Marcel Bucher, a professor of plant sciences at the University of Cologne in Germany, lost two years of carefully structured academic work in an instant when he temporarily disabled ChatGPT's "data consent" option in August to test whether the AI tool's functions would still work without providing OpenAI his data. All his chats were permanently deleted and his project folders emptied without any warning or undo option, he wrote in a post on Nature. Bucher, a ChatGPT Plus subscriber paying $20 per month, had used the platform daily to draft grant applications, prepare teaching materials, revise publication drafts and create exams. He contacted OpenAI support, first receiving responses from an AI agent before a human employee confirmed the data was permanently lost and unrecoverable. OpenAI cited "privacy by design" as the reason, telling Nature it does provide a confirmation prompt before users permanently delete a chat but maintains no backups. Bucher said he had saved partial copies of some materials, but the underlying prompts, iterations, and project folders -- what he describes as the intellectual scaffolding behind his finished work -- are gone forever.

Read more of this story at Slashdot.

Anthropic has a problem that most companies would envy: its AI model keeps getting so good, the company wrote in a blog post, that it passes the company's own hiring test for performance engineers. The test, designed in late 2023 by optimization lead Tristan Hume, asks candidates to speed up code running on a simulated computer chip. Over 1,000 people have taken it, and dozens now work at Anthropic. But Claude Opus 4 outperformed most human applicants. Hume redesigned the test, making it harder. Then Claude Opus 4.5 matched even the best human scores within the two-hour time limit. For his third attempt, Hume abandoned realistic problems entirely and switched to abstract puzzles using a strange, minimal programming language -- something weird enough that Claude struggles with it. Anthropic is now releasing the original test as an open challenge. Beat Claude's best score and ... they want to hear from you.

Read more of this story at Slashdot.

Selon le média américain The Verge, des milliers d’employés de Microsoft ont récemment été incités à utiliser Claude Code pour améliorer leur productivité au travail… alors même que Microsoft commercialise GitHub Copilot auprès de ses clients.​

Ancient Slashdot reader erice shares the findings from a recent study showing that while AI helped researchers publish more often and boosted their careers, the resulting papers were, on average, less useful. "You have this conflict between individual incentives and science as a whole," says James Evans, a sociologist at the University of Chicago who led the study. From a recent IEEE Spectrum article: To quantify the effect, Evans and collaborators from the Beijing National Research Center for Information Science and Technology trained a natural language processing model to identify AI-augmented research across six natural science disciplines. Their dataset included 41.3 million English-language papers published between 1980 and 2025 in biology, chemistry, physics, medicine, materials science, and geology. They excluded fields such as computer science and mathematics that focus on developing AI methods themselves. The researchers traced the careers of individual scientists, examined how their papers accumulated attention, and zoomed out to consider how entire fields clustered or dispersed intellectually over time. They compared roughly 311,000 papers that incorporated AI in some way -- through the use of neural networks or large language models, for example -- with millions of others that did not. The results revealed a striking trade-off. Scientists who adopt AI gain productivity and visibility: On average, they publish three times as many papers, receive nearly five times as many citations, and become team leaders a year or two earlier than those who do not. But when those papers are mapped in a high-dimensional "knowledge space," AI-heavy research occupies a smaller intellectual footprint, clusters more tightly around popular, data-rich problems, and generates weaker networks of follow-on engagement between studies. The pattern held across decades of AI development, spanning early machine learning, the rise of deep learning, and the current wave of generative AI. "If anything," Evans notes, "it's intensifying." [...] Aside from recent publishing distortions, Evans's analysis suggests that AI is largely automating the most tractable parts of science rather than expanding its frontiers.

Read more of this story at Slashdot.

An anonymous reader quotes a report from the Korea Herald: South Korea will begin enforcing its Artificial Intelligence Act on Thursday, becoming the first country to formally establish safety requirements for high-performance, or so-called frontier, AI systems -- a move that sets the country apart in the global regulatory landscape. According to the Ministry of Science and ICT, the new law is designed primarily to foster growth in the domestic AI sector, while also introducing baseline safeguards to address potential risks posed by increasingly powerful AI technologies. Officials described the inclusion of legal safety obligations for frontier AI as a world-first legislative step. The act lays the groundwork for a national-level AI policy framework. It establishes a central decision-making body -- the Presidential Council on National Artificial Intelligence Strategy -- and creates a legal foundation for an AI Safety Institute that will oversee safety and trust-related assessments. The law also outlines wide-ranging support measures, including research and development, data infrastructure, talent training, startup assistance, and help with overseas expansion. To reduce the initial burden on businesses, the government plans to implement a grace period of at least one year. During this time, it will not carry out fact-finding investigations or impose administrative sanctions. Instead, the focus will be on consultations and education. A dedicated AI Act support desk will help companies determine whether their systems fall within the law's scope and how to respond accordingly. Officials noted that the grace period may be extended depending on how international standards and market conditions evolve. The law applies to three areas only: high-impact AI, safety obligations for high-performance AI and transparency requirements for generative AI. Enforcement under the Korean law is intentionally light. It does not impose criminal penalties. Instead, it prioritizes corrective orders for noncompliance, with fines -- capped at 30 million won ($20,300) -- issued only if those orders are ignored. This, the government says, reflects a compliance-oriented approach rather than a punitive one. Transparency obligations for generative AI largely align with those in the EU, but Korea applies them more narrowly. Content that could be mistaken for real, such as deepfake images, video or audio, must clearly disclose its AI-generated origin. For other types of AI-generated content, invisible labeling via metadata is allowed. Personal or noncommercial use of generative AI is excluded from regulation. "This is not about boasting that we are the first in the world," said Kim Kyeong-man, deputy minister of the office of artificial intelligence policy at the ICT ministry. "We're approaching this from the most basic level of global consensus." Korea's approach differs from the EU by defining "high-performance AI" using technical thresholds like cumulative training compute, rather than regulating based on how AI is used. As a result, Korea believes no current models meet the bar for regulation, while the EU is phasing in broader, use-based AI rules over several years.

Read more of this story at Slashdot.

Intel says it struggled to satisfy demand for its AI data-center CPUs while new PC chips squeeze margins. CEO Lip-Bu Tan framed the turnaround as supply-constrained, not demand-constrained, with manufacturing yields (18A) improving but still below targets. Reuters reports: The forecast underscores the difficulties faced by Intel in predicting global chip markets, where the company's current products are the result of decisions made years ago. The company, whose shares have risen 40% in the past month, recently launched a long-awaited laptop chip designed to reclaim its lead in personal computers just as a memory chip crunch is expected to depress sales across that industry. Meanwhile, Intel executives said the company was caught off guard by surging demand for server central processors that accompany AI chips. Despite running its factories at capacity, Intel cannot keep up with demand for the chips, leaving profitable data center sales on the table while the new PC chip squeezes its margins. "In the short term, I'm disappointed that we are not able "to fully meet the demand in our markets," Chief Executive Officer Lip-Bu Tan told analysts on a conference call. The company forecast current-quarter revenue between $11.7 billion and $12.7 billion, compared with analysts' average estimate of $12.51 billion, according to data compiled by LSEG. It expects adjusted earnings per share to break even in the first quarter, compared with expectations of adjusted earnings of 5 cents per share.

Read more of this story at Slashdot.