A nephew’s hand-drawn plan for a family-themed beat ’em up began as a modest birthday project: five selectable heroes, twelve villains, a stolen vehicle and a puppy to rescue. The first version, assembled with coding tools and a game template, was playable but too shallow to match the child’s vision. With an overnight deadline and access to Claude Fable 5.2, the developer asked the system to rebuild the game as a polished arcade-style brawler, using the drawings as references and generating art, sound, levels and support for four controllers. By morning, it had produced a complete twelve-stage game with cutscenes, breakable crates, power-ups and an ending. It also created a bot to play through every stage in a browser and inspect screenshots for defects.
The family’s playtesting then drove further changes: Owen gained a Hulk transformation, Aiden received an exceptionally powerful attack, and relatives and two pet turtles became unlockable fighters, bringing the roster to eighteen. Each addition prompted new character art and story updates; the final rescue honors a recently deceased great-grandmother as an angel. The project used roughly 600 Claude requests over two days, about 500 generated images, eleven music tracks, around 100 sound effects and 90 announcer lines - roughly 340 million tokens in total. The developer presents it as evidence that modern coding agents can produce a complete, genuinely enjoyable game in a single intensive run, while acknowledging uncertainty about whether earlier models could have achieved the same result.
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.