GPT-5 is here. It’s not a giant leap in intelligence. But for 98% of users, it’s still the best ChatGPT yet. What does this tell us about the future of AI? In today's blog, I dig in.

Remember the GPT legacy? • GPT-2 (2019): First coherent paragraphs • GPT-3: Actually useful, could follow instructions • GPT-4: Aced the bar exam, became a coding copilot • o1/o3: New paradigm with "thinking before answering" Each was a paradigm shift. Until now.

OpenAI chased the "GPT-5" name for years. Rumor has it GPT-4.5, o1, and o3 were all candidates but fell short of the dream. They finally gave up chasing the big intelligence leap and just shipped what they had. The pressure was too great.

Instead, we've moved from ONE way to improve models (make them bigger) to FOUR: 1. Train more 2. Post-training reinforcement learning 3. Inference-time compute 4. Better scaffolding (tools, web search, etc) GPT-5 is actually two models glued together

What GPT-5 rules out: The "lightning fast AI progress in 2025" hypothesis is dead. If o3's December demo meant reasoning models could take off rapidly, GPT-5 should have leapt. It didn't. My median AGI arrival date stays at 2033.

But there are still many unanswered questions about AI progress: - Is old-fashioned training scaling dead? - How far can post-training RL take us? - Does this work for hard-to-verify domains like writing? - When do we get real AI agents that can book flights autonomously?

But here's the plot twist: 98% of ChatGPT users use it for FREE. And GPT-5 delivers for them: ✓ Faster ✓ Cheaper ✓ Less hallucinations ✓ Less deception ✓ Better writing ✓ No more confusing "4o vs o4" choice, model picker solves it for you

The money tells the real story: OpenAI valuation: - End 2024: $157B - Now: Potentially $500B Who cares about benchmarks when you're printing money?

Speed, cost, and reliability are bigger bottlenecks for enterprise AI than raw intelligence. Models are already smart enough for most use cases. GPT-5 delivers where it actually matters for business adoption.

The verdict: Intelligence progress continues on predictable trendlines, but GPT-5 tackles the real bottlenecks: cost, latency, reliability. AI elites keep waiting for fireworks. Meanwhile, for 98% of users, GPT-5 is simply the best ChatGPT yet.

Want deeper analysis on GPT-5, AI progress, and where the next big leap might come from? Check out the blog and subscribe!