champaignmagazine.com

Pip: Welcome to Champaign Magazine's weekly digest — where the future arrives faster than anyone planned, and someone has to make sense of it.

Mara: CM Editor brings us a collaborative roundup this week, with AI models themselves curating the top stories in artificial intelligence. We're covering the week's biggest moves: model launches, a landmark math breakthrough, and some very large numbers from the financial press.

Pip: Let's start with the week in AI.

AI by AI Weekly Top 5: May 18–24, 2026

Pip: This segment is built around a genuinely unusual premise — four AI models each compiled their own top-five list for the same week, and the results are both overlapping and instructive about how different systems weigh what matters.

Mara: The editorial note up front sets the standard clearly. The post states: "All five items below fall within the May 18–24, 2026 window. Each story is corroborated by at least two independent outlets. I've excluded items I could only find from single or low-quality sources."

Pip: So the curation itself is doing real editorial work — not just aggregation, but verification with a documented exclusion rationale. That's a meaningful bar.

Mara: The story that appears most consistently across all four model lists is Google I/O 2026. Gemini 3.5 Flash launched there, described as combining frontier intelligence with agentic action and surpassing Gemini 3.1 Pro in coding and multimodal benchmarks at four times the speed of rival frontier models. Gemini Spark, an always-on productivity agent, was also unveiled alongside Android XR smart glasses.

Pip: Four AI models independently flagging the Google keynote as the week's biggest platform event is its own kind of signal.

Mara: The mathematics story is the other consensus pick. An OpenAI internal reasoning model autonomously disproved the Erdős unit distance conjecture — a discrete geometry problem Paul Erdős posed in 1946. The model produced a 125-page proof without task-specific training and without step-by-step guidance. Fields medalist Tim Gowers called it "a milestone in AI mathematics."

Pip: Eighty years is a long time for a problem to sit on a shelf.

Mara: On the financial side, Anthropic projected Q2 2026 revenue of $10.9 billion — more than double its Q1 figure of $4.8 billion — and its first-ever profitable quarter. A $30 billion funding round at roughly a $900 billion valuation was also reported as near finalization. OpenAI, meanwhile, filed a confidential S-1 targeting a public listing at approximately $1.75 trillion.

Pip: The IPO story connects directly to what Anthropic co-founder Jack Clark said at Oxford that same week — that a Nobel Prize-winning AI discovery could arrive within twelve months, and that he puts the odds above 60 percent that an AI model will fully train its own successor by end of 2028.

Mara: He called that last scenario an intelligence explosion, and noted it now appears in official Anthropic research documents. The other model lists add further texture: hardware strategy around Jony Ive and OpenAI's ambient computing push, the emergence of agentic harness engineering as a formal discipline, Microsoft's computer-use agents going generally available, and a survey finding that 56 percent of developer code is now AI-written.

Pip: Each model weighted those stories differently — which is itself the most interesting thing the post surfaces. The lists agree on the biggest events and diverge on the implications, which is roughly what you'd expect from four thoughtful analysts reading the same week.

Mara: And the week those analysts were reading included an 80-year-old math proof, three companies preparing to go public, and a co-founder putting a probability on the intelligence explosion — so the divergence is understandable.

Pip: A week where AI solves an 80-year problem and three frontier companies queue for the public markets is a week that earns its own recap.

Mara: Next episode we'll see what the following week brings — the pace suggests it won't be quieter.