AI Hot Takes with Bengaluru's AI Builders: p(doom) Scores, Claude Code vs. Codex, Open Weight Models
summary
Group discusses concrete positions and tools: a community-calibrated p(doom)=33% belief about AI extinction risk, preference for open weights like Llama 2 and GPT set the bar, and practical tooling choices favoring Hugging Face ecosystems and K2.5 for code. One team reports running roughly 50,000 hours of data in week-one pilots and shipping foundation models for deepfake detection; another builds physics-based GPU telemetry and QC engines that already catch reliability regressions in production.
AI-written summary · about
