Godot Isn’t Making it

Source: Where’s Your Ed At Date: 2024-12-03 URL: https://www.wheresyoured.at/godot-isnt-making-it/

Summary

Zitron argues generative AI has reached peak capability with diminishing returns — transformers face insurmountable training data limits, scaling laws have failed to deliver expected improvements. OpenAI’s GPT-5 didn’t hit desired performance. OpenAI spends $2.35 per $1 earned; Anthropic burns $2.7B annually; GitHub Copilot loses $200M annually. NVIDIA’s Blackwell chips face critical cooling failures months before deployment.

Implications

• The scaling law failure claim is the most consequential technical assertion. If diminishing returns from scaling are real, the entire investment thesis — more compute → better models → more revenue — breaks.
• GitHub Copilot losing $200M annually is the enterprise unit economics benchmark. Microsoft has the largest enterprise distribution in software history. If they can’t make AI coding tools profitable, nobody can.
• The cooling failure for Blackwell is a physical constraint that delays deployment timelines regardless of financial pressure.