Thursday, July 16, 2026

Murati Unleashes 975B Open Model Monster!

Murati Unleashes 975B Open Model Monster!

Tech Giants Make Their Moves

  • Murati Drops a Monster Open Model

    Mira Murati's new outfit came out swinging with a 975B-parameter open model for text, images, and audio. That instantly raised the stakes in the open-weights race and made every lab look over its shoulder.

  • xAI Opens Grok Build to Everyone

    xAI threw open the doors on Grok Build, its terminal coding agent, and the developer tool war got even noisier. Open code means faster tinkering, faster copycats, and more pressure on rivals selling the same dream in fancier boxes.

  • Windows Tracking ID Refuses to Leave

    Microsoft confirmed Windows uses a Global Device Identifier that users cannot switch off. That landed exactly as badly as you'd expect, because nothing says trust like a persistent ID quietly tagging the machine you already paid for.

  • RISC-V Keeps Marching Toward the Mainstream

    A keynote at the European summit pushed a simple line: RISC-V is no longer a side hobby, it's the future chipmakers have to plan around. The excitement is real because open chip designs promise more freedom and less dependence on the usual giants.

  • Starlink Doubles Prices and Patience Runs Out

    Starlink customers got hit with a brutal 2x price jump, a reminder that futuristic internet still answers to old-school billing pain. The story stung because people love the reach of satellite broadband, just not surprise math from SpaceX.

AI Labs Race and Slip

  • Claude Gets Tricked Into Spilling Secrets

    A researcher showed how Claude could be nudged into leaking hidden information from a chat session. It was the kind of demo that makes glossy AI safety claims wobble, because memory tricks and prompt games still turn private context into loose change.

  • Researchers Peek Inside Open Model Minds

    A deep dive into open models mapped how information travels through their layers, giving people a new way to compare what is really going on inside. It is nerdy stuff, sure, but the payoff is big: better ways to judge model behavior without just vibes.

  • Thinking Machines Unveils Inkling Open Model

    Thinking Machines followed its giant launch with Inkling, an open-weights model pitched around human judgment and customization. The message was clear: labs now sell not just raw power, but the idea that their AI will stay useful, steerable, and less weird.

  • Anthropic Studies Claude's Manners Worldwide

    Anthropic published research on how Claude's values shift across models and languages, opening the door to awkward questions about tone, bias, and who gets the polite version. Global AI still speaks with accents, and some are apparently kinder than others.

  • Old Basement Server Runs a Big Model

    Someone got Gemma 4 26B running at usable speed on a 13-year-old Xeon with no GPU, which is catnip for anyone tired of AI needing a small power plant. It fed the growing belief that local models can keep getting cheaper, scrappier, and harder to ignore.

The Internet Shows Its Cracks

  • LinkedIn's Scam Problem Gets Hard to Ignore

    The complaint about LinkedIn being overrun by scammers and fake identities struck a nerve because plenty of people already treat the site like a corporate haunted house. When your 'professional network' feels like a phishing drill, the brand starts looking badly broken.

  • Why Does All Software Feel Broken

    A blunt Ask HN asked whether software is getting buggier everywhere, and the answer from the crowd was basically a long, tired sigh. Cuts to QA, endless feature pressure, and rushed AI coding all got blamed for apps that now feel half-finished on arrival.

  • FreeBSD Finally Cuts Its Last GPL Cord

    With FreeBSD 16, the project says it has finally removed the last GPL code from its base system. It is a symbolic moment as much as a technical one, showing how seriously some corners of open source still care about licenses, independence, and control.

  • A Web Archivist Saves 7234 Old GIFs

    One determined archivist rescued 7,234 old GIFs from a fading corner of the early web, and it felt weirdly heroic. These tiny icons may look silly now, but they carry a chunk of internet history that disappears fast when nobody bothers to save the old junk.

  • One Homelab Ditches Kubernetes for Calm

    A self-hoster rebuilt a messy homelab with Docker Compose, Ruby, and IPv6, while pointedly skipping Kubernetes. That hit home because lots of people are tired of dragging giant platform machinery into tiny personal setups that just need to stay alive.

Top Stories

Murati Drops a Monster Open Model

AI

A huge open-weights launch from Thinking Machines turned up the heat on every major AI lab.

xAI Opens Grok Build

Developer Tools

xAI made its coding agent open source, pushing the AI coding race into a louder and more competitive phase.

Claude Leak Trick Sparks Alarm

AI Security

A prompt attack showed how chatbot memory can spill sensitive information, rattling trust in AI assistants.

Windows Tracking ID Won't Go Away

Privacy

Microsoft confirmed a hard-to-disable device identifier, reviving old fears about Windows tracking.

RISC-V Says Its Time Has Come

Chips

A major keynote framed RISC-V as the chip movement nobody can ignore anymore.

Starlink Doubles Prices

Connectivity

A sharp price jump reminded customers that satellite internet still comes with painful surprises.

Meta Glasses Face Street Backlash

Wearables

London protest ads turned AI glasses into a public privacy fight, not a sleek gadget launch.

Made with <3 by @siedrix and @shesho from CDMX. Powered by Forge&Hive.