GPT-5 is here. So are the new problems
AI will save us hours. Then waste them differently.
OpenAI just released GPT-5. AI will solve old problems; it'll also create new ones - like the "chart screwup" that surfaced during its launch livestream.
We know all about what AI can do. So, this UnfilteredFriday, let's talk about the brand new problems that AI will create.
'Unintentional' chart crime
In a livestream early this morning, OpenAI launched GPT-5 with much aplomb. It's impressive, though I'll skip how good it is for now.
Anyway, it turns out that some charts showing GPT-5's performance were off, with equally-sized bars for different scores, or longer bars for smaller scores.
As reported by The Verge, it was bad enough that Sam Altman called it "a mega screwup." A marketing staffer also apologised for the "unintentional chart crime."
Did OpenAI use AI to create the charts, only to have it hallucinate? I don't know. But here are some real-world problems that AI is creating for editors and copywriters right now.
Too much to check
Creating content used to require effort and expertise. That natural quality control is gone. Now, those responsible for quality are buried.
The flood looks like this:
- Verbose drafts sent over that need complete rewrites.
- Teams pushing content live without quality checks.
- 50-page white papers that should be 10.
To be clear, AI does uplift general writing standards. But official content that needs to be accurate, concise, and on-brand? That's the challenge.
The devil's in the last 5%
I've previously related a project where I used carefully prompted, project-specific instructions to extract information from dozens of files.
It looked perfect until I found the hallucinations. Just 1-2 per batch, but enough to force me to manually verify everything. So much for time savings.
So the devil's in the last 5%. And that's the time killer. You can't trust the 95% that's correct because you don't know which outputs are in the 5% that aren't.
The only viable solution?
Check everything, turning a 10x productivity gain into maybe 1.5x. Unless, of course, your team member decides to skip the checks. That's another can of worms entirely.
What's your experience? What new problems is AI creating for you?
