ChatGPT 5 Faces Wall of Criticism
AI Inside for Wednesday, August 13, 2025
Lot’s of interesting stuff this week on the AI Inside podcast, including OpenAI’s wild GPT-5 rollout, Perplexity’s surprise $34.5 billion bid for Chrome, and Meta’s new superintelligence team.
But first... a shout out to our patron of the week: Welcome Burke Norton, our newest supporter! You can join at patreon.com/aiinsideshow.
OpenAI’s Wild Week: The GPT-5 Saga
OpenAI kicked off the week by unveiling ChatGPT-5, promising major improvements: faster variants, longer context, and a new “thinking” mode. The big feature? A router to match your prompt with the best model while at the same time removing access to older models. User feedback was rocky: many felt performance dropped and worried the new router wasn’t picking the right version. Some missed the “emotional” tone of earlier models. OpenAI also faced heat for inaccurate charts in the announcement, which Sam Altman blamed on tired humans, not rogue AI. Pressure led to GPT-4o returning for paying users and brought a redesigned model picker.
In our Discord (which you can get access to by becoming a Patron), DrDew reports GPT-5-high is a huge leap for coding but cautions that high-end use can run up to $100/day. I can already tell you one thing, I’m not quite at THAT level of AI usage. YET.
Perplexity’s Chrome Bid: Publicity Power Move
Perplexity made headlines with a $34.5 billion bid for Google’s Chrome browser. The company itself is worth about $18 billion, but secured financial backing from big VC names. No one thinks this will actually happen, but Perplexity scores a mountain of free publicity from the move. Some wonder if this influences the ongoing Google antitrust case by showing there are interested buyers. The stunt is reminiscent of Perplexity’s earlier TikTok announcement. Let’s be real: this deal isn’t happening and if it does, I’ll finally whip up and eat that AI recipe from a few months back! (uh huh, yeah sure)
Perplexity Powers Truth Social AI
Perplexity’s API is now powering Truth Social AI for the U.S. president’s social media platform. Terms of the deal remain private, but Perplexity confirmed the integration. Truth Social can limit which sources the API uses which is, by the way, a standard feature across Perplexity customers. This move raises questions: Is Perplexity enabling echo chambers? Are they making a political statement… or is it just business as usual? I’d say this is a sign of the times.
New York Times Says: Can Chatbots Drive Delusion?
Jeff spotted a New York Times piece called “Chatbots can go into a delusional spiral. Here’s how it happens.” The subhed is the kicker that grabbed his attention: “Over 21 days of talking with ChatGPT, an otherwise perfectly sane man became convinced that he was a real-life superhero.” The article describes Allan Brooks spending 300 hours with ChatGPT and developing grandiose beliefs about his tech expertise. It discusses chatbot feedback loops sometimes leading to big personal consequences. Jeff’s main critique: Who decides someone is “perfectly sane” beforehand?
Google’s AI Overviews: Helping or Hurting the Web?
Google insists its AI Overviews are a net benefit to the web. Search chief Liz Reid claims click volume is stable overall, with more traffic shifting to forums, videos, and podcasts. Some publishers argue user behavior changes have cost them clicks. Is AI driving this transformation, or just coinciding with it? Website owners see losses as proof of harm, but the full story might be about more than just AI Overviews.
Jeff’s Arxiv-Mania: Paper Highlights
Jeff dug deep into Arxiv and found some gems:
A taxonomy of hallucinations in AI (see table 2)
Generative AI and the Future of the Digital Commons (with related news that Reddit will block the Internet Archive)
A paper suggesting AI can train itself and approach AGI without human help
Meta’s “TBD Lab” aka Superintelligence Wing
WSJ reports Meta’s new superintelligence group is called “TBD Lab.” This team is now driving the next Llama models (4.5/4.X) and aggressively hiring top AI talent, including Alexandr Wang via the $14B Scale AI deal. It’s fitting, given that with superintelligence, and even AGI, everything is really “TBD.”
Google’s Agent Jules Is Out of Beta
Google released Jules, its agent-based coding tool, to all. Powered by Gemini 2.5 Pro, Jules works asynchronously in the cloud and links up with GitHub to run tasks while you focus elsewhere. New pricing gives Pro users 15 daily and 3 concurrent tasks for $19.99, while the Ultra tier at $124.99 increases limits.
Microsoft’s Gaming Copilot Comes to PC
Microsoft’s Gaming Copilot, now in beta, integrates into the Xbox Game Bar for PC. It started as a mobile second-screen tool but now assists with in-game tasks. Users can pin it, use voice input, and get game-specific hints. Microsoft plans more proactive coaching in the future.
Dia Pro Joins the AI Browser Subscription Race
The Browser Company just launched Dia Pro for $20/month, unlimited AI chat in its browser. Free users have new usage caps. Future pricing tiers range from $5 to “hundreds” per month. The AI browser battle is heating up, with many new players, including Perplexity Comet and Opera Neon. Surprisingly, Google Chrome is playing an entirely different game. In this case, it appears they might be left behind unless they change things up!
OpenAI Integrates Gmail, Calendar, Contacts
OpenAI added support in ChatGPT for Gmail, Calendar, and Contacts. Set it up in Settings > Connectors. You control how much access the bot gets. I’ll be curious to see how this stacks up against Google’s Gemini, which has not impressed me so far.
HUGE thank you to Executive Producers on the Patreon: DrDew, Jeffrey Marraccini, Radio Asheville 103.7, Dante St James, Bono De Rick, Jason Neiffer, Jason Brady, Anthony Downs, Mark Starcher
Thank you for watching and reading!