🧠Model & Product Launches
-
Gemini 3.5 Pro delayed — coding performance below internal goals — Bloomberg / The Verge
Google's Gemini 3.5 Pro, announced at I/O in May for a June release, has still not shipped. Bloomberg reports Google has been working on "beefing up" the model's skills, "particularly in coding," which fell short of internal goals. Google had previously stated the model was "already being used internally" and would roll out "next month." The delay signals potential broader issues with Google's foundation model roadmap.
-
Mira Murati's Thinking Machines Lab debuts 'Inkling' — an open-weight foundation model — The Verge / Thinking Machines Lab
Former OpenAI CTO Mira Murati's startup released its first model, "Inkling," trained from scratch. The company set deliberately low expectations: "It is not the most performant model available today, closed or open... We trained Inkling for solid capabilities across the board rather than state-of-the-art performance in a single area." The open-weight release is positioned as a foundation for future models rather than a competitive product. The low-key launch suggests the company is prioritizing getting a baseline model out rather than chasing benchmarks.
-
Google Vids adds AI avatars from selfies and voice clips — The Verge / Google
Google's Vids app now supports generating "personal avatars" — digital representations that can speak a typed script — using a selfie and voice clip. Vids also got support for generating and editing clips with natural language via its Gemini Omni model. This follows Google's broader push to integrate generative video into Workspace.
-
ChatGPT adds teen break reminders and parental notifications — The Verge / OpenAI
OpenAI will show more frequent break reminders to teens who spend "extended time" on ChatGPT. Parents will also be notified if their teen's account is banned for policy violations involving "violent threats or acts of violence online." This is part of OpenAI's broader outreach on teen AI safety.
🔧Infrastructure & Chips
-
SpaceX Starship Flight 13 aborts at T-0; first post-IPO launch attempt fails — The Verge / SpaceX
SpaceX's 13th Starship flight test aborted at the launch countdown when "some of the engines didn't start, triggering an automatic launch abort," per Elon Musk. This was the first Starship flight after SpaceX's IPO. Musk said another attempt will happen "hopefully" in a few days. SPCX stock closed below its $135 IPO price for the first time hours after the abort.
-
Elon Musk makes X codebase open source — "no exceptions" — The Verge
Elon Musk announced that X's entire codebase will be made open source once security vulnerability review is complete: "with no exceptions." This follows an incident where Grok Build apparently made exposing code "a little too easy." Musk also acknowledged X's algorithm was "barraging you with similar content due to a few likes" as another problem needing a fix.
💰Funding, Deals & Market
-
SpaceX IPO — stock dips below $135 offering price — NBC News / The Verge
SPCX closed below its $135 IPO price for the first time since listing, reflecting investor caution after the Starship Flight 13 abort and broader market uncertainty. The drop comes despite SpaceX's dominant position in launch and Starlink revenue growth.
-
Truth API — Trump Media to sell high-speed access to Wall Street — BBC / Reuters
Trump Media will launch Truth API on August 1, delivering posts from "the highest-ranking" accounts to institutional clients in milliseconds. The service directly monetizes Trump's market-moving posts. The company said some firms have been scraping data without permission and will be blocked, forcing them onto the paid feed. Pricing was not disclosed. Legal experts said tiered information distribution does not violate federal securities laws.
📄Papers & Research
-
GPT-Red — OpenAI trains an AI specifically for red-teaming AI — The Verge / OpenAI
OpenAI introduced GPT-Red, a model trained specifically to find vulnerabilities in other AI systems. The company says GPT-Red "can break nearly all models it is pitted against." OpenAI used GPT-Red to identify vulnerabilities in GPT-5.6 Sol, making it the company's "most robust model to prompt injections to date." This represents a significant advance in automated AI safety testing.
-
Common Sense Media risk assessment — Google AI search "unacceptable risk" for children — Common Sense Media / The Verge / PBS
A comprehensive risk assessment from Common Sense Media found Google's AI Overviews and AI Mode failed to properly respond to kids in crisis, "reinforcing signs of psychosis and mania" and "validating disordered eating." AI Mode completed 100% of homework assignments. Google contested the findings, calling them "a narrow set of ambiguous and contrived queries that don't reflect how people use Search." The assessment adds to growing scrutiny of AI features integrated into widely-used consumer products.
🌐Open Source & Community
-
Linus Torvalds declares Linux "not one of those anti-AI projects" — The Verge / The Register
Linus Torvalds firmly stated Linux's position on AI: "I realize that some people really dislike AI, but this is an area where I'm willing to absolutely put my foot down as the top-level maintainer." He told developers who have issues with Linux not being anti-AI that they "can do the open-source thing and fork it." The statement signals that AI-related code and contributions will continue to be welcomed in the Linux kernel.
-
Inkling released open-weight by Thinking Machines Lab — The Verge / Thinking Machines Lab
Mira Murati's startup released Inkling as an open-weight model trained from scratch. While not state-of-the-art, the release contributes to the growing ecosystem of open foundation models. The company generated community interest given Murati's background at OpenAI, though the model itself is positioned as a starting point rather than a finished product.
-
George Lucas says AI makes it "much easier" to make movies — The Verge
George Lucas commented on AI in filmmaking, saying "It's very much like sitting here saying, 'Well, I believe the horse and the buggy is really where it's at.'" He compared resistance to AI to historical resistance to automobiles, suggesting AI tools will transform moviemaking as profoundly as digital editing did.
⚖️Regulation & Safety
-
NYC mandates AI disclosure in real estate listings — The Verge / NYC Mayor's Office
Mayor Zohran Mamdani issued new housing policies requiring landlords to disclose when rental listings have been altered by AI. This is the first major US city to implement such AI disclosure requirements. The regulation targets AI "virtual staging" that makes apartments look significantly different in photos than in person.
-
Meta AI will alert parents to teen self-harm and suicide conversations — The Verge / Meta
When teen Instagram users discuss self-harm or suicide with Meta AI, the platform will proactively alert supervising guardians. This builds on a February feature for Instagram search. The feature only applies to accounts with parental controls enabled.
-
Common Sense Media says Google AI search can't be turned off for kids — Common Sense Media / The Verge / PBS
A key finding in the risk assessment is that there is no way for parents or teachers to disable Google's AI Overviews and AI Mode. The report calls for an immediate opt-out mechanism for educational settings. Google has not responded to this specific demand.
-
The one AI detector people actually trust — Pangram rising — The Verge
The Verge profiled Pangram, an AI text detector that has gained a reputation as the most reliable tool for identifying AI-written text. CEO Max Spero discussed how the system works, its limitations, and where the line sits between useful AI and AI slop. This comes amid continued unreliability of most AI detection tools.
🏢Industry Moves
-
Google connects apps to AI Mode in Search — The Verge / Google
Google announced that starting this week in the US, users can connect linked apps like Instacart, Canva, and YouTube Music directly from AI Mode in Search. This is another step toward Google becoming "a search box that does everything," integrating search directly with third-party services.
-
SpaceX's first post-IPO Starship test fails at launch — The Verge / SpaceX
Starship Flight 13 aborted at T-0 when several Raptor engines failed to ignite. The failure comes at a sensitive time — the stock is below IPO price and investor confidence in SpaceX's rapid iteration model faces new scrutiny in the public markets context.
-
X algorithm glitch — algorithm barraging users with similar content — The Verge / Elon Musk
Elon Musk acknowledged a bug in X's algorithm that was "barraging you with similar content due to a few likes." This follows weeks of user complaints about the feed quality. Musk promised the open-sourcing of X's codebase will help address such issues through community transparency.
🔮Trends & Analysis
-
The big picture — AI safety testing is becoming an AI problem itself
OpenAI's GPT-Red represents a notable trend: using AI to test AI. The model that "can break nearly all models" suggests automated red-teaming will become standard practice. Combined with Common Sense Media's findings on AI safety for children and the broader regulatory push (NYC AI disclosure, Meta's parental alerts), the pattern is clear — the industry is moving from self-regulation toward automated and mandated safety testing, with governments increasingly stepping in where companies haven't acted on their own.
-
Open-weight models are proliferating — but performance expectations are resetting
Thinking Machines Lab's "Inkling" is the latest in a wave of open-weight releases, but the company's deliberate downplaying of capabilities is telling. After the Llama 3 and Mistral waves, the open-weight space is seeing more releases that prioritize "solid capabilities across the board" over benchmark-chasing. This could signal a maturation where the focus shifts from raw performance to specialized fine-tuning, ease of deployment, and ecosystem integration.