Sam Altman addresses ‘bumpy’ GPT-5 rollout, bringing 4o back, and the ‘chart crime’ ↗
In a Reddit AMA, OpenAI CEO Sam Altman admitted GPT-5’s new real-time model router briefly underperformed, making it seem “dumber” than GPT-4o. He promised fixes to the autoswitcher, greater transparency on which model answers queries, and is weighing a GPT-4o option for Plus subscribers. Plus users will also get double rate limits. Altman also laughed off the infamous “chart crime” in the GPT-5 launch deck and vowed to stabilize performance and keep listening to feedback.
RIP, Microsoft Lens, a simple little app that’s getting replaced by AI ↗
Microsoft is retiring its popular Lens mobile scanner app this fall, steering users toward the Copilot AI chat app instead. Lens will be phased out on iOS and Android starting September 15 and fully removed from app stores by November 15, with scanning disabled after December 15. Launched in 2015 and downloaded over 92 million times, Lens offered free, fuss-free document, receipt and whiteboard scanning—and none of Copilot’s deeper integration or accessibility features.
ChatGPT is bringing back 4o as an option because people missed it ↗
OpenAI quickly restored GPT-4o for paid ChatGPT users after replacing it with GPT-5 by default. Fans mourned the loss of the older model’s personable, creative responses—some calling it a companion—and even canceled subscriptions. CEO Sam Altman confirmed Plus subscribers can revert to GPT-4o while OpenAI gauges usage. The company also vows to enhance GPT-5’s speed and transparency, letting users see which model responds and raising usage limits.
Former Googlers’ AI startup OpenArt now creates ‘brain rot’ videos in just one click ↗
Surreal “brain rot” clips—think sharks in sneakers or cappuccino-headed ballerinas—are lightning-fast to make with OpenArt’s new One-Click Story feature. Input a sentence, script or song and instantly generate a minute-long video using Character Vlog, Music Video or Explainer templates. The platform taps 50+ AI models and lets you tweak storyboards, but IP slip-ups remain a risk. Founded by ex-Google engineers, OpenArt boasts 3M MAUs, $5M raised and is on track for $20M ARR, with plans for dual-character chats and a mobile app.
Pinterest CEO says agentic shopping is still a long way out ↗
Pinterest’s CEO Bill Ready told investors that fully autonomous “agentic” shopping—where AI buys items on your behalf—is still years away. Instead, the company is focusing on AI-driven recommendations and personalization to act like a personal shopping assistant that “just gets” your style. Ready highlighted Pinterest’s AI-powered search, visual discovery and ad tools, while acknowledging user concerns over low-quality AI content. In Q2, Pinterest reported $998 million in revenue (beating estimates) but missed on EPS, sending its stock lower. Gen Z now makes up over half of its monthly users, with male users rising 95% year-over-year.
OpenAI priced GPT-5 so low, it may spark a price war ↗
OpenAI stunned the AI market by undercutting rivals with GPT-5 pricing—$1.25 per million input tokens and $10 per million output (plus caching fees). That matches Google’s Gemini rates and sharply beats Anthropic’s Claude Opus costs. Early testers praise its coding chops and aggressive cost-effectiveness versus GPT-4o. Industry watchers say OpenAI’s move could trigger an AI model price war, easing fees for startups and developers burdened by high inference expenses.
Apple Intelligence’s ChatGPT integration will use GPT-5 starting with iOS 26 ↗
OpenAI rolled out GPT-5 on Thursday, but Apple Intelligence won’t tap into it until iOS 26, iPadOS 26, and macOS Tahoe 26 arrive—likely next month, Apple told 9to5Mac. Today, Apple Intelligence relies on GPT-4o for Siri queries and its Visual Intelligence feature. Meanwhile, GPT-5 is live for all ChatGPT users, even on the free tier, with ChatGPT seeing 700 million weekly users.
Tesla’s Dojo supercomputer is DOA — now what? ↗
Tesla has disbanded its Dojo supercomputer project after the team’s leader and many employees departed for another AI effort, Bloomberg reports. The move disrupts Tesla’s plans to build an in-house training rig for Autopilot, Full Self-Driving, and Optimus AI models, and underscores ongoing challenges in retaining top talent at the automaker.
Meta acquires AI audio startup WaveForms ↗
Meta has acquired WaveForms, an eight-month-old AI voice startup known for its “Speech Turing Test” and emotional general intelligence tech, to bolster its Superintelligence Labs. WaveForms raised $40 million at a $160 million valuation from Andreessen Horowitz. Co-founders Alexis Conneau (ex-OpenAI) and Coralie Lemaitre are joining Meta. This undisclosed deal follows Meta’s recent purchase of PlayAI, marking its second major AI audio acquisition in a month.
Google tests revamped Google Finance with AI upgrades, live news feed ↗
Google is rolling out an AI-powered redesign of Google Finance in the U.S. Users can ask detailed financial questions and receive AI-driven answers with links, access advanced charting tools (including candlestick and moving average envelopes), and track real-time commodities and crypto data alongside a live news feed. The update aims to keep users engaged on Google Finance instead of turning to rivals like Yahoo Finance or AI chatbots. A toggle lets users switch between the new and classic designs as the rollout unfolds over the coming weeks.