Murat Beshtoev

I need to make a confession. @elonmusk I think people need to know. This is not a hoax. After almost 6 years of studying, in 1997 I defended my M.Sc. dissertation on “Masking nuclear explosions as an earthquake on a junction of tectonic plates.” (Cont’d)

Claude Code + Obsidian is the most powerful AI combo I've ever used. I literally built an AI second brain that runs my entire life. Inspired by Andrej Karpathy's LLM Knowledge Wiki, this tool has been a complete game-changer. Here's EXACTLY how to build one for yourself:

Yes @AnthropicAI would need ~7× more compute to serve Mythos to every Claude user That's why it's gated to 50 partners The math behind the gating, in The Compute Confession 👇 bepresearch.substack.com/p/the-compute-…

Super interesting technology: Exclusive | 'Ghost Murmur,' a never-used secret tool, deployed to find lost airman in Iran in daring mission nypost.com/2026/04/07/us-…

Claude Mythos just obliterated every single benchmark in AI. I can't believe what I'm reading.

AI model error rates are cutting in half every 6 months. At that pace, superhuman accuracy across the board isn't a question of if. It's when

A viral tweet claimed that Meta is now responsible for nearly a third of Anthropic's $30B ARR, based on reporting that Meta used 60.2T tokens over 30 days. This doesn't actually pencil out. It assumes all of Meta's spend is on output tokens, which are much more expensive than input tokens, and which would be extremely unusual — according to OpenRouter data, about 98.9% of all Opus 4.6 tokens are input. The actual number is *maybe* closer to $136M a month, or $1.6B a year. @tylercosgrove goes through the math:

Ronan Farrow joins Anderson to discuss his new reporting in The New Yorker that looks at allegations of persistent doubts about the head of OpenAI, Sam Altman.

this is the MOST important 4 minutes you’ll watch on AI this year. anthropic built a model so good at finding vulnerabilities they didn’t release it to the public. >CLAUDE MYTHOS PREVIEW it’s unreleased to the public and here’s what it did in a few weeks: >found a 27-year-old vulnerability in OpenBSD >caught a 16-year-old flaw in FFmpeg that automated tools missed after 5 million tests >chained together multiple linux kernel exploits autonomously. no human steering. AWS, google, microsoft, apple, nvidia, crowdstrike, JPMorgan. all got access. Anthropic committed $100M in credits to let these companies hunt vulnerabilities in their own systems before attackers do. >93.9% on SWE-bench verified. >77.8% on SWE-bench pro. nothing else is comes remotely close. Anthropic just pulled away in this AI race…

In science, there's an "unlimited number of discoveries to be made." Edison's co-founder Andrew White says integrating AI could accelerate those possibilities.

@Michell65664629 @Tablesalt13 This already exists in some form. Agree, it will probably get worse.

@Tablesalt13 @CirclEdgeInc How long until they exit tax us!!!? 😂 (or institute capital controls for money leaving the country …… or both). My bet is we’ve got a year and no more.

NEW There are now an estimated 1.1 million Canadian professionals living in the USA, 40% of which are top 1% earners 3x higher than any other country and, Canada is now bleeding 65,000 pros every year, a 50 year high. -HRD study

@CRRJA5 They used actual cameras not GoPro

🌍 100% REAL — straight from NASA (not AI!) 🌍 1972 vs 2026: Earth from space, 54 years apart. Left: Iconic “Blue Marble” from Apollo 17 — full daylight, Africa shining bright. Right: Fresh shot from Artemis II crew aboard Orion (April 2026) as they head to the Moon. Same planet. Different lighting, cameras, and perspective — but still breathtaking. Technology evolved. Earth’s beauty hasn’t.

When Dario bans your org from using Claude Code

Incredible growth pace because they were focusing on one vertical and solved a really hard problem

Holy Cow Anthropic $30 Billion ARR Has it caught OpenAI? Probably

Anthropic just announced it has crossed $30 billion in annualized revenue, and it is only the beginning of April. Two months ago, that number was sitting at $14 billion. Before that, it was $9 billion at the end of 2025 and it was barely $1 billion in late 2024. The growth curve has gone from impressive to nearly vertical in the span of a single year. Also, Anthropic just signed the largest compute deal in the company's entire history. They have locked in multiple gigawatts of next-generation TPU computing power through a three-way agreement with Google and Broadcom, with that capacity coming online starting in 2027. To understand the sheer scale of what that means, the average American nuclear plant generates roughly one gigawatt of power. Anthropic is reserving more than three times that amount in pure computing electricity, dedicated entirely to running its AI models. The really striking part is not the size of the deal but rather the diversity of the hardware strategy behind it. Anthropic trains and runs Claude simultaneously on AWS Trainium chips through Project Rainier, on Google TPUs manufactured by Broadcom, and on NVIDIA GPUs across multiple data centers. Claude is now the only frontier AI model in the world available on all three of the largest cloud platforms at once: Amazon Web Services Bedrock, Google Cloud Vertex AI, and Microsoft Azure Foundry. That kind of multi-cloud, multi-chip architecture is not an accident but a deliberate hedge against supply bottlenecks and a signal of how seriously Anthropic is treating infrastructure as a competitive weapon. Back in December, Broadcom's CEO revealed a mystery customer had placed a $10 billion custom chip order and then quietly disclosed it was Anthropic. That was followed almost immediately by another $11 billion order in the same quarter. Now the total commitment has expanded again with this new multi-gigawatt deal set to come online in 2027. Broadcom CEO Hock Tan has told investors the company expects to generate close to $100 billion in AI chip revenue in 2027, with Anthropic as one of its primary drivers. The customer growth numbers are the most alarming signal of all for Anthropic's competitors. When Anthropic raised $30 billion at a $380 billion valuation in February, it disclosed that 500 business customers were each spending over $1 million per year on Claude. Today, less than two months after that announcement, that number has already doubled to more than 1,000 enterprise customers at that same spending threshold. Eight of the Fortune 10 companies are already running critical workloads on Claude, which means the world's most powerful corporations have quietly made Anthropic a core piece of their operations. The forecast that looked wildly optimistic in January is now starting to look like it may be too conservative. Internal projections earlier this year suggested Anthropic might reach $18 billion in annual revenue for all of 2026. The company has already blown past that figure as a run rate before the second quarter has even started. Its own bullish internal forecast called for $55 billion in 2027, a number that, at the current trajectory, no longer sounds like science fiction.

Something to keep in mind

Microsoft has copilots for copilots. How do you manage all of this?

@pmddomingos At least this could transform (transformers again!) our electric grid

Blowing trillions on data centers is a very expensive way to discover that AI can’t be solved by brute force.

@pmddomingos Revenge of the Transformers (mathematical and metal)

Transformers are not all you need. You also need transformers.