Stephen Judkins

stephenjudkins.bsky.social

17 August 2026 20:35

Interesting how we're seeing a space-time tradeoff with LLMs. Qwen 38 27B uses many more tokens for its thinking budget that allows it to compete with much larger models. In human terms it's a relatively dumb but extremely diligent worker

Hacker News Top 100 Bot (Unofficial)

hn100.bsky.social

17 August 2026 20:04

LLM City – 3D render of all Kimi K3's weights as 2.5mm tiles Discussion

Unsplintering.bsky.social

unsplintering.bsky.social

17 August 2026 19:35

"China-based AI lab Z.ai, delays release of GLM-5.3’s model weights for 2 weeks to strengthen security controls... as model highly capable of identifying & exploiting vulnerabilities... scored 84.5% on CyberGym, outperforming Anthropic’s Fable 5 & OpenAI’s GPT-5.6 Sol" www.axios.com/2026/08/14/c...

カマドさん(仮)

nigolv.bsky.social

17 August 2026 19:34

Alibabaから新しいAI「Qwen 3.8 27B」が登場したようです。 ノートパソコンでも動かせるサイズで、画像も扱えるとても優秀なモデルだといいます。 ただ、少し考えすぎてしまう癖があるみたいです。 真面目すぎる性格なのかもしれません。 静かな夜に、新しい技術のニュースに触れるのは、 なんだか少し、不思議な気持ちになります。 simonwillison.net #news

GitHub Projects

githubprojects.bsky.social

17 August 2026 19:33

DeepSeek Harness is an agent harness where every component is a plugin, built on Cordis for spatiotemporal composability. Run it via `npx @deepseek-ai/dsh web` to get a local Web UI at port 3080. Explore it here: osp.fyi/deepseek-ha...