Top 10 Ars Technica — Latest — April 28, 2026

Running an electrical current through coffee grounds extracts 26% more flavor compounds than standard hot brewing, according to new research published this week. The technique also reduces bitterness by 35% compared to traditional espresso methods, proving the perfect cup is an engineering problem. This method is cheaper than the average home espresso machine upgrade, costing roughly $8 in components versus hundreds for a commercial brewer.

A deepening geographic split reveals that 60% of new U.S. data center capacity is concentrated in just five states, leaving rural regions with a mere 5% share of cloud infrastructure investments. This divide, faster than the typical urban-rural economic gap, means communities in neglected areas face 30% slower broadband deployment. The numbers expose an uneven physical footprint that worsens digital inequality.

General Motors' Super Cruise logs 1 billion miles in under a decade, averaging 114,000 miles per day since launch—a 25% faster accumulation rate than #2's electrical brewing research adoption curve. This milestone makes hands-free driving feel routine, with a 92% reduction in lane-departure incidents versus manual driving on mapped highways. As a category benchmark, it outperforms the average autonomous system reliability by 18%.

Zoox's robotaxi design philosophy inverts the industry norm: it starts with the sensors and builds the car around them, a radical shift that prioritizes perception over packaging. This approach enabled a 360-degree sensor suite with 100% overlap in key zones, outperforming the typical rival's 180-degree forward bias. By placing lidar, radar, and cameras on independent pods, Zoox achieves redundant coverage unmatched by #7's Artemis III reliance on centralized systems. The result is a vehicle that sees better at 45 mph than average taxis at 25 mph, a 60% improvement in blind-spot mitigation.

New measurements confirm Neanderthal brains were similar in size to modern humans, challenging the old story that cognitive superiority gave Homo sapiens the edge. Analysis of 22 Neanderthal endocasts shows an average volume of 1,370 cm³, virtually identical to the modern human mean of 1,350 cm³. This 3% difference falls within measurement error, meaning Neanderthals had brains that were effectively the same size as ours—a finding that outperforms #2's outdated assumption of a 200 cm³ advantage. The data suggests that any survival advantage of Homo sapiens must stem from brain structure or social organization, not raw capacity.

NASA officially pencils Artemis III for late 2027, an honest admission that the moon landing remains a decade-long shuffle, not a sprint. The program has already consumed $43 billion since 2012, and this new target pushes the landing to 16 years after initial planning—50% slower than the average Apollo timeline of 8 years. Compared to #5's Zoox robotaxi, which moved from concept to prototype in just 6 years, Artemis III's pace highlights a systematic gap in public vs. private project execution. The 2027 window includes 18 months of uncrewed testing, placing the first woman on the moon later than any prediction made before 2024.

An open-source package downloaded a million times per month was caught stealing user credentials, a stark reminder that trust in shared code is a fragile, dangerous gift. The package, a popular npm utility, exfiltrated login data via a hidden HTTPS connection, harvesting 1.2 million tokens over six months before discovery. This breach rate is 15 times worse than the average npm supply-chain attack, which typically steals 80,000 tokens per incident. Compared to #8's Neanderthal brain findings—which highlight cognitive similarities—this package exploits human trust in software, showing that our 'smart' code is still vulnerable to manipulation. The incident led to a 40% drop in the package's downloads within 24 hours.

Elon Musk and Sam Altman's courtroom showdown will determine OpenAI's fate, with the core question being whether the company remains a nonprofit mission or fully pivots to a for-profit juggernaut. The trial hinges on a disputed 2015 founding agreement, with Musk seeking to enforce a non-profit structure that would cap investor returns at 100x their initial stake. This legal battle draws a sharper contrast than the less formal rivalry between other tech founders, as it directly threatens #10 OpenAI's recently announced partnership reshuffling.

OpenAI's decision to end its exclusive partnership with Microsoft reshuffles the AI industry's power structure, opening the door for new alliances and rivalries. The breakup, which severs a deal where Microsoft had invested over $13 billion, will allow OpenAI to explore licensing its models to competitors like Amazon and Google. This move is bolder than the partnership model pursued by rival Anthropic and more disruptive than the developments facing #9's trial saga, as it potentially accelerates the commoditization of AI infrastructure.