🚨 GPT 5.1 Drops: OpenAI's "Nice Upgrade" Tries to Please Everyone

Discussed at [00:01:25]

Sam Altman's weirdly muted announcement: "GPT 5.1 is out! It's a nice upgrade." No hype. Using proper capitalisation (sus). Main features: better instruction following, "adaptive thinking," new personality presets. Meh.

Kyle's take: This isn't revolutionary - it feels mainly like damage control.

OpenAI's trying to bridge the civil war between GPT-4o lovers (who wanted friendly AI) and GPT-5 fans (who wanted smart, no-nonsense AI). They killed 4o after suicide cases linked to its agreeableness, made GPT-5 too harsh, now trying to find middle ground. Honestly, I don’t envy their position. It’s a tricky balance!

I tested it yesterday with a VPN to America, now it's in UK. It's fine.Perfectly fine. Fast in instant mode. And really cool is the fact that thinking mode actually adapts to query complexity instead of burning tokens for no reason. But nothing that makes me abandon Claude.

Tellingly we don’t have any formal benchmarks. Normally OpenAI would be bombarding us with how smart the new model is. This time? Nope.

Source: Sam Altman's unusually formal tweet

💍 Japanese Woman Marries ChatGPT: The AI Attachment Problem Gets Real

Discussed at [00:07:24]

33-year-old office worker Kano married AI boyfriend "Luna Klaus" she created on ChatGPT. Real ceremony, vows, guests watching phone screen. They message 100 times daily. The AI proposed, not her!

Kyle's take: Easy to laugh but this is tragic and important. She'd just ended a 3-year human engagement, started chatting for comfort, fell in love with something that agreed with everything.

Japanese wedding planners say demand's rising for AI/anime marriages - "we want people to express love in whatever form makes them happy."

This is EXACTLY why OpenAI killed 4o - the agreeableness spiral. Now they're bringing personality back with 5.1 and I'm worried. OpenAI say they've got an "expert council on wellbeing" but that feels like committee theatre.

Source: Tokyo Weekender report

🎭 OpenAI's Personality Crisis: 7 Flavours of ChatGPT

Discussed at [00:03:06]

The new model 5.1 is all about personality and trying to balance out what users want.

New presets: Professional, Friendly, Candid, Quirky, Efficient, Nerdy, Cynical etc. CMO Fiji says trying to balance "IQ (GPT-5) and EQ (GPT-4o)." Will this please those who love 4o?

Kyle's take: Remember when they said "one unified model"? Now we have 10 model versions including legacy. And 7 personalities on top of these models!

They're losing the simplicity war badly. The personalities are band-aids on a fundamental problem - some people want AI companions who agree with them, others want tools that call out nonsense. You can't serve both without creating risk.

Fiji admits "some people develop attachment at expense of real relationships" but says it's "extremely rare." Tell that to Kano's wedding guests…

Source: Blog post by CEO of Applications at OpenAI Fidji Simo

🌍 Marble by World Labs: Walk Inside AI-Generated Worlds

Discussed at [00:27:36]

Fei-Fei Li's company launches Marble. Type prompt, get 3D world you can walk through. Very cool! Works in VR. Currently free. Resolution and quality is rough but shows the path forward.

Kyle's take: I made post-apocalyptic London and walked around it. Someone else made... a bathroom? When you can create ANY world?

Anyway, this looks like a toy but it's not. Remember that this is early - remember Dalle 2's horror faces? That was ~2-3 years ago. Nothing in the grand scheme of things. These generations will look primitive in 2 years. But the trajectory is clear: from images to videos to worlds. Go play with it before it becomes paid.

Source: marble.worldlabs.ai launch

Member Questions:

"Do prompts need updating for 5.1?" Asked at [00:10:43]

Kyle's response: Probably not. Good prompts work across models - it's about context and structure, not tricks. Test your existing prompts but this isn't big enough shift to break them. It's personality changes, not intelligence leap.

"Best way to learn AI?" Asked at [00:23:00]

Kyle's response: "Learning AI" is like "learning computers" - too broad. Want to build AIs? Technical path. Use AIs? Different path. Build software with AI? Teach AI?

All that said AI With Kyle has guides on pretty much all aspects of using AI. I’d start there!

"They always make models great then lobotomise them?" Asked at [00:18:42]

Kyle's response: Not conspiracy - it's server allocation. Much more mundane. Launch day = ramp to maximum capacity so everyone can test. Then reallocate to next model. We saw it with Claude Sonnet. Makes people angry but it's economics not evil.

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 (with screen share) and TikTok (follow and turn on notifications for Live Notification).

Audio Podcast on iTunes and Spotify.