Top Stories
OpenAI’s AI Model Autonomously Disproves 80-Year-Old Erdős Conjecture
OpenAI announced that one of its internal reasoning models autonomously disproved a central conjecture in discrete geometry that Paul Erdős posed in 1946. The problem asks how many pairs of points among n points in a plane can be exactly distance 1 apart. For nearly eight decades, mathematicians believed square grids were essentially optimal.
The model discovered an infinite family of constructions using deep algebraic number theory — specifically Golod-Shafarevich theory and infinite class field towers — that achieve polynomial improvement over square grids. Princeton mathematician Will Sawin subsequently refined the result, confirming the mathematical validity.
Fields medalist Tim Gowers called the proof “a milestone in AI mathematics.” This isn’t a model retrieving an existing solution or verifying a human hypothesis. It’s an AI system finding a novel construction in a field where the best human mathematicians had been stuck since 1946. The implications for mathematical research — and the broader question of machine creativity — are significant.
Source: Roborhythms, ExplainX
Meta Begins 8,000 Layoffs as Zuckerberg Bets $145 Billion on AI
Meta has started cutting 8,000 jobs — roughly 10% of its workforce — with an additional 6,000 open positions cancelled. Another 7,000 employees are being reassigned to AI-focused teams. In a memo to staff, Mark Zuckerberg wrote that “success isn’t a given” in the AI race, framing the cuts as necessary to fund the company’s AI ambitions.
The restructuring accompanies a staggering capital expenditure increase: Meta has raised its 2026 CapEx guidance by up to $10 billion, reaching as high as $145 billion, almost entirely directed toward AI infrastructure. The scale of the bet is remarkable — Meta is simultaneously eliminating human roles and massively expanding its compute footprint, a pattern that’s becoming the template for Big Tech’s AI pivot.
Microsoft’s Suleyman: All White-Collar Work Automated Within 18 Months
Microsoft AI CEO Mustafa Suleyman told the Financial Times that most tasks involving “sitting down at a computer” will be fully automated by AI within 12 to 18 months. He specifically named accounting, legal work, marketing, and project management as vulnerable, predicting “human-level performance on most, if not all professional tasks.”
Suleyman’s timeline is aggressive even by Silicon Valley standards, and it’s worth noting that Microsoft has a direct financial interest in selling AI automation tools. But the prediction is notable coming from someone running one of the largest AI operations in the world. He also suggested that creating custom AI models would soon be as simple as “creating a podcast or writing a blog” — a claim that will test well against reality.
Source: Fortune
Quick Hits
-
Anthropic’s Clark predicts AI Nobel discovery: Co-founder Jack Clark told an Oxford University audience that AI will collaborate on a Nobel prize-winning discovery within 12 months. He also warned that AI still has “a non-zero chance of killing everyone on the planet.” ResultSense
-
Karpathy starts at Anthropic: OpenAI co-founder Andrej Karpathy officially began work at Anthropic this week, joining the pre-training team under Nick Joseph. He’s building a team focused on using Claude to accelerate pre-training research. TechCrunch
-
MCP Security launches: Trust3 AI announced Model Context Protocol Security to protect enterprise agentic AI workloads, as the agentic AI attack surface continues to expand. Help Net Security
-
Colorado AI law delayed to 2027: Colorado’s AI legislation, originally set for mid-2026, was pushed back to January 1, 2027 after Governor Polis signed SB 189 on May 14, significantly scaling back the law’s scope. A federal court had also stayed enforcement in April. Hunton Privacy Blog
Worth Watching
The Erdős conjecture result is the story to track here. Previous AI math achievements involved verification or incremental improvements. This is a novel construction that moved a field forward — the kind of result that, if it holds up to full peer review, changes the conversation about what AI systems can actually do in research. Meanwhile, Meta’s layoffs and Suleyman’s predictions paint a picture of an industry that’s simultaneously promising human-level AI and eliminating the humans. The tension between those two narratives will define the next year.