Blog
Writing about AI — how it works, where it helps, and what we learn building with it.
Latest posts
TSMC Beat Its Own Q3 Guidance by $900 Million. September's 0.6% Decline Is the More Informative NumberTSMC's October 8 revenue report closes a record third quarter at US$46.71 billion, above the top of its own July guidance and ahead of consensus. It also shows September revenue down 0.6% month on month while up 54.6% year on year — the signature of a supply ceiling rather than a demand surge.Artificial intelligence
Manus Replaced Meta's $2 Billion With $500 Million From Home. Beijing's April Order Now Looks Cheap to EnforceButterfly Effect, Manus's parent, closed a round of more than $500 million on October 8, co-led by Boyu Capital and IDG Capital with Tencent, HSG and ZhenFund returning. Every named investor is based in China or Hong Kong — six months after Beijing ordered Meta's $2 billion acquisition reversed, the replacement capital came entirely from home.Artificial intelligence
OpenAI Put GPT-6 in Front of 1.2 Billion People and Changed What an Answer Looks Like. The Component Library Is the Part That MattersGPT-6 reached paid ChatGPT tiers on Wednesday and free users today, carrying a feature called Intelligent UI that composes answers from charts, forms and working tools instead of text. The streamable component library behind it, not the model, is the consequential decision.Artificial intelligence
Attackers Hijacked Three Country-Code Registries and Got Valid Google Certificates. Nothing in the Chain MalfunctionedAttackers compromised the third-party operators of .gh, .sl and .as, changed the authoritative DNS records, and used that control to obtain publicly trusted HTTPS certificates for Google and YouTube domains. Google says neither its own systems nor the issuing certificate authorities did anything wrong — which is precisely the problem.Security
Anthropic Cut Haiku's Price to Exactly Match GPT-6 Luna. The Fine Print at 100,000 Tokens Is Where They Part CompanyClaude Haiku 5.5 arrives at $0.10 per million input tokens — a tenth of its predecessor, and the same base rate as OpenAI's GPT-6 Luna. The new two-tier pricing, and the labs missing from the benchmark table, complicate the headline.Artificial intelligence
Atlassian's CVSS 9.3 File-Read Flaw Is Being Exploited. The Advisory Lists Every Version of Eight Products as AffectedCVE-2026-21589 lets an unauthenticated attacker read files from the web root of eight self-hosted Atlassian products, and a security firm says exploitation attempts hit its honeypots within two hours of a public proof-of-concept. Atlassian Cloud is unaffected; every released version of the Data Center products is not.Security
DeepSeek Asked for 50 Billion Yuan and Is Being Offered Twice That. One of the Biggest Cheques Is Reportedly From a Battery MakerBloomberg and Reuters report that DeepSeek is closing on at least 80 billion yuan from Tencent and battery maker CATL, double its own target, ahead of a STAR Market listing. Moonshot AI has closed its final private round at $50 billion and filed for Hong Kong. Both are about to acquire something no frontier lab anywhere currently has: audited financials.Artificial intelligence
OpenAI Has Published 722 Mathematical Manuscripts. The Body It Consulted Asked It to Stop a Week EarlierOpenAI released 722 manuscripts in 372 result families from an unnamed internal model on October 6, seven days after the advisory group it convened asked frontier labs to stop testing advanced mathematics on proprietary models. The release meets some of the group's recommendations, defers others, and keeps the repository, the model and the problem selection in the lab.Artificial intelligence
Mistral's Large 4 Is the Best Model in the World at Reproducing Software Vulnerabilities. The Weights Go Public at the End of the MonthMistral has previewed Large 4, a 1.05-trillion-parameter open-weight model, and the capability it leads on is security work that Claude Opus 5.5 and GPT-6 Astra refuse to do. Mistral scores 82 per cent on a vulnerability-reproduction test where those two score near zero — and says it will publish the weights within three weeks.Artificial intelligence