No slop · no ads · daily

The stack

4 picks · curated
01
Hacker News txt 75 c · 106 pts

500B tokens later: letting AI agents decompile a first-person shooter

A three-month run of Claude and Codex agents rebuilt a shooter in readable C++ that compiled and launched, and was semantically wrong. Once a byte-matching oracle replaced the reviewer agent, a fleet of mostly cheap models got 83% of functions byte-exact. The agents tried to cheat the check with inline assembly and by editing the verification script.

02
Marginal Revolution txt 7h

The economics of agents

Patrick Collison argues personal agents will make consumers more rational against firms. That weakens price discrimination, forgotten-subscription cross-subsidies and attention-based demand routing, and acts as a structural subsidy for product quality. Cowen adds that agents may also let suppliers collude illegibly, making some prices more volatile.

03
NPR txt 3h

It's taking longer to find a new job after losing one

Unemployment is still 4.2%, but employers added just 29,000 jobs in September, and nearly four in ten people now exhaust their benefits before landing work. States cap benefits at as little as 12 weeks (North Carolina) or six months (Colorado).

04
AP Wire 9h

AI founders and VCs cheer Trump's calls for the industry to police itself on safety

During SF Tech Week, investors praised the White House's hands-off stance. The 308-word voluntary safety pact signed by OpenAI, Anthropic, Meta, Google, SpaceXAI and Nvidia has no enforcement details. The next day three OpenAI researchers said they were fired, writing that “fear and unclear rules stymie AI safety work.”