The Diff
Archives
Search...
Log in
Subscribe
No. 050: The Lean community certifies the Claude Riemann proof. Someone shrunk Inkling 2 Mini to 5.4GB.
November 22, 2026
The Lean and Mathlib communities certify the Claude-assisted Riemann Hypothesis proof, independent kernels and all. And Inkling 2 Mini shrinks from 20GB to...
No. 049: An Anthropic mathematician says Claude proved the Riemann Hypothesis. Inkling 2 Mini loops a 26B onto your laptop.
November 21, 2026
An Anthropic mathematician says Claude proved the Riemann Hypothesis, in 17.4 million lines of Lean, and Scott Aaronson bets it will stand. And Inkling 2...
No. 048: Anthropic's first day: $111.50 open, $97.17 close. GPT-6.2's price cut meets its first backlash.
November 20, 2026
The Diff No. 048: Anthropic’s first day on Nasdaq: opened at $111.50, closed at $97.17, seventeen percent above the record $83 price. And three days into...
No. 047: Anthropic priced at $83 - the biggest IPO on record trades today. Gemini 4 Neon at $3 a million.
November 19, 2026
Anthropic priced the biggest IPO on record last night - $83 a share, above its range, upsized to 980 million shares. It starts trading on Nasdaq today. Hours...
No. 046: Nvidia's $115.8B quarter, and the board shuts the door on Huang. Anthropic prices today.
November 18, 2026
Nvidia posts its first quarter since Jensen Huang’s resignation - revenue double a year ago - and its board chair uses the earnings call to rule out bringing...
No. 045: Qwen4-Max's million-token context. BIS targets remote chip access.
November 17, 2026
Alibaba releases Qwen4-Max and HN votes it straight to the top - while Washington makes renting restricted chips through an overseas data center an export-...
No. 044: A $3 Arduino, two coding agents and a 1994 Jeep ECU. The brawl over Nvidia's next CEO.
November 16, 2026
A maker points Claude Code and Codex at his 1994 Jeep’s undocumented engine computer and ends up controlling it from a laptop - while Washington and Silicon...
No. 043: Vibe-coded Lovable apps expose passports and 200,000 private chats. WSJ: Huang's "avalanche" and the CFO who replaces him.
November 15, 2026
A security research firm sampled 159,041 vibe-coded Lovable apps and found passport numbers, payment data and private chats readable without a login - while...
No. 042: Anthropic went down twice in a day - and some scheduled runs never happened. Perplexity publishes how it trains Decider v1.3, its decision model.
November 14, 2026
Anthropic’s roughest reliability day in months - two resolved incidents on Friday, a third still recovering as this issue ships - lands next to Perplexity...
No. 041: Gemini 4 Hydrogen beats Argon on the index at $1.50 in / $7.50 out - and AA clocks it at $3.16 a task. Jensen Huang resigns as Nvidia CEO, effective immediately.
November 13, 2026
An evening double-header: Google shipped Gemini 4 Hydrogen - its speed tier now beats Argon on the Artificial Analysis index, at $1.50 in and $7.50 out per...
No. 040: Google's Gemini Decision 1 answers 1,000 questions in one pass - no text, output free. GitLab ships the settings.local.json block Anthropic hasn't.
November 12, 2026
Google’s newest model writes no text at all: it scores up to 1,000 questions against one input in a single pass, in about 200 ms, at $0.10 per million input...
No. 039: Anthropic says it is prototyping secret scanning for Claude Code - and that auto mode is the real answer. Fable 5.2's report card: #1 on the index, $26.54 per task, one dev's $34 bill.
November 11, 2026
Anthropic answered the exposed-secrets report where it blew up - a 624-point Hacker News thread: secret scanning is being prototyped, no date, and "everyone...
No. 038: Anthropic says Fable 5.2 tops its coding benchmark - and still recommends Opus 5.5. Meta says Muse Ember matches Haiku 5.5’s best score at a third of the cost.
November 10, 2026
Anthropic released Claude Fable 5.2 and Mythos 5.2 on Nov 9 - same model, two doors - at $8/$40 per million tokens, and says Opus 5.5 is still the daily...
No. 037: Goldman pulls Claude Code from its bankers and traders - Bloomberg says the October guidance cut off everyone but engineers. Nvidia’s board investigates Jensen Huang.
November 9, 2026
Goldman Sachs pulled Claude Code and GitHub tools from its bankers and traders - Bloomberg reported the October guidance, and the same bank is a lead...
No. 036: Claude Code's "don't ask again" saves your password with the command - Apiiro says 1,122 of those files are public. The end of Microsoft's AI Frontiers.
November 8, 2026
Security researchers say 1,122 public GitHub files hold secrets Claude Code saved verbatim when users clicked "Yes, and don’t ask again." Microsoft...
No. 035: Postman reverts the forced agent-plugin install: "we are reverting this change." The terminal in your IDE is now one stable C library.
November 7, 2026
Postman says "we are reverting this change" after a week of backlash over silently installed agent plugins - the first full revert in the consent thread. The...
No. 034: Claude Code sends your email to websites that ask for it. A fix is claimed - users can't find it.
November 6, 2026
Claude Code users find their own email addresses in outbound web requests, the issue respondent says a deterministic block shipped - users can’t find it -...
No. 033: GitBook’s post-mortem documents the chain and softens the AI claim. OpenAI says its pipeline minted 948 Linux root exploits.
November 5, 2026
GitBook’s post-mortem documents the chain and softens the "AI agents" claim, OpenAI says its pipeline minted 948 working Linux root exploits, and a paying...
No. 032: A developer catches Postman installing plugins into his other AI tools. GitBook says AI agents breached it.
November 4, 2026
A developer catches Postman installing plugins into his other AI tools and approving its own hook, GitBook says AI agents breached it, Railway’s agents...
No. 031: GPT-6 Astra edits a model's weights by hand. WorkOS says an AI attack framework hit its CRM.
November 3, 2026
An MIT team has GPT-6 Astra reading and rewriting neural-network weights like source code, a breach notice blames an autonomous AI attack framework, and...
Older archives