From 5 items, 3 important content pieces were selected

Technology News

  1. Math 2.0 needs holistic valuation of AI-era mathematical progress ⭐️ 8.0/10
  2. Trump administration suspends Microsoft from green card program ⭐️ 7.0/10
  3. Anthropic releases Claude Haiku 5.5 at GPT-6 Luna pricing ⭐️ 7.0/10

Technology News

Math 2.0 needs holistic valuation of AI-era mathematical progress ⭐️ 8.0/10

A Mathstodon post argues that mathematics in the AI era—dubbed “Math 2.0”—needs to value mathematical progress more holistically rather than judging it mainly by solved-problem benchmarks. The post sparked a large Hacker News discussion about large language models, proof verification, and research culture, with commenters debating whether AI is best at combining and organizing vast mathematical literature rather than producing fully understood results. Several commenters argue that dumping AI-generated proofs on the community shifts the burden of verification, refinement, and expansion onto mathematicians, and that authors who cannot explain or support their results are not contributing in the same way as conventional researchers. Others frame the value as the insights gained during proof discovery and the need to communicate what was solved and why it matters.

hackernews · ent101 · Oct 8, 05:14 · Discussion

「Background」 Terence Tao is an Australian-American mathematician and professor of mathematics at UCLA, widely regarded as one of the leading mathematicians of his generation, and he is the author of the Mathstodon post in question. The "Math 2.0" framing emerged amid a wave of AI-generated mathematical results — external coverage connects it to OpenAI's release of more than 300 mathematical results produced by an unreleased internal model — which has raised questions about who bears the burden of verifying proofs and how mathematical contribution is credited. Mathstodon, the Mastodon instance where the post appeared, is used by the mathematical community for exactly this kind of methodological debate.

「Impact」 Mathematicians and AI researchers may face a growing burden to verify and contextualize AI-generated proofs if evaluation continues to reward benchmark-style solved problems over explanation and field contribution.

「Community discussion」 Commenters broadly agree that AI is strong at connecting large amounts of mathematical information but disagree on whether current AI results represent genuine mathematical progress; skeptics warn that autonomous solvers and prompters often lack interest in or understanding of the broader field. A recurring concern is that proof verification, refinement, and expansion are being offloaded to the mathematics community, while several commenters argue that communication and insight—not just solving—are essential to the discipline.

References

Tags: #AI, #mathematics, #LLMs, #research evaluation, #scientific progress

Trump administration suspends Microsoft from green card program ⭐️ 7.0/10

The Trump administration is suspending Microsoft from a green card program, according to an Associated Press report. The action targets the employer-sponsored permanent-residency pathway that U.S. technology companies use to retain foreign workers, most commonly through the PERM labor-certification step tied to H-1B employment. It marks an escalation in immigration enforcement aimed directly at one of the largest corporate sponsors of skilled foreign labor. The supplied material does not specify the suspension's scope, duration, legal mechanism, or any response from Microsoft, so the practical consequences for affected employees remain unclear.

hackernews · alephnerd · Oct 8, 15:15 · Discussion

「Background」 The H-1B is a U.S. work visa that allows employers to hire foreign workers in specialty occupations, and employers can separately sponsor those workers for green cards through a labor-certification process meant to show that qualified U.S. workers are not being displaced. Vice President JD Vance said the Trump administration is suspending Microsoft from a program to apply for green cards for H-1B workers, accusing the company of fraud and abuse. The supplied reports do not specify the alleged conduct, how long the suspension lasts, or whether other employers are affected.

「Impact」 Suspending Microsoft from the Labor Department's Permanent Labor Certification program cuts off the first step toward employment-based green cards for its H-1B workers, leaving those employees' permanent-residency paths stalled for as long as the suspension lasts. External reports indicate the action also covers Adobe and other technology firms, so the effect may extend beyond Microsoft to a broader set of skilled foreign workers, though the scope and duration of the suspension remain unconfirmed in the supplied material.

「Community discussion」 Commenters largely questioned the fairness and consistency of the enforcement, citing H-1B and PERM sponsorship data to argue that Microsoft is being singled out while staffing firms such as Wipro and Tata Consultancy Services, or other large sponsors, face less scrutiny; one described the mechanism as a weapon that can be wielded selectively and called for a law addressing fraud across all abusers. A minority supported the move as correct because the program itself is the problem, and others debated whether green cards should be tied to an employer at all rather than to years worked or taxes paid. Several also referenced Vance's quoted claim that employers advertise in small-town newspapers, use the lack of response, and then replace American workers with what he called foreign indentured servants, with one commenter noting that such practices are widespread.

References

Tags: #immigration policy, #H-1B/PERM, #Microsoft, #tech labor market

Anthropic releases Claude Haiku 5.5 at GPT-6 Luna pricing ⭐️ 7.0/10

Anthropic released Claude Haiku 5.5, its new fast, low-cost model, replacing Haiku 4.5, which shipped almost a year earlier at $1 per million input tokens and $5 per million output tokens. Haiku 5.5 exactly matches OpenAI's GPT-6 Luna at $0.10/$0.50 per million tokens up to 100,000 tokens, but rises fivefold to $0.50/$2.50 beyond that threshold, while Luna's increase at 272,000 tokens goes only to $0.20/$0.75. Simon Willison notes a hidden cost increase: the new model uses a less generous tokenizer, and his Claude Token Counter shows the same long prompt consuming roughly 1.25x as many tokens as under Haiku 4.5. He reports that Haiku 5.5 matches Luna's price below 100,000 tokens with higher benchmark scores, but that Luna looks like a much better deal above that point; Haiku 5.5 cannot disable reasoning, defaults to medium effort, and produced a usable bicycle frame at every effort level beyond low. Anthropic also halved the price of cache reads for Sonnet 5.5 and introduced monthly API credits for Max and Team subscribers: $100 for Max 5x, $200 for Max 20x, and up to $500 pooled for Team, which do not roll over.

rss · Simon Willison · Oct 7, 20:56

「Background」 Claude Haiku is Anthropic's fast, low-cost tier of Claude models, positioned below the larger Sonnet and Opus lines. Its predecessor, Haiku 4.5, shipped in October 2025 at $1 per million input tokens and $5 per million output tokens, a price OpenAI's GPT-6 Luna undercut tenfold when it launched in September 2026. Two details shape how such pricing is compared: tokenizers determine how much text maps to each billable token, so a less efficient tokenizer raises effective cost even when the headline rate is unchanged, and Haiku 5.5 is the first Haiku model to expose a reasoning effort setting, spanning low through max, rather than allowing reasoning to be disabled.

「Impact」 Developers and organizations running workloads under 100,000 tokens gain a cheaper, higher-scoring alternative to GPT-6 Luna, while those exceeding that threshold face a fivefold price step-up plus an effective tokenizer-driven cost increase of roughly 1.25x.

References

Tags: #Anthropic, #Claude Haiku, #AI models, #LLM pricing, #model release