Domingos: Log-Perplexity Fell to 0.75 Bits While Cost Rose Orders of Magnitude
pmddomingos · x · 2026-11-23
Pedro Domingos uses OpenAI's newly published perplexity data to argue diminishing returns: log-perplexity dropped from 10 bits (bag of words) to 5 bits (best pre-GPT model), 1 bit (GPT-3.5), and 0.75 bits (GPT-6.2 Astra) — all while costs rose by orders of magnitude. A prominent skeptic's take on the transparency chart.
Related event: OpenAI Chart Leaks GPT-6 Astra as Researcher Questions Scaling Returns(2 posts)→
More from Models
- Voyager releases open-source shot-reader-8b multimodal model on Hugging Face — ravisparikh · 2026-11-23
- Matthew Berman baffled: Gemini weekly limits stay at 100% after a weekend of heavy Neon use — MatthewBerman · 2026-11-23
- 5 free uncensored GGUF models on Hugging Face: MiniMax, Qwen, Kimi, Nemotron — Promptmethus · 2026-11-23
- Pretraining sets the floor, RL raises the ceiling: on how the two divide the work — JacquesThibs · 2026-11-23
- Google AI Studio's periodic-table model selector reveals Gemini 4 Neon: 1M context, $3/$15 pricing — patloeber · 2026-11-23
- Nathan Lambert: RL Is 'Frying' Models and Reducing Their Diversity — natolambert · 2026-11-23