DeepSeek vs Codex: Token Waste Hits 200M+
▸ Anthropic's valuation reportedly lost $100B+ after Kimi emerged as a serious competitor. Insiders predict DeepSeek V4 Pro's official release could wipe out another $100B from the AI incumbent's market cap. [@aigc1024]
▸ DeepSeek consumes vastly more tokens than Codex — not due to model size alone, but because it is "dumb enough to make many more mistakes." Each instance can overshoot 200M+ tokens of wasted inference before reaching the correct answer. The blunt self-diagnosis: DeepSeek's intelligence gap relative to Codex directly drives excessive token burn. [@inside1024]
▸ The "Mian Shen" (冕神) phenomenon under scrutiny — AI community members are debating whether the cult-like reverence of certain Chinese AI influencers is an orchestrated personality campaign rather than an organic movement. The discussion reflects growing awareness of manufactured hype in the AI ecosystem. [@aigc1024]
Bottom line: The gap between frontier models is real enough that weaker models burn massive token budgets correcting their own mistakes — and the market is starting to price in the winners.
Top comments (0)