opinion: The micro-computer makers shares surged this week after an X post tied the AI agent to Pi demand (03:33 | register)

: Big Blue has been saying this itself since 2023 (03:33 | register)

: Says DeepSeek, Moonshot, and MiniMax are using 'distillation' to gin up their own models (03:33 | register)

: One massive round of firings is apparently better for morale than a drip-drip-drip of death (03:33 | register)

: Legal language change aims to make longstanding policy clear (00:55 | register)

: When it gets stuck, the bot will escalate rather than hallucinate (00:55 | register)

: Bezos-corp blames user error for outage, 'specifically misconfigured access controls' (Tegnap | register)

: Up to 8 exaFLOPS of super sparse AI compute (Tegnap | register)

: Russinovich and Hanselman say firms must train juniors to fix agent mistakes not replace them with prompts (Tegnap | register)

: Company tries to curb strain by banning customer accounts for 'malicious' usage (Tegnap | register)

: Pretending the software is sentient makes it sound more powerful (Tegnap | register)

exclusive: Just less than before, according to the ORCA test (Tegnap | register)

: Because nothing says hospitality like a bot counting your pleases (Tegnap | register)

: Consultancy to monitor usage by meatbags with corporate aspirations (Tegnap | register)

: Watchdogs warn models that can generate realistic images of people must comply with data protection laws (Tegnap | register)

: Security is 'dangerously behind' though, as devs 'treat it as something to solve later' (Tegnap | register)

: 'Just trust us' Big Tech's hackneyed catchphrase makes an unwelcome return (Tegnap | register)

: OpenAI CEO takes really, really long view on energy efficiency (Tegnap | register)

: Complaints pile up from users after months of conversations disappear. Google insists its just a temporary bug (Tegnap | register)

: Whac-A-Mole season continues as Redmond finds yet another corner to stuff its 21st century Clippy (Tegnap | register)

: TrendForce says eight hyperscalers are set to pour $710B into servers and infrastructure (Tegnap | register)

: Report claims more vulnerabilities created than fixed as remediation gap widens (Tegnap | register)

: Off-the-shelf tools helped Russian-speaking cybercrime group run riot (Tegnap | register)

: Analyst warns soaring DRAM and NAND costs could push entry-level devices out of reach (Tegnap | register)

: Skill at buzzword bingo also required as company seeks innovative and disruptive visionary (Tegnap | register)

: Ministry of Justice wowed by Ontario's paperless system, announces £12M for AI unit (Tegnap | register)

Asia In Brief: PLUS: Chinas sword-wielding humanoid robots; Australian court swamped by AI filings; Vietnams 25km overwater drone delivery; And more! (Tegnap | register)

A new technique leverages computing downtime to accelerate the training process for reasoning large language models (LLMs) without any additional computational overhead. (Tegnap | mit)

: MIT CSAIL's 2025 AI Agent Index puts opaque automated systems under the microscope (Tegnap | register)

: Selling so many agents they've cooked up a way to measure what they do (Tegnap | register)

: You'll find these days that there's no hiding place (Tegnap | register)

: Anthropic fixed the flaws - but the AI-enabled attack surfaces remain (Tegnap | register)

: Claude, ChatGPT, and Gemini all had different personalities and reasoning tactics, but the endgame was the same (Tegnap | register)

: From AI conflation to thin evidence, a new report calls many climate claims greenwashing (Sze. | register)

: AI model said to show improved reasoning capabilities (Sze. | register)

: AI firm drops key safety pledge as Pentagon dispute drags on (Sze. | register)

: Agent-making tool that mimics human workers is about to get its enterprise close up. (Sze. | register)

A system known as PhysiOpt complements generative AI models with physics simulations to create 3D models of personal items. Users can prompt the system using text or images, and get a blueprint that works in the real world when fabricated. (Sze. | mit)

Sponsored Post: Feeling stuck with your AI initiatives? Gartner has the answer (Sze. | register)

: For now, it might not function outside of a lab (Sze. | register)

: Uses Vite and Claude to sidestep Vercel lock-in (Sze. | register)

: Sometimes the 'S' word slips through even the best media training (Sze. | register)

: Rush is on to push forward sympathetic candidates from both parties ahead of midterms (Sze. | register)

: Study of 11 LLMs shows they rarely refuse to answer, even when they probably should (Sze. | register)

: 'Why not 12?' says lawyer (Sze. | register)

: Self-generated skills don't do much for AI agents, study finds, but human-curated skills do (Sze. | register)

: It's not chatbot psychosis, it's 'math and engineering and neuroscience' (Sze. | register)

Opinion: The DECwriter got me hooked in 1975. 'Clawdine' feels like a wonderful new beginning (Sze. | register)

Interview: Co-author Jon Kern says AI coding tools amplify strengths and expose weaknesses (Sze. | register)

Sponsored Post: Companies pursuing AI face a gap between perception and practice (Sze. | register)

: Note to secret agents: ChatGPT is NOT a private diary (Sze. | register)

A new AI framework identifies which data about a cell are captured by one measurement modality and which are shared across multiple modalities. This gives researchers a more complete picture of the cell state and could help them understand disease mechanisms and plan treatments. (Sze. | mit)

: IT services companies are largely immune to AIpocalypse, although the outlook is not good for entry-level jobs (Sze. | register)

Strahinja Janjusevic, a masters degree student in the MIT Technology and Policy Program, focuses on applying AI techniques to cybersecurity problems and examining the policy implications of these advancements, especially in the context of maritime cybersecurity. (Sze. | mit)

: Who needs to express themselves through music when a bot will do it for you with nothing but a prompt? (Sze. | register)

: Discovery is getting cheaper. Validation and patching arent (Sze. | register)

opinion: Protect the robot, sacrifice the human (Sze. | register)

: Data Loss Prevention? Yeah, about that... (Kedd | register)

: Plants expected to begin operations as early as 2028 pending approval by state government (Kedd | register)

: Survey says 80% of firms see no gains from the tech (Kedd | register)

: GitHub itself to blame for AI slop pull requests, say devs (Kedd | register)

: EV maker avoids 30-day license suspension after state ruling on self-driving claims (Kedd | register)

: Seemingly complex strings are actually highly predictable, crackable within hours (Kedd | register)

: No, customers aren't laughing either as pressure from memory shortages bites (Kedd | register)

: CEO lauds security researchers, insists they're not 'inputs' (Kedd | register)

: Upstart's 5th-gen RDU aims to undercut Nvidia's B200 on speed and cost (Kedd | register)

: Parliament committee finds AI BS helped shape a real-world decision (Kedd | register)

: Visa applications down, executives emigrating, and AI blamed for the rest (Kedd | register)