Domingos: AI is the leakiest abstraction yet, and what leaks is the LLM mess
pmddomingos · x · 2026-08-28
Pedro Domingos, prominent ML researcher, argues that AI is the leakiest abstraction yet — and what leaks through is the mess of the LLM itself. The point: users expect a clean encapsulated interface, but the underlying model's flaws and inconsistencies keep seeping into higher-level applications.
More from Models
- MiniMax-H3 on 8×H200: 1.95× Lossless Speedup, Up to 6.24× — ying11231 · 2026-08-28
- Deep Dive: Engram Enables Smaller Models to Reason Like Giants by Decoupling Memory — chocolateUI · 2026-08-28
- Grok-4.6 Ranks #15 in Agent Arena with 13% Success Rate Boost — arena · 2026-08-28
- Opinion: Model choice should be boring infrastructure — route by task, not by vendor — reddebtt · 2026-08-28
- Ling-3.0-flash-Fin: 124B Finance-Enhanced Model, Free API and Open Source — niacolhealth · 2026-08-28
- Study Finds LLM Confidence Tone Does Not Correlate with Correctness — ClickOk5811 · 2026-08-28