AI For Humans: Weekly AI News, Tools & Trends · Kevin Pereira & Gavin Purcell

OpenAI's Code Red Vs Google Gemini's AI Assault

·49 min·1 clip

OpenAI has declared a "code red" in response to the competitive threat from Google's Gemini 3 Pro, as discussed by hosts Gavin and Kevin Pereira. The leaked memo, reported by The Information, signals a strategic refocus on OpenAI's core product amid market share concerns. Hosts Gavin and Kevin, veterans of tech commentary, analyze the implications for the AI industry's balance of power. The episode details OpenAI's potential shift away from side projects like Pulse and the rumored development of two new internal models codenamed "Garlic" and "Shallot Pete." Mark Chen, OpenAI's Chief Research Officer, stated in an interview that the company has internal models performing at Gemini 3's level and is focusing on pre-training advancements. Google's Gemini is gaining traction, with hosts noting exploding download charts for Google AI products and its utility for coding and video analysis. The video AI landscape saw major updates with Kling's O1 model for real-time video editing and its 2.6 model with generated audio. Runway announced its 4.5 video model, while Apple researchers introduced the Starflow V model, which uses a different, less compute-intensive method called normalizing flows. A social phenomenon dubbed "bird game three," consisting of AI-generated clips about a fictional bird-fighting video game, emerged from Sora 2. In robotics, a Chinese company named Engine AI unveiled a humanoid robot called the T-800. A surprising claim is that Google's Gemini 3 Pro proved significant advances in AI pre-training are still possible, countering a prevailing industry belief that scaling was nearly complete. The hosts find OpenAI's "code red" a significant admission, with Kevin Pereira noting it's the "AI equivalent of 'my girlfriend is a model in Canada.'" A memorable technical flaw occurred when the Kling 2.6 model bizarrely mispronounced the famous "You can't handle the truth!" line as "You can't handle he trot." The Apple Starflow V model is highlighted for its potential to enable advanced video generation directly on phones, bypassing cloud compute costs. The "bird game three" meme cycle exemplifies Sora 2's unique, culturally-aware output, generating believable faux-gaming content like lobby chatter and midnight launch reports. The tone is fast-paced, conversational, and laced with humorous banter between the two co-hosts. The style blends breaking news analysis with hands-on testing of new AI tools, often sharing the hosts' personal experiences and failed experiments. Listeners interested in the business rivalry between OpenAI and Google, or those who enjoy weekly deep dives into the latest AI model releases and capabilities, will love this episode. Anyone seeking a highly structured, purely technical lecture or who is fatigued by the rapid-fire "news of the week" format might want to skip it.

As heard by us

A code-red AI race runs into garbled courtroom dialogue and a sharp question about signal in synthetic video.

OpenAI's code red over Google's Gemini 3 Pro gives this episode its live-wire question: has the model race settled, or could rumored OpenAI releases scramble the leaderboard again?

Read the full review in PlayNext →

Why you'd press play

You can hear a famous courtroom monologue get mangled by Kling, then follow the panic back to the AI leaderboard.

Read the full recommendation in PlayNext →

Listen to the show on

Apple Podcasts Spotify

Play clips from this episode More from AI For Humans: Weekly AI News, Tools & Trends