Coder of 20+ years teaching non-technical people how to build their own software business in 30 days with AI. No devs or code required.
don't use GPT-5.1 (here's why)
|
Hey Reader, You can also listen to this episode here. So, GPT-5.1 was released last week and... it feels completely broken. I'm not being dramatic. I spent 6 hours testing it, and something's off. And judging by the 8,200 people who saw my tweet about it... I'm not alone. The Model ExplosionCursor dropped 15 (!) GPT-5.1 models at once. What? GPT-5.1 Fast. GPT-5.1 High Fast. GPT-5.1 Codex Mini. GPT-5.1 Codex Mini High... It's really overwhelming. I stuck with GPT-5.1 Fast and High Fast because I value speed over money. I'd rather pay double and get the job done fast. Here's what I found... GPT-5.1 Is WeirdThe model behaves... differently. Think of it like this: Imagine you hired a developer who keeps asking you "Are you SURE you want me to change that file?" every 2 minutes. That's GPT-5.1. It tells you what it wants to do. Then doesn't do it. Sometimes it straight up refuses, saying "that would be too much work." I'm not making this up. I posted a quick tweet not thinking much about it, and the responses flooded in: "Yes 100% it's way lazier"
Look, I'm gonna be brutally honest with you: Right now, I've got exactly ONE client in the just launched Inner Circle. That means this Thursday's 90-minute call is basically a private session. Even when 2-5 people join, you're still getting 18-45 minutes of my undivided attention per call every two weeks. But this window closes FAST. Once I hit 10 members, that same 90 minute call gets you 9 minutes. The people who join in the next 72 hours get founder-level access at a fraction of what it'll be worth when this group fills. I can't offer this ratio again. The math won't let me.
Join before Thursday (with lifetime access)
What You Should DoI'd stick with GPT-5.0 for now. Chances are that based on the overwhelming feedback, OpenAI will probably tune it. There's zero shame in using a model that actually works. GPT-5.0 is a solid, proven model that doesn't ask for permission 47 times before doing the work. I'll keep testing GPT-5.1 over the next week or two and let you know if it gets better. Talk soon, P.S. If you've tried GPT-5.1, reply and tell me if I'm crazy or if you felt the same weirdness. I read every response. I read every reply and love to hear from you! |
Robin Ebers
Coder of 20+ years teaching non-technical people how to build their own software business in 30 days with AI. No devs or code required.
Reader, Got a bunch of replies from yesterday's email! A few of you already bought, and I just sent them an email explaining what they're actually getting. Let's just say... it's more than a workshop. I recorded a quick 2-minute video explaining why this first group is different. There's only 5 spots left Grab one of the last seats Once they're gone, I'm closing this and building the app. If you receive a refund, I'm sorry in advance - the limited spots are real! See you there!Rob
Hey Reader, Everyone wants to build a mobile app, but nobody shows you how. On Dec 10th, I'm building an AI-powered camera app for iPhone from scratch - and I'm recording the whole thing. What You Get Full "build with me" recording (watch at your own pace) Live Q&A session where I answer your questions and troubleshoot Step-by-step checklist to get it live in Apple AppStore Lifetime access to the AI Builder's Circle community But here's the thing... This is NOT for complete beginners. You...
Hey Reader, Google dropped Gemini 3 Pro this week. Everyone lost their minds. Then 24 hours later, OpenAI was like: "F*ck you Google! Here's our new model and it's even better!" So here's what happened in these 24 hours of model insanity. And more importantly: If you should even care. What's Really Going On I've watched this play out for a year now. Every single time a new SOTA (state of the art) model drops, the same thing happens: Before release: "This is going to change EVERYTHING!" After...