Top Stories

Apple sues OpenAI, accuses ex-employees of stealing trade secrets

1125 points · 9to5mac.com

The gloves are off between two of tech’s heaviest hitters. Apple has filed suit against OpenAI, alleging that former Apple employees who jumped ship carried confidential trade secrets with them. It’s the kind of talent-war escalation the HN crowd has been predicting ever since the two companies moved from partners to rivals, and it lands squarely in the middle of the broader poaching frenzy over AI researchers.

The thread is dissecting everything from the specifics of California’s (limited) trade-secret enforcement to what this signals about Apple’s own on-device AI ambitions. When a company this litigation-averse goes to court, people pay attention.

GPT-5.6 Sol Ultra produces proof of the Cycle Double Cover Conjecture

465 points · openai.com

OpenAI’s latest frontier model reportedly generated a proof of the Cycle Double Cover Conjecture, a decades-old open problem in graph theory. If it holds up under peer scrutiny, it’s a genuine milestone for AI-assisted mathematics — and the “if it holds up” is exactly what commenters are chewing on.

Mathematicians in the thread are split between excitement and healthy skepticism, poring over the PDF for the kind of subtle gaps that have sunk machine-generated proofs before. Either way, the bar for what these systems can attempt keeps moving.

Show HN: Getting GLM 5.2 running on my slow computer

859 points · github.com

A developer shared Colibri, their project for coaxing the hefty GLM 5.2 model into running on modest, non-GPU hardware. This is catnip for HN: the ongoing quest to run capable open-weight LLMs locally without a data-center budget resonates with anyone who’s watched VRAM requirements balloon.

The comments are a goldmine of quantization tricks, offloading strategies, and real-world tokens-per-second numbers on aging laptops. It’s a reminder that a big chunk of the AI community cares less about the frontier and more about squeezing usefulness out of the hardware they already own.

NYC to ban deceptive subscription practices

534 points · theguardian.com

New York City is moving to outlaw the dark patterns that make subscriptions trivial to start and maddening to cancel. Think buried cancel buttons, endless retention flows, and “are you sure?” gauntlets. For a community that has ranted about these tactics for years, the mood is celebratory.

Discussion ranges from whether local rules can meaningfully bind national services to how this dovetails with the FTC’s broader “click to cancel” push. Several commenters are already imagining the compliance scramble ahead for subscription-heavy businesses.

QuadRF can spot drones and see WiFi through my wall

601 points · jeffgeerling.com

Jeff Geerling is back with a hardware deep-dive, this time on QuadRF, a software-defined-radio setup capable of detecting drones and mapping WiFi signals through walls. It’s the sort of accessible RF hacking that turns invisible spectrum into something you can actually visualize.

The thread blends genuine enthusiasm for the tech with sober notes about privacy and surveillance implications. As RF sensing gets cheaper and more capable, HN is wrestling with what it means when anyone can see what’s radiating around them.

Good Tools Are Invisible

448 points · gingerbill.org

Ginger Bill (of Odin language fame) argues that the best tools disappear into the work — they don’t demand attention, ceremony, or constant configuration. The essay is a quiet rebuke of over-engineered developer tooling that ends up being the project rather than serving it.

It struck a nerve. Commenters are trading their own examples of tools that got out of the way versus those that became a second job to maintain, with the usual detours into editors, build systems, and the eternal simplicity-versus-power debate.

AI 2040: Plan A

287 points · ai-2040.com

A speculative long-form scenario mapping out how AI could plausibly unfold over the next decade and a half, framed as a concrete “Plan A” rather than a vague forecast. It’s ambitious enough to draw both engaged readers and pointed critics.

The thread is a microcosm of the AI-discourse spectrum: some find the exercise clarifying, others see confident extrapolation dressed up as planning. The debate over how seriously to take detailed AI timelines is very much alive here.

After 7 years in production, Scarf has reluctantly moved away from Haskell

148 points · avi.press

A candid engineering retrospective on why Scarf, after seven years running Haskell in production, decided to migrate away. The “reluctantly” in the title sets the tone — this is admiration mixed with hard-won pragmatism about hiring, tooling, and operational friction.

Functional-programming devotees and pragmatists are having exactly the debate you’d expect, but the post’s honesty about the tradeoffs keeps it from becoming a flame war. It’s a useful data point for anyone weighing a niche-but-powerful language for long-term production work.