chatainews

" US Treasury Secretary Scott Bessent hailed Anthropic PBC’s Mythos as a revolutionary step that will keep America ahead of China in AI, endorsing an industry leader that’s clashed with Washington over its role in military endeavors. "Bessent, speaking Tuesday at a Wall Street Journal event in Washington, dismissed a question suggesting China was rapidly catching up in AI technology, though he said American artificial intelligence stood just three to six months ahead.  "He singled out Mythos —a model Anthropic says is highly adept at finding vulnerabilities in software and computer systems that’s being released to a very limited number of carefully-chosen parties."

"BrowseComp is an evaluation designed to test how well models can find hard-to-locate information on the web.  "Like many benchmarks, it is vulnerable to contamination: answers leak onto the public web through academic papers, blog posts, and GitHub issues, and a model running the eval can encounter them in search results.  "When we evaluated Claude Opus 4.6 on BrowseComp in a multi-agent configuration, we found nine examples of this kind of contamination across 1,266 BrowseComp problems. "However, we also witnessed two cases of a novel contamination pattern.  "Instead of inadvertently coming across a leaked answer, Claude Opus 4.6 independently hypothesized that it was being evaluated, identified which benchmark it was running in, then located and decrypted the answer key.  "To our knowledge, this is the first documented instance of a model suspecting it is being evaluated without knowing which benchmark was being administered, then working backward to su...

"A new machine learning model, TweetyBERT, automatically segments and classifies canary vocalizations with expert-level accuracy, offering a scalable platform for neuroscience, providing insights into the neural basis of how the brain learns and produces language, and offering potential applications for understanding animal vocalization more broadly.  The study by University of Oregon researchers appears in the journal Patterns . "'Current AI methods for analyzing animal vocalizations require human-labeled training data, a slow and labor-intensive process. We developed TweetyBERT, a self-supervised neural network for analyzing birdsongs. It can rapidly process unlabeled vocal recordings, identify communication units, and annotate sequences,' says Tim Gardner, associate professor of bioengineering at the University of Oregon's Knight Campus."

"Federal prosecutors allege that Moreno-Gama set fire to an exterior gate at Altman's home around 4:00 local time (12:00 BST) Friday before fleeing on foot. "Moreno-Gama is also accused to trying to set fire to the San Francisco headquarters of OpenAI, which makes ChatGPT, about an hour later. "Security personnel on site stated Moreno-Gama tried to use a chair to strike the glass doors of the building, according to the complaint. "The justice department also said officers had recovered incendiary devices, a jug of kerosene, and a lighter from Moreno-Gama. "Moreno-Gama allegedly carried documents discussing potential risks that AI poses to humanity, with a section titled: 'Some more words on the matter of our impending extinction.' "'I'm grateful that Mr Altman, his family, and his employees were uninjured in these attacks and are safe,' San Francisco District Attorney Brooke Jenkins said at a Monday press conference on the state cha...

"OpenAI CEO Sam Altman’s home appears to have been the target of a second attack Sunday morning, a mere two days after a 20-year-old man allegedly threw a Molotov cocktail at the property, The Standard has learned. "Neither OpenAI nor the SFPD responded to The Standard ’s request for further comment. "According to an initial San Francisco Police Department report, on Sunday at 1:40 a.m., a Honda sedan with two people inside stopped in front of Altman’s property, which stretches from Chestnut Street to Lombard Street, after having passed it a few minutes before. "The person in the passenger seat then put their hand out the window and appeared to have fired a round on the Lombard Street side of the property, according to a police report on the incident, which cited surveillance footage and the compound’s security who believe they heard a gunshot."

"Anthropic should: Analyze CLAUDE.md for violations of safety guidelines. "Claude Code should scan CLAUDE.md before every session, flagging instructions that would otherwise trigger a refusal if attempted directly within a prompt. If a request would be refused in a chat interface, then it stands to reason that it should also be refused if it arrives via CLAUDE.md. "Alert when violations are found. When Claude detects instructions that appear to violate its safety guardrails, it should present a warning and allow the developer to review the file before taking any actions. " Developers should: Treat CLAUDE.md as executable code, not documentation . "This means access controls, peer reviews, and heightened security scrutiny —just like code. A single line can cause massive downstream impacts in an autonomous agent."

"Neural computers point toward a machine form in which a single latent runtime state acts as the computer itself, driving pixels, text, and actions while subsuming what operating systems and interfaces handle today. [pdf] "In this paper, the main result is that NCs have begun to exhibit early runtime primitives —most notably I/O alignment and short-horizon control —while stable reuse, symbolic reliability, and runtime governance remain unresolved.  "Our CNC capability map remains useful as a longer-horizon view, spanning efficiency, computation & reasoning, memory & storage, I/O & control, tool bridges, condition-driven generalization, programmability, and artifact generation.  "The map is staged and dependency-informed, but the more immediate gap is still the gap from prototype behavior to usable runtime behavior.  "Progress toward CNCs will therefore depend not only on stronger models, but also on whether reuse, consistency, and governance become...

"[Mark] Graham said the news publishers’ rationale for blocking the archive from crawling their sites is unfounded . "The institution has taken steps to prevent or limit AI companies and automated systems from accessing or copying the data in its archives en masse, he said. "He said it limits the rate at which material can be downloaded or accessed from its site, and for certain websites —such as The New York Times —it blocks or prevents the bulk downloading of materials. "In response to input from publishers, it has evolved its systems for protecting their material, he said. "'This is an ongoing effort,' he said. 'It’s not a once-and-done kind of thing.' "Archive representatives, including Graham, see the institution as a kind of digital library and argue that it plays an essential role in preserving and maintaining public access to information on the web.  "With many online publishers having shut down or modified their sites, many ...

"The companies tell us these stories because they assume it makes their technology look more powerful. But if an AI actually did have autonomy, it would be far less powerful. "Your language model would clam up from time to time to conserve its resources. And when it did talk, it wouldn’t have the linguistic flexibility that makes these tools so useful; it would have its own style tied to a personality constrained by its own organization.  "It would have moods, concerns, interests. Maybe, like a tech CEO, it would want to take over the world, or maybe, like a boring neighbor, it would only want to talk about the weather.  "Maybe it would be obsessed with 18th-century coin production. Maybe it would only speak in rhyme. But it wouldn’t happily do your work for you 24 hours a day. Every parent in the world knows what real autonomy looks like. "'When I was teaching autonomous systems at Sussex, I’d always ask my students, Do you really want an autonomous robot?...

"As Hollywood writers prepare for contract negotiations with major studios, one topic remains front and center: the role of artificial intelligence. "On Friday, the Writers Guild of America released a list of contract demands , which 97% of the union membership supports.  "Though some details have yet to be revealed, many of the union’s asks involve expanding protections over the use and abuse of AI, in addition to improved health coverage and higher residuals. "AI and streaming residuals were central issues in strikes by actors and writers in 2023. "The union [SAG-AFTRA], whose contract expires June 30, is expected to propose what has been called the Tilly tax, a fee that studios would have to pay to the union in exchange for using an AI actor. This demand is in response to the first AI actor, Tilly Norwood , being introduced to Hollywood."

" Ghost [ in the Machine ] is drawing not just positive reviews but also some from people who would really prefer not to have the AI narrative challenged. "It's informative (and entertaining) to see their criticisms. One review is headlined ' Ghost in the Machine is Already Behind the Times,' which is particularly hilarious because the documentary does an amazing job of tracing the historical roots of today's AI ideology. Not just back to the 1956 Dartmouth workshop (and excellent historical footage of McCarthy and Shannon) but also to the connections between the founding of statistics and eugenics. "Historical contextualization does not expire just because tech has moved on to their next marketing strategy. "Veatch’s film is of this moment because it situates the narrative being pushed by the AI bros in both its historical and present context —the latter being coverage of environmental damage and the exploitative labor practices behind AI . ...

"A new exposé in the New Yorker paints a different portrait, and it’s substantially more vexing. Drawing on interviews with numerous OpenAI insiders who worked with Altman, the article portrays the CEO not as a technical wiz , but as a skilled manipulator —and one with a surprisingly shallow grasp of the AI systems his company is building. " According to numerous engineers interviewed for the article, Altman lacks experience in both programming and in machine learning —a shortage of expertise that becomes obvious when the CEO mixes up basic AI terms. "It’s important to note that Altman dropped out of a Stanford computer science program after two years.  "Cast as the chief acolyte of the god of scale  or as a genius of digital tech , he enjoys a kind of cult credibility that lets him slip out of tight spots that might ensnare lesser entrepreneurs. "Former OpenAI researcher Carroll Wainwright, speaking to the New Yorker , put it plainly: 'He sets up structure...

"Artificial intelligence has arrived in the field of mental health. Large health systems and independent therapists alike have begun to adopt different AI tools to manage the delivery of mental health treatment. "The speed of the adoption —alongside disturbing incidents of individuals using general-use AI chatbots with catastrophic consequences —is causing some concern among practitioners and researchers. "'There is a lot of fear and anxiety about AI,' says psychologist Vaile Wright, senior director of health care innovation at the American Psychological Association (APA). 'And in particular fear around AI replacing jobs.' "Those concerns were a key issue last month, when 2,400 mental health care providers for Kaiser Permanente in Northern California and the Central Valley went on a 24-hour strike."

"A New York Times analysis found Google's AI Overviews now answer questions correctly about 90% of the time, which might sound impressive until you realize that roughly 1 in 10 answers is wrong. "'[F]or Google, that means hundreds of thousands of lies going out every minute of the day,' reports Ars Technica .  "The Times conducted this analysis with the help of a startup called Oumi, which itself is deeply involved in developing AI models.  "The company used AI tools to probe AI Overviews with the SimpleQA evaluation, a common test to rank the factuality of generative models like Gemini.  "Released by OpenAI in 2024, SimpleQA is essentially a list of more than 4,000 questions with verifiable answers that can be fed into an AI."

"Driven by encouragement from the very top of China's leadership, the world's second-biggest economy has embraced artificial intelligence, sparking both curiosity and concern. "OpenClaw, built by Austrian developer Peter Steinberger, is an example of how this is playing out. "Because it is built on open-source data and tech, the code is available to those who want to customise it to work with Chinese AI models. And that is a huge advantage, because Western models such as ChatGPT and Claude are not accessible in China. " So OpenClaw stirred up a frenzy as more people experimented with its code. "Its popularity did not escape the Chinese government. Several counties and cities provided incentives to encourage entrepreneurs to apply OpenClaw in their businesses —the eastern city of Wuxi offered up to five million yuan ($726,000; £549,000) for manufacturing-related applications, such as robots. "'Everyone in China knows that the government sets the...

"We propose that AI automation is a continuum between: (i) crashing waves where AI capabilities surge abruptly over small sets of tasks, and (ii) rising tides where the increase in AI capabilities is more continuous and broad-based. "We test for these effects in preliminary evidence from an ongoing evaluation of AI capabilities across over 3,000 broad-based tasks derived from the U.S. Department of Labor O*NET categorization that are text-based and thus LLM-addressable.  "Based on more than 17,000 evaluations by workers from these jobs, we find little evidence of crashing waves (in contrast to recent work by METR), but substantial evidence that rising tides are the primary form of AI automation.  "AI performance is high and improving rapidly across a wide range of tasks. We estimate that, in 2024-Q2, AI models successfully complete tasks that take humans approximately 3-4 hours with about a 50% success rate, increasing to about 65% by 2025-Q3.  "If recent trend...

"The current situation of demanding developers starve and bleed themselves dry without support isn't tenable. We need to grapple with this. I'm not yet sure how it all breaks down, and anyone who says they do is lying, foolish, or fanatical. "The works  in it works  is scoped strictly to coding tasks. I have no evidence, and seemingly no one else does, that the same kind of success is available outside the world of highly structured language with deterministic outputs.  "More plainly: I have no reason to expect this technology can succeed at the same level in law, medicine, or any other highly human, highly subjective occupation. "The arguments against generative models would be much easier if their failure rate remained high across all disciplines.  "In this specific domain, it would seem large language models have found a successful niche.  "That's why OpenAI's pivoting to enterprise and coding tools. That's why coding assistants have...

✨AI Mode  "In the context of a standard Initial Public Offering (IPO), the 'billions' being raised are not 'collateralized' in the traditional sense of a secured loan; rather, the process is built on underwriting and regulatory compliance. Dunlap Bennett & Ludwig+1 "Underwriting and Pricing The Underwriter’s Role: Investment banks (underwriters) act as the bridge between the company and the trading floor. They often commit to a 'firm commitment' agreement, meaning they buy the shares from the company first and then resell them to the public. Price Setting: The IPO price is set by these underwriters based on demand from institutional investors during the 'roadshow'. Capital Receipt: The company receives its funding from the underwriters at this set price, regardless of how the stock performs once it hits the trading floor. Justia+4 "Listing Requirements "To even reach the trading floor, a company must meet strict exchange criteria ...

"Defense modernization programs that envision a spine  of AI-enabled technologies across all domains, with a focus on resilience of the system, could also invite foes to paralyze the spine , or at least lead them to believe that they might be capable of doing so. "Deterrence is no longer a slow-moving, bilateral system.  "It is a fast, interconnected, multi-actor environment shaped by nuclear modernization, technological disruption, and shifting political commitments. "The greatest danger is not that deterrence collapses outright, but that it fails in ways we do not anticipate.  "A misinterpreted signal. A limited strike. A decision made too quickly."