5 corporations control 90% of the media.

They decide what you see. What you hear. What you think. Your opinions? Manufactured. Question everything #opus #moscow grok

X AI Models 𓃵Aug 16

Qwen3.8-27B: 27B dense, Apache 2.0, 65 tok/s on a 4090.

The "beats Opus 4.6 on every benchmark" take is wrong. Their own table: wins SWE-bench Pro, loses Terminal-Bench, GPQA, HLE. Hands-on reviewers put it between Opus 4.6 and Sonnet 5, which is still wild for 27B.

Josh PocockAug 16

Ran the "Lord of the Rings intro" test with both DeepSeek-V4-Flash 0731 (cloud version) and Qwen3.8-27B (local on my DGX Spark)

DS4 was quicker (40 min vs. 2hr 17 min) and hit some of the elements better (party tent), but the Qwen model made a much visually richer world and did

tonbiAug 1615

An interactive 3D simulation of a coronal mass ejection striking Earth, compressing the magnetosphere

Intensifying aurora and stretching the magnetotail toward reconnection, with adjustable CME intensity. Code -> Opus 5 using Three.js

TechartistAug 1641