Chat GPT5 is here

AI
1

and they emphasize the improved coding capabilities
any progress with xojo code ?

2

What do your tests show?

3

I made a quick comparison. I asked both models for code to “Calculate proper text bounding box that includes descenders by using TextHeight instead of FontAscent.” Both models make the same errors, but these are usually easy to identify. What I noticed is that v5 was much faster in response compared to pre-v5.

Interestingly, ChatGPT gave an additional advice:

If you want the truly tight bounding box without guesswork, Xojo has a trick:
Instead of relying solely on Graphics properties, you can measure a string with Graphics.TextHeight after drawing it to an offscreen picture, which gives you exact pixel bounds (including descenders and without the font’s internal leading). That’s a bit more code, but it removes the “mystery padding.”

2 Likes
4

! cool, pretty precise advise !

5

i wrote somewhere that OpenAi is overrated,
some people think LLM may reach a a ceiling
Maybe sam Altman over sold this, just read :

GPT-5 Broke Everything That Was Working

After half a day of testing what was supposed to be the “revolutionary leap forward,” I’m watching my workflows collapse in real-time.

The breakdown is stark:

  • Model selection gutted. You now get 3 options instead of the full range. No more choosing the right tool for the task.

  • Instructions that worked for months? Dead. My production frameworks now confuse it - especially in “thinking mode.” It apologizes, then repeats the exact same errors.

  • It’s literally GPT-4o in disguise. Using GPT-5 for content creation - it has the same choppy structures as 4o. Same flow issues and slop patterns. No way to use 4.1 or 4.5, my preferred tools.

  • Network crashes. Three “connection lost” errors in 12 hours. Not my internet - their new internal ‘routing’ system.

  • Tools fail, then it gaslights you. Python execution breaks, then it either lies about why or claims it “isn’t capable” of tasks it just attempted (even though it is).

  • Even image generation seems to have got worse.

The trillion-dollar question: Are we watching the rocket run out of fuel while everyone’s still selling tickets to Mars?

Sure, it is early days for GPT-5 and perhaps it will improve rapidly.

But, for the first time, I’m wondering if the skeptics were right about the wall.