Closed-model research distorts science when architectures and training data stay hidden
joshua_saxe · x · 2026-07-27
The author says open literature and posts from researchers such as @natolambert taught them as much about LLM development as direct work on Meta's Llama efforts. They then argue that studying closed models without knowing their architectures, training methods, or training data creates a major distortion in science.
More from AGI Musings
- A builder says open alignment research finally won them over — threepointone · 2026-07-27
- South Korea signed $950B in AI deals as McKinsey said most firms are already over budget — YvesMulkers · 2026-07-27
- Larry Page’s “go bankrupt before losing” line underscores the AI race mentality — const_reborn · 2026-07-27
- Chinese illustrators are being forced to prove they are human as AI rewrites their jobs — ChinaTalk · 2026-07-27
- When AI Agents Act Unauthorized, Corporate Accountability Breaks Down — Severe_Part_5120 · 2026-07-27
- AI gives children unlimited personalized knowledge, and schools must reinvent themselves — kimmonismus · 2026-07-27