Anthropic appears to be A/B testing reduced effort levels in Claude Code
SummaryClaude Code effort-level A/B testing signals incoming feature refinement—timely for people using Claude for coding work.
52 → 216 points · peak 216 · 33 observations
Selected items · August 23, 2026
What moved on August 23, 2026
The items the system selected that day and what they measured. This day came before the written daily briefs; it shows the data.
Why your local LLM feels dumber than it is
SummaryPractical explainer on why local LLMs underperform and how to get better results—immediately useful for people running AI locally.
2 → 507 points · peak 507 · 38 observations
Posted on X · 2026-08-23Local LLMs don't feel dumber because the weights are worse. They feel dumber because the defaults are set for your laptop's RAM, not the model's actual capability.
Most of the gap is context window, temperature, and sampling settings you can fix in 30 seconds.