GPT-5: Flop or next step in AI?

On Thursday, August 7, OpenAI released its latest product, GPT-5. What happened next was probably predictable - half the commentators thought it was a flop, half thought it was a solid step forward. Almost no one thought it was a smashing success. So, what happened?

For a start, they genuinely messed up the rollout - some of the plumbing broke, and it acted less intelligently than earlier models. Second, though, it is more important. AI is now at a place where some of the models are too smart for some questions, which is why ChatGPT now has an automated model picker. That is actually what broke. 

Why this matters now is that “AI” was never just one thing; it was always different software products that did other things. But now, within one product, there will be multiple models that can do various things - in fact, what Grok, Claude, and now ChatGPT also do is create AI agents that work in parallel to do increasingly difficult things.

The reality check: AI is continuing to develop like all technologies do, getting better, evolving into niches that serve different needs. What OpenAI did with GPT-5 is start the process of creating an AI that can govern other AI models. In time, we can expect AI that can govern specialist AI tools, like one for accounting, another for document processing, etc. Not to replace humans, but to coordinate data and information. So no, it was not a flop as a product, but pretty floppy as a marketing campaign.

← Back to AI Reality Checks