Antithesis reports that its system can find a winning route through NetHack in under five minutes on nearly 80% of game seeds, with successful runs taking a median of less than two minutes. The goal is not merely to reach the dungeon’s depths: players must retrieve the Amulet of Yendor, escape, and complete the final journey across the Planes. NetHack’s permadeath, vast range of possible states, scarce information, and need for long-term planning have made it difficult for both human players and automated agents. Earlier bots relied on specialized exploits or extensive hand-coded rules, and AI victories were rare and dependent on favorable conditions.
The account traces Antithesis’s effort from early random-input fuzzing, which quickly stalled on unproductive inputs, infinite loops, and the problem of choosing among countless possible dungeon paths. Work on other games helped the team develop techniques for measuring progress, following a deliberate game plan, managing inventory, and discarding states likely to fail. The central claim is that Antithesis is more than a tool for exploring program states: it can act like a probabilistic programming system, where people or AI agents specify parts of a task and search fills in the rest. With capable AI agents directing that search, the system can produce rapid victories. NetHack also shaped Antithesis’s broader ideas about exploration, coverage, visualization, and the need to reward curiosity when solving problems that require sustained strategic progress.
Summary generated by AI from the linked article. hn.today is not affiliated with Hacker News or Y Combinator.
A Hacker News poster asks which practical AI skills individuals can learn online to benefit economically from LLMs and agents. They want specific, hands-on recommendations suited to someone with a moderately technical background, without requiring a degree. (news.ycombinator.com)
Former OpenAI researcher Will Depue claims hidden initialization patterns in six Chinese open-weight AI models match Nvidia Blackwell hardware, including systems restricted from export to China. He says the matches were reproduced on real chips and independently checked by researchers, though the claims are based on his forensic analysis. (twitter.com)
OpenCyclone is a free, open-source Android agent that uses your existing apps to carry out tasks through a hidden screen. It asks for approval before actions like sending messages and lets users connect their own AI models. (opencyclone.vercel.app)
Sam Bankman-Fried’s $500 million investment gave Alameda nearly 14% of Anthropic, a stake that would be worth about $37 billion at the AI company’s expected $1.5 trillion IPO valuation. The stake was sold in bankruptcy to repay FTX customers and creditors after Bankman-Fried’s businesses collapsed. (ft.com)
llama.cpp and ggml have reached v1.0, establishing stable public C APIs and clear versioning rules for breaking changes. The release strengthens the foundation for running local, multimodal AI across a wide range of hardware. (huggingface.co)
Gen Z is spending more on premium AI subscriptions to boost productivity, run businesses, and manage everyday tasks. Financial advisers say paid tools can be worthwhile when they deliver clear value, but users should check costs and prioritize essential savings. (reuters.com)
Today's best Hacker News stories, summarized and screenshotted, one email a day.