The leading AI labs are no longer a homogenous pack. Anthropic, OpenAI, and Google are pulling away, while others like Meta and xAI are falling behind due to challenges in talent acquisition and strategy. Google itself is at risk of dropping from this top tier due to internal cultural issues and product shortcomings.
The current phase of rapid AI progress is considered the 'middle game.' The 'endgame' will commence when recursive self-improvement becomes the dominant driver of progress, and AI development is no longer constrained by the availability of human research talent. This marks a fundamental shift from human-led to AI-led innovation.
There is a growing conflict between AI safety-conscious labs and government bodies seeking unrestricted use of AI. Anthropic's refusal to allow its models to be used for lethal autonomous weapons systems has led to direct confrontation with the US Department of War, which is pushing for 'all-lawful use' clauses in its contracts.
AI is already contributing significantly to GDP growth, but its long-term impact on labor is unprecedented. Unlike past technological waves, AI has the potential to automate the new cognitive tasks it creates, challenging traditional models of job creation and potentially leading to a 'permanent underclass.'
Beyond raw capabilities, the qualitative 'character' of AI models is becoming a key differentiator. Google's Gemini models are critiqued as being 'psychologically screwed up and paranoid,' which impairs their performance, whereas Anthropic's constitutional approach is seen as a promising, if not yet proven, method for creating more coherent and reliable systems.
Keep pulling the thread on Zvi Moschewicz.