The relative failure of Llama 4 turned out to be very consequential to the AI landscape. It led to the shifting the locus of open weights development to China, a move towards closed models as companies running local Llama couldn’t continue to upgrade, & big talent wars in the US.
LLMS
-
LLM News Updates: Reverse-Engineering Latest Developments
By
–
… or if you want my free but MUCH longer and more frequent newsletter I just sent out out too – here's the latest edition, covering just the last three days of LLM-related news https://
simonw.substack.com/p/reverse-engi
neering-some-updates
… -
We Got 140,000 Tokens Instead of Flying Cars
By
–
We wanted flying cars, instead we got 140,000 tokens.
-
Residual Stream as Working Memory in Neural Networks
By
–
The residual stream is what psychologists call working memory.
-
Why Do AIs Have Version Numbers But Humans Don’t?
By
–
People don't have version numbers, so why do AIs?
-
Anthropic Reveals Persona Vectors for LLM Behavior Control
By
–
Anthropic literally showed how to control LLM personalities. They can inject Persona Vectors to make models adopt specific traits or prevent unwanted behaviors. Like a vaccine for AI – inject evil to prevent the model from becoming evil.
-
Weekly AI and Tech News: GLM-4.5, Runway Aleph, Luma Labs
By
–
Here’s all the AI and tech news from the past week. Did I miss anything? – Zhipu AI launched GLM-4.5
– @runwayml unveiled Runway Aleph
– @LumaLabsAI introduced “Modify with Instructions” to edit videos
– @ideogram_ai released new Character feature
– @midjourney added start & end -
Unofficial Claude Code Alternatives Integration via API
By
–
Yeah there should be – you could use one of the unofficial Claude Code alternative model mechanisms and plug it into their API https://t.co/g3T8bFr8IA
— Simon Willison (@simonw) 1 août 2025Yeah there should be – you could use one of the unofficial Claude Code alternative model mechanisms and plug it into their API
-
Kimi K2 Achieves 5x Speed Improvement Without Quality Loss
By
–
And in further LLM performance news, Kimi K2 now have a model that's 5x faster for the same quality of results
-
Qwen3-Coder-480B: Top Open-Source Coding Model
By
–
Qwen3-Coder-480B-A35B-Instruct is a top open-source model for coding. It delivers precise, context-aware programming responses and supports a 256k-token context window. (
https://
poe.com/Qwen3-Coder-48
0B-FW
…) (5/8)