Model outputs are no less buggy, but they now take hours to spot
jsuarez · x · 2026-08-04
The author says model outputs are still producing the same amount of garbage, but it is getting much harder to spot.
- Early models were obviously bad within seconds.
- By the start of this year, it took about 15 minutes to notice the garbage.
- Now it can take 2–3 hours, even though the underlying garbage rate has not dropped.
More from Models
- Community-built MiniMax H3 weights target 12GB–24GB GPUs with INT4 and NVFP4 variants — VoidAsuka · 2026-08-04
- Epoch AI’s MirrorCode benchmark sees Claude Fable solve C preprocessor and Pkl tasks — Jsevillamol · 2026-08-04
- X post jokes that Anthropic could hit $100B in revenue by year end — Jsevillamol · 2026-08-04
- DeepSeek chatter says the company is moving beyond a V4-Ultra-style model — teortaxesTex · 2026-08-04
- Yann LeCun says strong code generators are not just pure LLMs — suchenzang · 2026-08-04
- Windows reset restores RTX 3070 throughput for local Qwen3.6-35B inference — campaigner_ · 2026-08-04