Anthropic Expands Post-Training Entities, But ICL Capabilities Lag
Sauers_ · x · 2026-07-29
Observations indicate that Anthropic has expanded post-training to include various entities beyond the 'user' (e.g., graders, malicious users, automated systems, simulated users, and other Claude instances). However, Claude does not yet adequately utilize in-context learning (ICL) to identify which type of entity it is interacting with. Scaling these entities seems to have incrementally improved ICL abilities with each recent model release.
Related event: Claude Expands Multi-Entity Interactions, But Context Flaws Remain(2 posts)→
More from Models
- Perplexity adds Kimi K3 for Pro and Max users, hosted on U.S. servers — perplexity_ai · 2026-07-29
- Claude Opus 5 hits 74% on DeepSWE, topping long-horizon coding models — brandon_galang · 2026-07-29
- PostTrainBench v1.1 flags 234 contaminated runs and tightens anti-cheat rules — scaling01 · 2026-07-29
- Nvidia’s LatentMoE is already shaping MoE pretraining after a paper from six months ago — peterjliu · 2026-07-29
- Internal chart compares how many tokens models need to center a div — BLUECOW009 · 2026-07-29
- Microsoft Unveils Mage-VL: Codec-Native Streaming VLM with 3.5x Inference Speedup — pmttyji · 2026-07-29