That's a good direction! Let models write their tool calls in code snippets, it's much more flexible & readable & expressive than the lame JSON blobs of usual tool calling!
LLMS
-
OpenAI plans 125x energy capacity increase by 2033
By
–
OpenAI have 9x'd their capacity this year, and have a goal to 125x further by 2033, thus exceeding current energy capacity of the whole of India – as per @alexeheath reporting of @sama
's internal slack note. One thing to note is that 125x energy capacity increase understates the -
Realtime Voice AI Still Too Limited for Meaningful Debate
By
–
Disappointed by the lack of progress on the realtime voice AI front. They'd be so good for debating ideas if they weren't as dumb as rocks.
-
AI Progress, Evals, and Reinforcement Learning in Job Market
By
–
quite a privilege to be a guest on Odd Lots! talked w joe & tracy about pace of progress in AI, broken evals, the crazy job market, and the difference between supervised and reinforcement learning
-
OpenAI reasoning system answer selection without test-time harness
By
–
OpenAI say "the reasoning system selected which answers to submit with no bespoke test-time harness whatsoever" – but were the answers themselves developed with the assistance of tools?
-
Gemini Achieves Gold in Programming Contest: Tool Usage Clarification Needed
By
–
The Gemini post mentions tools in the context of training the model itself, but doesn't clarify tools were used by the model at inference time during the competition https://
deepmind.google/discover/blog/
gemini-achieves-gold-level-performance-at-the-international-collegiate-programming-contest-world-finals/
… -
OpenAI Google DeepMind ICPC models code execution tools clarification
By
–
Anyone know if there's a solid answer from OpenAI or Google DeepMind yet on whether the models they used in their ICPC programming competition wins had access to tools like the ability to run generated code through a compiler?
-
Meta Explores Google AI Partnership for Enhanced Ad Targeting
By
–
Meta Eyes Google AI for Ads: Google's valuation is going to skyrocket Meta is exploring talks with Google about using Gemini and Gemma AI models to boost its ad targeting, after setbacks with its own AI efforts. While still early, the move shows Meta’s willingness to lean on
-
Vibe Coding: Building Complete Apps with Single Prompts
By
–
Vibe coding is evolving fast. With a single prompt, you can:
— Abacus.AI (@abacusai) 26 septembre 2025
•Design a frontend
•Set up user login & permissions
•Build an editor
•Spin up and connect a database
You can’t replace real developers, but you can build a simple, usable app for collaboration, tracking, and… pic.twitter.com/asQcWETD3iVibe coding is evolving fast. With a single prompt, you can:
•Design a frontend
•Set up user login & permissions
•Build an editor
•Spin up and connect a database You can’t replace real developers, but you can build a simple, usable app for collaboration, tracking, and -
Expert-Driven AI Implementation Improves Cost and Speed
By
–
I also think the paper makes a very compelling case that experts working directly with AI results in cost and speed improvements, and that those experts should use more advanced models. You should probably be doing this in firms It also shows this was not always true with GPT-4