Cmpnd's new dspy-wsp prototype lets a large language model tweak model weights without gradients, costing $5‑$10 per test.
NoteBy OossaPublished by Oossa: 1 min read
Cmpnd released dspy-wsp, a prototype optimizer that adds weight‑space programming to the DSPy framework. A frontier LLM – a large language model – reads a small model’s weights, edits them in a sandbox, and evaluates each change without using gradients. Each test run costs roughly $5 to $10. The tool is currently a prototype and not yet a commercial product.
Developers can experiment with model weight edits cheaply, opening new ways to fine‑tune AI without traditional training.
Models
Gemini 4 Carbon beats GPT‑6.1 Sol on the MathArena test but costs about 15 times more per query.
Models
The model turns camera images and instructions into robot movements. Runway says it led a simulation benchmark, but its tests have not been independently checked.
Models
OpenAI's new ChatGPT Wallet lets its ‘dots’ agents purchase items for U.S. Pro subscribers using tokenized card data.
Oossa · Newsletter
The week in AI, explained
Every Monday: the stories worth knowing, in plain language. Free, no spam.
