontier of LLMs will be especially formative. I am very excited to join the team here and get back to R&D. I remain deeply passionate about education and plan to resume my work on it in time.
You @ClaudeDevs withdraw the free credits you give us and then started charging us without any notice. When we try to make a complaint we get a chat that did not response and there is no way to get a human. That's not even legal #claudecode@tuQuejaSuma
"MACRI, TERRIBLE HIJO DE P..." 😅
Carlitos Tevez recordó las negociaciones con el ex presidente de Boca y la deuda que le quedó cuando se fue al Corinthians. "Le habíamos ganado al Milán y yo le debía plata a Boca. Mauricio es una cosa increíble, nunca vista".
🎙️ @Liberotyc
Renunció la secretaria de Trabajo de Donald Trump por "conducta indebida". ¿Cuál fue su conducta indebida? Hizo viajes personales en avión financiados con fondos de los contribuyentes. 🌚
I'll be honest: I thought I could replace parts of my team with AI agents.
I kept trying. It kept failing.
What I'm actually seeing as a founder: my most AI-pilled employees are becoming more effective than they have ever been. The right employee who is 10x-ing with AI, using it better than I do and teaching me how, is more valuable than anyone I could hire today.
Most employees don't see this. Only CEOs and founders do. The frontier is moving so fast that the people learning it best are pulling away from everyone else on the team.
All the alpha right now is in being at the frontier edge of learning AI.
The way you get there is to follow the handful of people who are actually doing the work and distilling what matters. The ones I follow: Greg Eisenberg, Riley Brown, Lenny Rachitsky, Carl Vellotti, Hannah Stulberg, Ruben Hassid, Paweł Huryn.
The empty pattern is scrolling X for three hours a day and absorbing noise. The winning pattern is a few high-signal digests from people you trust who are testing the latest stuff so you don't have to.
That's why I've shifted most of my own time to using the latest tools and distilling what actually works.
The punchline: jobs for people who can use AI well are going up, not down. Same thing happened with Excel. Same thing happened with the computer. The people who learned them early got the high-paying jobs.
Be the 10x employee. Or hire one.
Hoy, programando con LLMs, hay dos tendencias:
Agentic engineering: entender tu sistema para generar código tan bien que casi no necesitas mirar lo que hace.
Vibe coding: no entenderlo. No mirar. Confiar en el LLM.
🚨 CRYPTO: NYT REVEALS ARGENTINE PRESIDENT MILEI MADE 7 CALLS TO LIBRA PROMOTER ON NIGHT OF TOKEN LAUNCH
The New York Times reported that Argentine President Javier Milei exchanged seven phone calls with crypto lobbyist Mauricio Novelli on the night he promoted the LIBRA token on X in February 2025. Federal investigators have named Milei as a person of interest. LIBRA briefly hit $4.6B market cap before crashing 90%.
Forensic analysis of Novelli's phone revealed a draft $5 million payment agreement tied to Milei's endorsement, monthly payments to Milei dating back to 2021 that doubled after he became president, and WhatsApp messages referencing payments to his sister Karina. Milei's government later dissolved the investigative task force probing the scandal.
A congressional committee found Milei provided "essential collaboration" to the project. Blockchain data showed eight wallets linked to LIBRA cashed out $107 million. Approximately 114,410 investor wallets suffered losses. Milei denies wrongdoing.
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.
I thought people were exaggerating. Then I burned $100+ of extra Claude usage in 1 hour.
What happened: agentic coding with Claude in Chrome enabled. 10-20x multiplier on already expensive operations.
A survival guide until the fix ships:
Claude Code (terminal):
→ npx @anthropic-ai/claude-code (bypasses the Bun cache bug)
→ Avoid --resume (breaks cache)
→ /compact to compress context mid-session
→ /effort medium for routine work, /effort low for simple tasks
→ --model sonnet for exploration. Opus for final passes only
→ /mcp — disable servers you're not actively using
VS Code / Cursor:
→ Same engine. Same bugs. Same fixes.
Cowork (desktop):
→ Default to Sonnet. Not every task needs Opus.
→ Auto-selects effort
→ Start a fresh session if context bloats
Dispatch (mobile):
→ No model picker in UI
→ Delete session, start new one if it spirals
→ You can ask it to dispatch sessions with specific models
The biggest lever most people miss: stop using Claude in Chrome for scraping. Dev-browser gives you DOM snapshots instead of screenshots. Text is 10-100x cheaper than pixels.
Anthropic knows. Fix is coming. Until then — /compact, /effort medium, and kill your unused MCPs.
Más de 10.000 docentes renunciaron a la universidad pública en los últimos dos años por los bajos salarios. Algunos hacen Uber. Otros se van del país. "Nunca hubo un desguace tan rápido y tan grosero, ni en los 90 ni en 2001". Odian la educación. Odian el conocimiento.