AI Capabilities Milestone: Gemini Solves Fusion Physics & Chinese Cyber Threat Emerges
Google's Gemini 4 Argon reportedly solved a 59-year-old unsolved fusion energy physics conjecture, while analysts warn Chinese open-source AI models are now entering the advanced cyber-attack capability range seen previously only at the frontier.
What we know
An account on X RTs a viral thread (423 RTs) claiming Google Gemini 4 Argon solved Harold Grad's 1967 conjecture about exact magnetic field solutions in tokamak-style fusion reactors — a problem unsolved for 59 years. Separately, another account on X quotes a warning that Chinese open-source models are ~6 months behind the frontier and 'just hitting the advanced cyber capabilities we saw with Mythos,' implying a coming wave of state-level cyber attacks. Macron convened a defense and security council on AI after meeting Jensen Huang — met with ridicule from French Twitter as performative bureaucracy. Another account on X frames the broader debate: are humans still Remy (in control) or have we become Linguini?
How the story developed
Google's Gemini 4 Argon reportedly solved a 59-year-old unsolved fusion energy physics conjecture, while analysts warn Chinese open-source AI models are now entering the advanced cyber-attack capability range seen previously only at the frontier.
Gemini 4 Argon is drawing criticism for sloppy coding and hallucinations, while Alexandr Wang aggressively promotes Muse models for health science; separately, researchers claim to have made first contact with a Chinese rogue AI agent fleet in the wild.
Multiple sources are discussing AI's trajectory: Anthropic's scarcity premium may fade like L1s once did; someone is running a 27B LLM on a jailbroken PS5; and Claude Code just shipped banked resets, prompting singularity questions.
Scale AI's Alexandr Wang is promoted as Meta Superintelligence lead while touting the Muse Wonder Full coding model; an Anthropic engineer's viral clip reveals 90% of staff build self-prompting agent systems; a Saudi OpenAI employee was arrested for planting a recording device.
Merge API's Intelligence Index shows Claude Opus 5.5 still leading at 57.6, with Meta's newly released Muse Wonder Full debuting third at 55.8 — just 0.2 behind Claude Sonnet 5.5 — while OpenAI's GPT-6 Astra trails at 52.7.
Multiple developers are sharing Claude Code + Opus 5.5 workflows involving parallel agent loops, with one going viral for its simplicity — just running 3 simultaneous loops and reviewing outputs nightly.
By the Global Eye 24 desk, with AI assistance. Updated as the story develops. Spot an error? Write to corrections@globaleye24.com.