GPT-5's big new feature: less lying? ↗
On this week’s Vergecast, Adi Robertson and Alex Heath dive into OpenAI’s GPT-5 launch alongside the open-source GPT-OSS. They unpack odd survey charts, debate how to measure an AI’s deception when it’s too good to spot, and highlight GPT-5’s subtle but broad upgrades—especially for developers and coding. The deprecation of the model switcher hints at a smoother experience for all ChatGPT users.
Everything we think we know about the Google Pixel 10 phones ↗
Google’s Pixel 10 lineup—including the Pixel 10, 10 Pro, 10 Pro XL and a delayed 10 Pro Fold—arrives August 20 in NYC. The base Pixel 10 gains a triple-camera array with telephoto but uses smaller sensors; Pro models keep superior optics. Leaks reveal bright new colors for the standard model and muted tones for Pros; the Fold skips black. Expect 25W magnetic Qi2 charging via a new Pixelsnap ecosystem, a faster Tensor G5 on TSMC’s 3nm, on-device AI tools like Speak-to-Tweak and Pixel Sense, and familiar pricing.
Microsoft’s new Copilot 3D feature is great for Ikea, bad for my dog ↗
Microsoft quietly launched Copilot 3D, a free feature that converts any clear 2D image into a fully editable 3D model—no text prompts required. Ideal inputs have clean backgrounds, good lighting, and clear depth cues. Once generated, these models can be used in game creation, animation, 3D printing, VR/AR, and more, all in just a few clicks.
OpenAI unveils ChatGPT-5 and its hyped ‘PhD level’ intelligence struggled with basic spelling and geography ↗
OpenAI’s new GPT-5, touted as “PhD-level” AI, has surprised users by mangling simple spelling and geography. Testers found it insisted “blueberry” has three Bs, miscounted Rs in Northern Territory and invented states like “Krizona.” The hiccups stem from GPT-5’s real-time router choosing among multiple submodels. Despite these flaws, OpenAI says the system will improve as it learns from user feedback.
When a journalist uses AI to interview a dead child, isn’t it time to ask what the boundaries should be? | Gaby Hinsliff ↗
Ex-CNN journalist Jim Acosta recently interviewed an AI-generated Joaquin Oliver, the 17-year-old Parkland shooting victim, raising urgent questions about ethical boundaries of resurrecting the dead through technology. Gaby Hinsliff argues that while AI avatars can comfort grieving families and bolster campaigns for gun control, they also risk exploiting vulnerabilities, blurring lines between life and death, and challenging legal rights, selfhood and our societal readiness for permanent digital ghosts.
xAI’s legal chief steps down after whirlwind year ↗
Robert Keele, xAI’s first legal chief, is stepping down after just over a year to spend more time with his toddlers and due to “daylight between our worldviews” with Elon Musk. He joined in May 2024 ahead of xAI’s $6 billion Series B and Musk’s X acquisition. Former NASA rocket scientist–turned–lawyer Lily Lim will succeed him, adding to a string of executive shakeups across Musk’s companies.
OpenAI gives some employees a ‘special’ multimillion-dollar bonus ↗
On the eve of GPT-5’s rollout, OpenAI CEO Sam Altman surprised staff with a Slack message announcing a special one-time award for researchers and engineers in applied engineering, scaling, and safety teams. He cited market shifts and pledged ongoing compensation increases as the company continues to perform better.
Andrew Bragg on AI, housing and what the Coalition got wrong - podcast ↗
Two weeks before Labor’s economic roundtable, Coalition shadow minister Andrew Bragg warns against over-regulating AI, admits the Coalition bungled migrant community outreach, and reveals how he still finds time for long runs and audiobooks amid federal politics chaos.
Tesla shuts down Dojo, the AI training supercomputer that Musk said would be key to full self-driving ↗
Tesla is disbanding its in-house Dojo supercomputer team, ending its chip-making push for full self-driving. Lead Peter Bannon is leaving, and remaining engineers will shift to other data-center roles. The move follows 20 departures that spun out AI startup DensityAI and signals a strategic pivot toward external partners like Nvidia, AMD and Samsung under Elon Musk’s revamped AI roadmap.
OpenAI gets caught vibe graphing ↗
During OpenAI’s GPT-5 livestream, several bar charts meant to highlight the model’s prowess in deception detection and other metrics featured mismatched scales and mis-sized bars—showing GPT-5 underperforming smaller models with larger bars. The gaff prompted a self-deprecating “mega chart screwup” from CEO Sam Altman and an apology from marketing staff. While OpenAI hasn’t commented further, the sloppy visuals clash with its message of improved accuracy and hallucination reduction on launch day.