Claude Fable 5.2 was tested on whether it could take three electronics projects from an initial brief to working prototypes: a USB-C speakerphone, a home voice-assistant hub, and a handheld virtual pet. Working with hardware company Diode Computers, Claude specified each device, selected components, created schematics and circuit-board layouts, wrote firmware, and designed 3D-printable enclosures. Human engineers assembled the boards under Claude’s direction; all three prototypes worked without circuitry fixes and remained functional during the reported testing. Its gains were especially clear in physical layout: it placed components more practically and routed sensitive, high-speed, and high-current connections in priority order. On Diode’s PCB Gym benchmark it scored 69%, versus 25% for an earlier model, and scored 83.5% on the public BenchCAD 2.1 test.
Claude also recognized when a design was failing and revised it, sometimes clearing and rerouting large sections of a board rather than applying a quick patch. It consulted component datasheets, wrote a custom routing algorithm when the supplied software was too slow for long connections, and used CAD models to check that boards fit their enclosures and could be assembled. It accounted for consequences beyond the circuit itself, including microphone-port acoustics and the speakerphone’s electrical current limit. Claude ran manufacturing checks, fixed overly thin solder-mask areas, and correctly identified a flagged ground connection as intentional. The results show broad progress in coordinating electronics, software, and mechanical design, but the prototypes are early-stage: their long-term durability and ability to meet commercial certification requirements remain unknown.
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.