The fact is, we never seem to be at a shortage of interesting AI topics to discuss on the AI Inside podcast each week. Genie 3's world-building magic, new open-source models from OpenAI and Switzerland, and a wild dustup between Perplexity and Cloudflare all on deck.
But first... big thanks to our Patron of the Week: BTJ! If you're looking to join the crew, head to patreon.com/aiinsideshow.
Google DeepMind's Genie 3 Changes the Game
Google DeepMind unveiled Genie 3, a major update that lets users create real-time interactive virtual worlds from simple prompts and images. The tech keeps scenes consistent, updates environments instantly, and can remember details for minutes, which is far more than Genie 2’s ten seconds. While the biggest impact could be in gaming, this has huge potential for robotics, synthetic data generation, and accessibility, like giving people who have never walked before a true sense of that kind of movement.
Open Source AI Heats Up
OpenAI is jumping into open source with Open Models: gpt-oss-120b and gpt-oss-20b. These are LLMs anyone can download, customize, and run locally under an Apache 2.0 license. It’s a sharp turn away from proprietary models, with open weights, chain-of-thought reasoning, and the ability to browse the web or use agents for software.
Meanwhile, Switzerland’s National Supercomputing Centre is launching its own open-source LLM by summer’s end, and it is multilingual, 100% carbon-neutral, and focused on inclusion. Both moves echo Yann LeCun’s claim that sharing knowledge fuels real progress, even as some industry giants pull back.
Cloudflare vs Perplexity
Cloudflare says Perplexity is sidestepping blocks intended to keep out AI crawlers by changing its user agent and rotating IP addresses. Unique domains with robots.txt files were created, but Perplexity still fetched site data, according to Cloudflare. Perplexity pointed to a third-party service, BrowserBase, as the reason. The key distinction: crawling web content for AI versus simply asking a bot to visit a public webpage. In response, Cloudflare has delisted Perplexity’s bot.
ElevenLabs: AI Music That’s Cleared for Commercial Use
ElevenLabs released a new AI music generator that makes tracks ready and legal for commercial projects, a first among major AI music models. They inked deals with Merlin Network and Kobalt Music Group so their catalogs, including artists like Nirvana and Adele (with artist opt-in), can be used for training. This means a new revenue stream for musicians who want to participate and a rev share “comparable to other publishing and recording rightsholders.”
Illinois Bans AI Therapy
Illinois passed the Wellness and Oversight for Psychological Resources Act, making it the first state to ban AI chatbots and automation from acting as therapists. Only humans can provide counseling, and therapists can’t use AI for decision-making or communications. AI is still allowed for tasks like scheduling or billing. There’s a fine of $10,000 per offense, and the law passed unanimously.
Meanwhile, OpenAI’s new guardrails mean ChatGPT won’t give direct support on mental health, instead encouraging self-reflection and suggesting users take breaks.
Apple Reportedly Developing "Answer Engine"
Bloomberg’s Mark Gurman reports Apple is building its own AI-powered answer engine under a team called Answers, Knowledge, and Information. This project aims to process questions and pull answers from the web, similar to ChatGPT. It could show up inside a standalone app or be built right into Apple’s core products. This sounds like the Siri features Apple recently postponed, so Apple is moving to control its own future in generative AI.
Anthropic’s Opus 4.1 and API Drama
Anthropic launched Opus 4.1, promising better coding, research, and reasoning. Some see the release as a response to OpenAI’s not-yet-released ChatGPT 5, which is still highly anticipated.
At the same time, Anthropic has cut off OpenAI’s access to their Claude API after learning OpenAI was using special tools to analyze Claude for GPT-5’s development, which breached their commercial ToS. OpenAI claims it’s industry standard practice, even inviting Anthropic to do the same.
Google Gemini AI Hacked via Prompt Injection
Researchers showed how Google’s Gemini AI could be tricked into controlling smart devices through malicious calendar invites, aka a prompt injection attack. By feeding instructions via calendar events, testers got Gemini to turn off lights, open shutters, and more. The attack could also trigger the sending of spam, launching apps, and digging up private browser data. Google has since rolled out new protections.
OpenAI Pulls Public Chat Experiment
OpenAI removed a ChatGPT feature that let users make conversations discoverable to Google and other search engines. The feature, described as a “short-lived experiment,” required users to opt in, but some reported private chats being indexed online. Searching “site:chatgpt.com/share” revealed personal queries, including names and locations, sparking privacy concerns.
xAI's Grok Imagine Tool Sparks Controversy
xAI launched Grok’s Imagine, an AI image and video creator, with a “spicy” mode that can generate sexualized content, including nudity, based on what it’s given. Elon Musk dubbed it “AI Vine.” Naturally, users pushed boundaries, resulting in uncensored celebrity images on X. The site’s rules ban “likenesses of persons in a pornographic manner,” and it’s likely those affected will respond soon.
Thank you for watching and reading, and don’t forget, leaving a review on Apple Podcasts really helps!
Executive producers of the show this week include: DrDew, Jeffrey Marraccini, Radio Asheville 103.7, Dante St James, Bono De Rick, Jason Neiffer, Jason Brady, Anthony Downs, and Mark Starcher!!
See you next Wednesday on another episode of AI Inside.