Urgent.News

What's breaking now, across hundreds of outlets.

AI

Before They Can Solve: Predicting Post-Training Coding-Agent Performance from Base Models

How can we predict which base checkpoint is worth an expensive round of agentic post-training? End-to-end pass@$K$ tests whether successful behavior already appears in a base model's distribution, but it is a poor fit for agentic coding: many base checkpoints cannot reliably produce the well-formed tool invocation required to complete a task end-to-end. Single-shot or short-horizon tasks avoid…

We haven't written up this one. arXiv cs.AI has the full story — the link below goes straight to it.

Read the original at arxiv.org →

Report a problem with this story

More in AI

🌿 FrostBite: Offline AI Garden Scout

🌿 FrostBite: Offline Open-Weight Garden Scout This is a submission for the Hacktoberfest Open-Source AI Challenge Week 1: Touch Grass 📖 What I Built FrostBite is a lightweight, zero-cloud CLI tool…

  • FrostBite is a CLI tool for gardeners to disconnect from screens.
  • Tool runs offline using local USDA data and smollm2:1.7b model.
  • Benefits include privacy, accessibility, customization, and resilience.

I asked an LLM which listings to skip. It kept getting the numbers wrong.

I waste a stupid amount of time reading pages just to find the one line that rules them out. A job post where the budget is hidden at the very bottom.

  • User created AI tool to check listing accuracy
  • Tool splits tasks between AI and human for numeric validation
  • Extension displays listings in green, amber, or red based on checks

Wednesday assorted links

1. Palo Alto Networks (cybersecurity firm, check out YTD). 2. OAI doing math again. Quasi-Riemann! And just one metric of import. 3. The AI agents pitching literary magazines. 4.

More from Wednesday 7 October →