The skinny on what's happening in AI - straight from the previous live session:

Highlights

🔧 Sam Altman's Mea Culpa: "We Totally Screwed Up GPT-5 Launch"

Discussed at [1:43]

Sam Altman held a dinner with reporters and admitted OpenAI "totally screwed up some things on the rollout" of GPT-5. Beyond the apology, he revealed plans to spend "trillions" on data centres to scale globally, positioning OpenAI as infrastructure rather than just software.

Kyle's take: This isn't just about admitting mistakes - Altman's pushing past and positioning himself for massive government funding. By drumming up fears about China's AI capabilities (which he also did this week), he's making the case that OpenAI needs strategic government support to compete. Trillions of dollars moves us into a different realm of investment entirely.

Source: Fortune magazine

🎯 AI Recruiters Beat Humans in Massive 70,000-Person Study

Discussed at [27:32]

University of Chicago researchers tested AI voice agents against human recruiters with 70,000 job applicants in the Philippines. Results were brutal for humans: AI led to 12% more job offers, 18% more people actually starting jobs, and 17% better retention after 30 days. When given the choice, 78% of applicants selected the AI recruiter. The AI also extracted more relevant hiring information during interviews. Across the board win for AI.

Kyle's take: This is catastrophic news for recruiters and more broadly anyone in agency businesses - estate agents, I'm looking at you.

If you sit between two parties in a transaction, AI is coming for your lunch. The matching process is exactly what generative AI excels at - pattern recognition at scale. What's worse is people actually preferred talking to the AI. I imagine that’ll only increase over time too.

Source: University of Chicago research study

🎮 90% of Game Developers Secretly Using AI

Discussed at [42:34]

Google's study of 615 game developers across five countries found 90% are already using AI in their workflows, despite massive public backlash against AI in the gaming community.

95% say it reduces repetitive tasks, whilst 36% use it for creative work like dialogue and animation. The disconnect is stark - gamers rage when they discover AI usage, but developers are quietly adopting it everywhere…just…quietly!

Kyle's take: We're in this bizarre situation where using AI has become shameful, so people do it secretly. This mirrors what I see across creative industries - everyone's using it but keeping quiet to avoid the mob.

Honestly the 90% using it for workflow stuff makes perfect sense - get rid of the mundane so you can focus on actual creativity. And that 36% doing creative work with AI will only grow as the tools improve.

Source: Google Cloud research study

Member Question: "How can I make sure an AI replicates my writing style fully? My custom GPT is struggling despite analysing PDFs."

Kyle's response: You need way more data than one PDF. I use a system with 400-500,000 words of my writing plus hours of transcripts. Here's the full process:

Step 1: Gather massive amounts of data. You need hundreds of writing samples, not just one PDF. If you don't have enough, start creating more content specifically for this purpose.

Step 2: Use a larger context window. Custom GPTs are limited to ~128,000 tokens. A little more with GPT-5 but still limited. Switch to Claude Projects or Gemini which can handle much more text.

Step 3: Create a detailed style guide. Feed all your writing into a large context model like Gemini and ask it to analyse your style patterns, common phrases, tone, and create specific guidelines for another AI to follow.

Step 4: Give continuous feedback. Treat it like marking a student's essay. When it outputs something, annotate exactly what's wrong - "I don't use this many exclamation points," "I use British spelling not American," "This phrase isn't how I'd say it." Feed this back to refine the system.

Step 5: Consider moving beyond Custom GPTs. If you need something more sophisticated, build a proper application using tools like Lovable rather than relying on ChatGPT's limited custom options. This will allow you to put in place a proper database + system for replication.

The key is massive amounts of data plus iterative feedback. It's still hard work, but that's what gets results.

This question was discussed at [17:20] during the live session.

Want the full unfiltered discussion? Join me tomorrow for the daily AI news live stream where we dig into the stories and you can ask questions directly.

Streaming on YouTube and TikTok (follow and turn on notifications for Live Notification).

Audio Podcast on iTunes and Spotify.