AxiomCode Graph is a developer tool that turns a repository into a queryable map of code relationships for programmers and AI agents. Its central claim is that accurate change analysis requires resolving calls as a compiler does, using declared and inferred types, rather than guessing from syntax or matching names. That helps expose calls through interfaces, overrides, generics, and callbacks, so agents can trace a change’s downstream effects and identify relevant tests instead of relying on incomplete search results. The graph records source locations and confidence tiers, and leaves unresolved calls explicit rather than treating uncertain links as facts. The project says this focused context can reduce exploration and tool use while helping agents make more complete changes.
The README reports tests on 748 held-out Defects4J bugs: AxiomCode selected every bug-revealing test for 94.9% of bugs, compared with 61.5% for the best listed tree-sitter-based system, and achieved an F1 score of 72.4 against Defects4J’s observed test selection. Against compiler-derived call targets, it linked 96.5% of Java calls and 88.8% of TypeScript calls to the exact target. Cross-file relationship F1 was 0.976 for Java and 0.896 for TypeScript. The engine supports Java, TypeScript, and Python, with JavaScript and C# in beta; an agent plugin provides MCP tools, a skill, and hooks for several coding environments. Installation requires Node 22.13 or later and Python 3.
Summary generated by AI from the linked article. hn.today is not affiliated with Hacker News or Y Combinator.
Researchers tested 16 LLMs on all 55 TIS-100 puzzles, giving each one attempt and time to refine its code. The models found 133 new performance tradeoffs across 18 puzzles and set three outright cycle records, though human players improved on most solutions within days. (tis-100.wincon.dev)
Australia is considering a new criminal offence for AI companies that fail to take reasonable steps to prevent harm caused by their models. The government is also weighing an AI safety watchdog with power to impose significant fines, following a hack of a Medicare statistics portal involving OpenAI. (abc.net.au)
AI is increasingly becoming a gateway through which people find information. Audiences still resist having AI produce their news. (niemanlab.org)
Nvidia holds a central position in the AI industry, making disruption to the company potentially catastrophic for the wider ecosystem. The race for advanced AI is also intensifying economic and geopolitical conflict, while concentrated corporate power and infrastructure face growing political resistance. (france24.com)
LLM inference can require reading about 26 GB of model weights to generate each token from a 13-billion-parameter model. The explanation shows how model size drives memory bandwidth demands during text generation. (youtube.com)
Deletion Tuning matches full fine-tuning across three LLM post-training methods by zeroing fewer than 0.05% of model weights. Its compact masks also preserve prior-task accuracy better than full fine-tuning, LoRA, and random sparse tuning. (arxiv.org)
Today's best Hacker News stories, summarized and screenshotted, one email a day.