AI Research Pulse

AI reasoning breakthroughs: OpenAI disproves Erdős conjecture, Karpathy joins Anthropic, Google Gemini 3.5 Pro

AI reasoning breakthroughs: OpenAI disproves Erdős conjecture, Karpathy joins Anthropic, Google Gemini 3.5 Pro

Key Questions

What mathematical breakthrough did OpenAI achieve?

OpenAI's general-purpose model disproved an 80-year-old Erdős conjecture through a creative mathematical leap. DeepMind also solved nine additional problems using formal verification methods.

What role did Andrej Karpathy take at Anthropic?

Andrej Karpathy joined Anthropic to lead pre-training research. This move strengthens Anthropic's efforts in advancing model capabilities.

What features will Google's Gemini 3.5 Pro include?

Google's Gemini 3.5 Pro will feature a 2 million token context window and Deep Think reasoning capabilities. It is scheduled for a June launch and signals advances in long-context reasoning.

OpenAI's general-purpose model disproved an 80-year-old Erdős conjecture, a creative mathematical leap. DeepMind solved nine more problems via formal verification. Andrej Karpathy joined Anthropic to lead pre-training research. Google's Gemini 3.5 Pro with 2M token context and Deep Think reasoning is nearing a June launch, signaling a new frontier in reasoning capabilities. These advances accelerate AI's scientific contribution and competitive dynamics.

Sources (2)
Updated Jun 7, 2026