10.17.26 | Tesla Roadster, Claude reveals ECDSA origins, 90% US tasks lack AI benchmarks

SPEAKER_1 (0:00)

Welcome to the Hacker News Highlights, where we explore the top 10 posts on Hacker News every day. Today, we dive into showcasing Tesla's Roadster, uncovering Ecdsa curve origins through timestamps, and revealing benchmark blind spots across American work. Let's jump right in.

Title, Tesla Roadster. Source, tesla.com. The article link returned an access denied message instead of information about the Tesla Roadster. The Hacker News discussion centered on a reported demonstration of a Roadster prototype using suction fans and cold gas thrusters to drive upside down, accelerate rapidly, and perform brief aerial stunts. The production versions were described as planned for 2028 with the flying version limited to invitations and not approved for public roads. In the comments, sentiment was mostly skeptical, although users found the demonstration impressive. Users debated whether the prototype had truly flown or had merely jumped and controlled its fall. Concerns focused on safety, high pressure gas tanks, debris from the suction fans, road legality, and Tesla's history of delayed or reduced product promises. Overall, the community viewed the Roadster as a spectacular marketing demo rather than a practical production car.

Title, Claude Reveals Ecdsa Curve Origins, Source Anthropic. The article explained how Claude traced the seeds for most of the NIST Ecdsa Curves to the Loren 48 pseudorandom number generator initialized with Unix timestamps from the 1990s. Claude also linked several older curvebase points to the free-lip random number generator. The findings made deliberate backdoors much less likely, but left the B163 seed and several base point origins unexplained. Anthropic said the work did not reveal a cryptographic attack and donated the $36,864 bounty to the Internet Security Research Group. In the comments, sentiment was largely positive, with users treating the discovery as a convincing end to decades of speculation. Users debated whether the timestamp-based process still left room for deliberate selection of weak curves, especially because viable candidates appeared to have been skipped. Supporters argued that the available search space was far too small for a practical hidden weakness, while skeptics said the unexplained choices still prevented complete certainty. Overall, users saw the result as strong evidence against a backdoor, not proof that every question had been answered.

Title, 90% of US working hours are spent on tasks not covered well by AI benchmarks. Source, mercor.com. The article mapped 29 AI benchmarks onto 18,797 tasks across 923 US occupations. It found that only 208 occupations had tasks covered by at least five benchmark prompts, suggesting that about 90% of working hours involved tasks with little or no benchmark coverage. The project also found that benchmarks favored AI-exposed tasks in high-wage occupations, especially software engineering, law and finance. Mercor said the results were provisional and that future benchmarks should cover more occupations, physical work, human-AI collaboration and real-world task quality. In the comments, the sentiment was mostly skeptical about what the figure proved, with users saying it measured benchmark gaps more than AI capability. Users debated whether AI would mainly replace high-paid knowledge workers or eventually handle physical jobs through robotics. Supporters of rapid automation argued that software and mathematics were easier to verify and could help improve robotics, while critics cited hardware costs, messy environments and weak economic incentives. Overall, commenters concluded that AI could raise productivity without delivering broad gains to workers or consumers.

Title. How the Claude code team does code review. Source. Anthropic. The article described how Anthropic reviewed code written largely by Claude. It shifted routine review to automated agents, tests, security checks, risk classifiers, feature flags, staged releases and fast rollbacks. Humans focused on product direction, architecture, high-risk changes, evidence from tests, and accountability. Anthropic said this process more than doubled merged pull requests from June to September, while reducing reverts by more than 40 percent. In the comments, this sentiment was mostly skeptical, with users questioning whether rapid code churn represented real progress. Users debated whether automated review could replace human understanding, especially when agents produced duplicated tangled or misleading code. Commenters also warned that revert rates missed long-term architecture problems, security risks, rising technical debt, and the loss of shared knowledge. Overall, the community saw Anthropic's process as an interesting experiment, but poor general advice without stronger human oversight.

Title I love software rendering. Source James F. mckenzie.com. The Post explained why the author preferred software rendered versions of classic DOS games such as Quake, Duke Nukem 3D, Blood, and Screamer. It praised their sharp nearest neighbor textures, limited color palettes, stepped lighting, and distinctive visual style. Modern processors could run these renderers at much higher resolutions than computers in the 1990s, although demanding games still needed powerful hardware. The author preferred native DOS on real hardware over DOS Box and noted that sound support remained a separate challenge. In the comments, sentiment was largely supportive of the retro aesthetic and the appeal of controlling the whole rendering pipeline. Users debated whether the author really loved software rendering or mainly loved nearest neighbor textures, low color depth, and period-specific lighting. Supporting Points noted that modern graphics processors could reproduce much of this style, but historical software renderers also used different algorithms and effects that were not always easy to match. Overall, users concluded that the look was real and appealing, even if software rendering was not its only possible source.