The Archive

Recycling startups are using AI to improve the recovery of critical minerals like aluminum, aiming to build a massive source of the metal.

Earlier this month, SpaceX and Anthropic announced a new compute partnership that provides access to the rocket company's Colossus data centers in Memphis, TN. Now, with the release of SpaceX's IPO filing, we have more details about that deal, including how much Anthropic is paying to Elon Musk's company. In its S-1 filing, SpaceX said that Anthropic agreed to pay $1.25 billion per month through May 2029 for access to SpaceX's AI training centers at Colossus I and Colossus II. That's $15 billion annually, or nearly double the $18.7 billion in revenue that SpaceX reported in all of 2025. The a...

That’s my own Android app. Bad, yet impressive. | Photo by Sean Hollister / The Verge Yesterday, I built my first Android app. Then, I made two more - three in one afternoon. For one, I literally typed 148 words into my web browser and walked away. Ten minutes later, I had an entire new app on my actual Android phone. I did have to prep that phone by enabling a USB debugging mode and plugging it into my PC, but as advertised, Google's AI Studio did literally everything else for me. I typed in words, I hit install, and voilà: an entire working program. I was nearly ready to agree with David, A...

Reddit user expresses frustration over unspecified Anthropic identity verification policy; lacks detail on actual policy change or decision basis.

Andrej Karpathy joins Anthropic to work on self-improvement mechanisms for Claude without human feedback.

Tencent releases Hy-MT2 multilingual translation models (1.8B–30B MoE) supporting 33 languages with 1.25-bit quantization.

Reddit user shares personal experience using Claude for computer engineering studies over a year.

AdventHealth deploys ChatGPT for Healthcare to reduce administrative workload and improve clinical workflows.

User reports 110 tok/s inference speed on Qwen 35B with 12GB VRAM using ik_llama.cpp, outperforming standard llama.cpp after MTP merge.

User shares personal workflow integrating bookokrat TUI ebook reader with Claude for summarization and study assistance.

Anthropic seeks $900B valuation; Q2 revenue projected at $10.9B, signaling strong commercial traction in enterprise AI.

Reddit discussion claims companies deploy AI at work to extract employee workflows and habits for eventual workforce replacement.

User reports delegating Claude Code tasks to Mistral/DeepSeek via vibe-skill tool, achieving 90% cost savings over 10 days while maintaining output quality.

Parag Agarwal (Parallel) discusses content valuation and creator incentives in an agent-driven web economy.

Meta says it needs to “offset the other investments we're making.” | Image: Cath Virginia / The Verge, Getty Images Meta has reportedly notified thousands of employees that they've been laid off as the company attempts to compensate for its hefty AI investments. In an email from Meta management shared by Business Insider, impacted staffers were told that the planned headcount reduction was part of the company's "continued effort to run the company more efficiently and to allow us to offset the other investments we're making." Reports of an upcoming wave of layoffs started circulating in March...

AMD announces Ryzen AI Halo developer platform and Ryzen AI Max PRO 400 series for on-device agent inference; availability details follow previous announcement.

OpenAI's GPT-next model solved the 80-year-old Erdős planar unit distance conjecture computationally for under $1000, demonstrating AI capability in pure mathematics.

User shares llama.cpp configuration for running Qwen 3.6 27B locally with ROCm acceleration and optimizations.

Comparative evaluation of coding agents (GitHub Copilot, Pi, Claude Code, OpenCode) using Qwen 3.6 27B isolates model vs. harness performance.

Reddit discussion comparing inventor dishonesty to AI researcher integrity; lacks specifics on claims or evidence.

Reddit anecdote about building a SaaS with Claude in 7 days; mostly satire/hype without technical substance.

Hobbyist trains DCGAN on 350 iPod Touch 4 photos of a red solo cup to study sensor artifacts.

Google Gemini 3.5 Flash tops Zapier Automation Bench, outperforming competitors at lower cost.

Reddit post comparing Descartes' philosophical argument against machine cognition to modern LLM capabilities; no new technical claims.

The SpaceX IPO is here, and it's more than just an historic public offering that could make Elon Musk the world's first trillionaire. It also reveals more ways in which Elon Musk's companies interact and overlap with each other, shuffling money around in ways that are often difficult to keep track of. This is evident in ways that are both obvious and less so. A CTRL-F search for "Tesla" yields 87 results, xAI is mentioned 356 times, and X 267 times. Even the Boring Company (7 times) and Neuralink (3) get a few mentions. Throughout its 330 pages of rocket launches and interplanetary wishes, yo...

Developer announces stability improvements to smallcode local LLM project after fixing 90+ bugs.

Vague speculation about year-end protests; no substantive claim or context provided.

OpenAI model disproves Erdős geometry conjecture; researcher claims breakthrough significance will compound through 2025.

Reddit user questions whether Claude's standard plans expose PII when used for accounting/financial services, citing data sharing concerns.