I'll change my benchmark when a model produces a genuinely good picture of a pelican riding a bicycle! These bicycles see still pretty obviously bad
@simonw
-
GLM-4.5 Models Released with MIT License by Zai
By
–
Pretty decent pelicans from the new GLM-4.5 and GLM-4.5 Air models. Both models are MIT licensed, released by @Zai_org this morning https://
simonwillison.net/2025/Jul/28/gl
m-45/
… -
Cheapest Mac Hardware for Running AI Models Locally
By
–
Thanks! Looks like cheapest 128GB Mac at the moment is a $3,500 Mac Studio, I'm interested in finding the lowest price for realistically running these models
-
Running Quantized Models on M2 Mac with 64GB RAM
By
–
What are my chances of running that (or a further quantized version) on an M2 with 64GB?
-
System Prompts: Workflow vs Agent Implementation Examples
By
–
Maybe what we need here is concrete examples of a system prompt that implements a workflow compared to a system prompt that implements an agent
-
Creating AI Workflow System Prompts and Tool Ordering
By
–
By "create the workflow" do you mean a system prompt which spells out which tools they should use in which order?
-
What does an AI agent architecture really look like?
By
–
I genuinely do not understand what an "agent" looks like that is not implemented as a system prompt and a bunch of carefully selected tools that can then be run by the LLM in a loop – what am I missing here?
-
Model-Driven vs System Prompts: Tool Planning Approaches
By
–
Is "model driven" different from a system prompt that says "plan which tools to use"?
-
System Prompts and Tools: Defining AI Agent Architecture
By
–
Is that definition materially different from a system prompt combined with a small collection of tools, assuming the system prompt says something similar to "plan how to use your tools to solve the user's problem and then keep on calling them until your goal has been reached"?
-
AI Models for Code Assistance Not Yet Mature Enough
By
–
AI-assisted programming was happening, but the models weren't yet strong enough to support true don't-even-look-at-the-code vibe coding