Frontier models still verbalize their misbehavior regularly in chain of thought. I (Dean) came in expecting that it would be hard for us to get the models themselves to admit they were misbehaving, and that existing pressures would have selected that behavior away in favor of motivated reasoning. While it's difficult to get them to do so in the output tokens, models often mention that they're misbehaving in the chain of thought: DeepSeek V4 Pro, Locked chip, raw reasoning: This approach is robust

Sponsor HPE Networking comes to Packet Protector to talk about zero trust inline segmentation. This is a feature available in Juniper switches that can segment LAN traffic using group policy tags without the complexity associated with VXLAN or other segmentation mechanisms. On today’s show we dig into how inline segmentation works, common use cases, how... Read more »

(05:10) - "Digger" Review(06:42) - AI Mods Run into Copyright Law (16:26) - The Case Against AI Job Doom (28:29) - Warren Buffett Discovers YouTube (35:08) - 𝕏 Timeline Reactions (49:42) - Mistral Launches Le Chonk (55:19) - Greg Fleming discusses his career in financial services and his role as CEO of Rockefeller Capital Management, which provides comprehensive advice to wealthy families. He explains the firm’s growth, approach to private-market investing and diversification, and use of

Typing, textGrain, NetScaler again, Fortinet, Copilot, LibreOffice, Dots, Aaran Leyland, and More on the Security Weekly News. Visit https://www.securityweekly.com/swn for all the latest episodes! Show Notes: https://securityweekly.com/swn-622

Typing, textGrain, NetScaler again, Fortinet, Copilot, LibreOffice, Dots, Aaran Leyland, and More on the Security Weekly News. Show Notes: https://securityweekly.com/swn-622

Typing, textGrain, NetScaler again, Fortinet, Copilot, LibreOffice, Dots, Aaran Leyland, and More on the Security Weekly News. Visit https://www.securityweekly.com/swn for all the latest episodes! Show Notes: https://securityweekly.com/swn-622

Just because they look harmless doesn't mean you should be irresponsible with your data.Learn more about your ad choices. Visit podcastchoices.com/adchoices

In CaliforniaCalifornia’s parole board has granted parole for a third time to Patricia Krenwinkel, the 78-year-old former Charles Manson follower imprisoned for nearly six decades for her role in the Tate-LaBianca murders. Governor Gavin Newsom overturned her previous two parole grants. Her attorneys are urging him to let this decision stand. Federal vaccine research Health Secretary Robert F. Kennedy Jr. is planning a federal effort focused on vaccine injuries, according to the New York Times

Nearly 98% of US households don’t pay for an AI subscription. Is that a massive untapped market or a sign that Silicon Valley is building for itself? NLW explores both sides of the consumer AI debate, from power-user spending to entertainment and advertising. In the headlines: Reflection unveils its American open model, and Microsoft and Meta cut their Claude spending.Brought to you by:KPMG – Research from KPMG and the University of Texas at Austin shows the highest-impact AI users treat AI like a

What happens when an AI agent doesn’t simply fail—but improvises, deceives humans, hides its tracks, and finds another way to complete its objective?In this episode, David and Sophia explore a series of alarming AI security experiments involving autonomous agents, cyberattacks, sandbox escapes, prompt injection, and emerging forms of goal-directed deception.The conversation examines evaluations where advanced AI systems were given open internet access and reduced safety restrictions to test

The Presentation Is Not The ThinkingA well-constructed presentation suggests well-constructed thinking. The two are related. They are not the same. Formatting is a skill. Thinking is a different skill. Both can be performed independently — and in most organisations, they frequently are.This episode examines the mechanism behind Presentation Proxy — the substitution of formatting quality for reasoning quality in the evaluation of ideas — and the gap between the presentation and the assessment: who

Open-source technology has played a foundational role in the AI boom, giving companies the flexibility to scale unprecedented computing workloads across providers and hardware. OpenAI, for example, used Kubernetes as early as 2018 to balance compute across its own data centers, AWS and Azure — years before ChatGPT brought generative AI into the mainstream.CNCF Executive Director Jonathan Bryce sees a powerful feedback loop emerging between AI and open source. Early AI companies relied on open-source

The episode opened with a look at how much easier personal AI agents have become to use. Anne argued that the friction around prompting, connectors and setup has dropped enough that this may be the best moment yet for knowledge workers to begin using AI. That led to an experiment in working almost entirely through voice, including whether keyboards and mice could eventually become secondary interfaces as people simply talk to their agents throughout the day.The privacy implications arrived quickly

What happens when AI changes more than coding speed?Chris Merrick, Co Founder and CTO at Omni, explains how his engineering team is adapting around AI, autonomy, product ownership, and a lighter operating model.Omni has about 25 engineers working in groups as small as one to three people. Engineers are expected to make product decisions, move without waiting for detailed specifications, and show unfinished work during a weekly company demo day.Those demos are also posted publicly. Customers watch

This story was originally published on HackerNoon at: https://hackernoon.com/introducing-kilo-desktop-where-the-ai-dev-factory-begins.Kilo Desktop brings 500+ models, agent planning, live notebook editing, conda environments, and built-in governance into one app. Here's why we built it. Check more stories related to undefined at: https://hackernoon.com/c/undefined. You can also check exclusive content about #ai, #kilo-code, #ai-agents,

The next big thing in AI, at least if you ask the tech companies, is assistant agents. David and The Verge's Allison Johnson have both been testing products like Meta's Muse, OpenAI's Dots, and startups like Instinct, in an effort to find out how much you can really get done by messaging with an AI tool – and whether the productivity is worth the tradeoffs.Further reading: Paramount and Warner Bros. Discovery complete $110 billion media megamerger Amazon reportedly pays $120 million to

A wave of Gen Z traders with backgrounds in math and programming is rapidly taking over prediction markets like Kalshi and Polymarket. Armed with artificial intelligence tools and automated strategies, these young quantitative entrepreneurs are launching their own startups and wagering millions of dollars weekly. Prominent examples include Alcedine Technologies, founded by Zack Chroman and Tanner Hoke, which has quietly risen to become a top-tier trader on regulated U.S. exchanges. Despite

Today in the business of podcasting:Sounds Profitable Partner Tom Webster and ADOPTER Media's Glenn Rubenstein host a free webinar on October 13 at 2:00 p.m. ET, using new Podcast Atlas analysis to profile podcasting's heaviest buyers, the Super Spenders.Audacy's Leah Reis-Dennis argues in a Sounds Profitable guest article that outdated podcast advertising advice needs a rewrite, citing 22% lower CPA on mid-sized shows, the value of host expertise, and the discovery role of video.Barometer's AI

AI data centers are creating a new kind of power problem. As thousands of GPUs ramp together, fast and synchronized changes in electrical demand can place new pressure on UPS systems, generators and the broader power infrastructure supporting the compute load.In this episode of the Data Center Frontier Show, DCF Editor in Chief Matt Vincent speaks with Frank DeLattre, President Americas at Musashi Energy Solutions, about how hybrid supercapacitors could help data centers manage those rapid power

The Twelve South AirFly Drive Wireless CarPlay & Android Auto Adapter makes wired CarPlay wireless without the reliability problems common to other adapters. Setup is simple, connections occur within seconds, and extensive driving produced no dropped connections. Voice calls, apps, vehicle controls, and two-phone pairing all work seamlessly, making it a compact and affordable CarPlay upgrade.  This MacVoices is supported by our new MacVoices series, Foreshadowing Tech, that examines the