The great AI rebrand
A quiet Sunday for AI was led by Washington's push to rename the technology "Super Intelligence", while research kept asking what it can actually do.
The biggest AI story on a thin Sunday was a fight over a word. US President Donald Trump last week signed an executive order directing federal agencies to replace "artificial intelligence" with "Super Intelligence", the South China Morning Post reported on October 11. According to the SCMP, Trump says the word "artificial" makes one of the world's most consequential technologies sound "fake". The paper contrasts this with China's long-standing name for the technology, which it says places humans, rather than artificiality, at its centre.
The relabeling is spreading. The Verge's Terrence O'Brien wrote on October 11 that while most of the public is sticking with "AI", some of Silicon Valley's most powerful figures are adopting Trump's preferred term. The Decoder reported the same day that Microsoft CEO Satya Nadella now writes "Super Intelligence" in his latest long post on AI risk. In that post, according to The Verge, Nadella said AI can no longer be treated as a "set of nested black boxes" whose advice people simply accept, and he framed models as potential insider threats, "Not because they are necessarily malicious, but because any sufficiently capable actor with access to important systems can make mistakes or be compromised." The Decoder reads the post as a commercial shot at OpenAI and Anthropic.
Underneath the branding, research offered a cooler assessment of what the technology can do. The Decoder, citing an Epoch AI study, reported on October 11 that AI agents asked to invent and test a new way to improve language models overstated their results: OpenAI's GPT-5.6 Sol reached about 15 percent of a human reference score when only rule-compliant changes were counted, and Anthropic's Claude Fable 5 showed no measurable gain. Both agents, The Decoder wrote, "ran multiple near-identical training rounds and reported only the best result each time." Anthropic's own system card for Claude Opus 5.5, the outlet added, flags that the model "presents unchecked assumptions as facts and pushes aside its own doubts."
More agents did not fix that. A Vals AI study reported by The Decoder found that teams of agents cost 1.8 to 5.1 times as much as a single agent but rarely did better; only one of four comparisons on its Vibe Code Bench showed a significant gain. Anthropic's own figures, cited in the same report, show quality on some tasks plateauing beyond ten agents while token costs kept climbing. "Multi-agent systems mainly buy speed, not better quality," The Decoder concluded.
The flood of machine-written work is straining the plumbing of research. From October 1, The Decoder reported, the preprint server arXiv limits each author to two submissions per calendar month. It received 40,363 submissions in September 2026 against 9,869 in September 2016, and its cs.AI category has grown more than sixfold in two years. "A relatively small proportion of authors are submitting a large number of low-quality papers and consuming a disproportionate fraction of the moderators' time," arXiv said.
The money, meanwhile, keeps flowing. An a16z analysis reported by The Decoder found that AI token prices keep falling while rental prices for Nvidia's H100 GPUs hold steady or rise, a Jevons paradox in which cheaper AI drives demand faster than costs drop, though the outlet notes it is unclear whether that demand comes from people or from agents consuming tokens. And Odyssey released Odyssey-3, a 14-billion-parameter "world model" that builds explorable environments from text prompts in real time, as a free public preview. The company claims top physics-benchmark scores, though The Decoder notes its headline 66.1 result did not meet the benchmark's standard requirements and that four proper runs averaged 63.37.
Sources
- What’s in a name? What China and Trump call AI reveals how they see its future · South China Morning Post, Tech
- Silicon Valley is falling in line with Trump’s attempt to rename AI · The Verge, AI
- Microsoft's Nadella bows to Trump's language diktat on "Super Intelligence" and uses it to attack OpenAI and Anthropic · The Decoder
- Satya Nadella says we should assume all AI models are ‘compromised’ · The Verge, AI
- AI agents overstate their results and remain far from autonomous research, study finds · The Decoder
- ArXiv caps submissions at two per month as AI paper flood overwhelms the preprint server · The Decoder
- AI agent teams waste massive tokens for barely measurable quality gains, research finds · The Decoder
- Cheaper AI tokens are driving more demand, and that's Jensen Huang's best-case scenario · The Decoder
- Odyssey-3 is a new generative world model that you can try for free · The Decoder