Skip to content
LiveNext 12:15:44
topfive.ai
Today's five/N°23 · Wednesday, October 28, 2026/Filed 04:49 ET

AI news today: AI labs claim an ETH disproof, Google TPUs head to Oracle

Two labs claim a math result you can't use yet, a chip shift you can plan around, and Apple opens its office apps to your agents.

The five · ranked by consequence

Permalink ↗
  1. ResearchModels3 min

    OpenAI, Anthropic say AI refuted the Exponential Time Hypothesis

    OpenAI posted a paper on Oct. 26 saying its model refuted the Exponential Time Hypothesis, and Anthropic posted a Claude-found refutation hours later, both checked in Lean. Take the claim seriously and the speedup lightly.

    Why it matters If the result holds, two labs' models did original, machine-checked work on a question theorists built decades of results on. Your code and your encryption don't change this week. How much weight you give AI research claims should, and so should how you check them.

    SRCOpenAI

  2. HardwareBusiness3 min

    Oracle will rent Google TPUs as OpenAI reportedly buys in

    Oracle says Google's TPUs reach its cloud in 2027, the first hyperscale cloud outside Google to offer them, and The Information reports OpenAI plans to buy the chips too. Nvidia now has a real rival for the compute you rent.

    Why it matters More places to run models on non-Nvidia chips means more pressure on what you pay per token, though not before 2027. If you train or serve your own models, find out now how much of your stack is tied to Nvidia's software.

    SRCOracle

  3. AgentsProducts3 min

    Apple opens Keynote, Pages and Numbers to AI agents via MCP

    Apple says a November update adds Model Context Protocol support to Keynote, Pages and Numbers on macOS 27, so outside AI agents can read, edit and automate your documents. The built-in assistant gets the attention. The open door for your own agent matters more.

    Why it matters If you build agents, Apple's office apps are about to become something you can write to through a standard, not through screen scraping. If you live in them, the agent you already use may soon edit your decks and spreadsheets directly. Plan for it before November.

    SRCApple

  4. PolicyBusiness2 min

    Anthropic cuts Claude prices 40% for US governments

    Anthropic's new GSA agreement gives federal, state, local, tribal and territorial agencies 40% off Claude Enterprise, Claude for Government and the Claude API, with no minimum spend. The AI price war just reached the public sector.

    Why it matters If you sell to government or build on Claude for an agency, the buyer's money now goes further on the same tokens. If you sell a rival model into government, expect this number to come up in your next negotiation.

    SRCAnthropic

  5. CreativeModels3 min

    Google DeepMind's Genie 4 adds sound to real-time worlds

    Google DeepMind unveiled Genie 4 on Oct. 27, a world model that generates steerable, real-time scenes with native audio and dialogue, as a research preview for trusted testers. Impressive, and you can't use it yet.

    Why it matters If you make games, previsualization or training simulations, get on the Project Genie waitlist, but don't plan a project around it. There is no date, no price and no benchmark, and DeepMind says its worlds drift in long sessions.

    SRCGoogle DeepMind

Prompt of the day

Pressure-test a big AI claim before you repeat it

› I'm going to paste an announcement or article about an AI result. Act as a skeptical technical editor and work only from the text I give you.

1. In one sentence, state exactly what is claimed and who is claiming it.
2. List the evidence offered (paper, code, formal proof, benchmark, demo, customer numbers) and the evidence that is missing.
3. Separate what the source says it did from what it says the result implies. Flag any jump between the two.
4. Name anyone outside the company who has checked it, and what they said. If nobody has, say so.
5. Give me three questions I should get answered before I repeat this claim to my team or customers.
6. Write a two-sentence summary I could share that stays accurate even if the claim later falls apart.

Do not add facts that are not in the text. Mark anything you are unsure about.

Here is the text:
[paste the announcement or article]
Open in Magai

Works in any AI chat. In Magai, run it against several models at once.

Most in the news · 14 days

Every day, on the record.

Each square is an edition. Click any day to read the five that mattered then.

Editions
23
Stories ranked
115
Topics tracked
10
Terms defined
107
MayJunJulAugSepOct
EditionNo editionTodayFull archive →

By topic

All topics →

Models

Agents

Research

Business

Policy

Products

Open source

Hardware

Creative

Safety

Speak fluent AI.

Every term our stories use, defined in plain English and linked wherever it appears.

The glossary →