GPT-5 is here... Can it win back programmers?

Tech for Change
4 followers
#ai #gpt5 #coding
Latest updates and insights on AI and technology – simple, clear, and useful for everyone.
Yesterday was the last day of the rest of our lives.
Yesterday was the day Sam Altman asked his ego this three-word question.
What have we done?
Because yesterday was the day the human intelligence monopoly officially ended with the long-awaited release of OpenAI's GPT-5,
the first AI model to outperform biological humans on the SimpleBench benchmark.
Meanwhile, it's quickly overtaking virtually every leaderboard on LM Marina,
which means every programmer that still has a job will be getting a layoff notice shortly.
Or at least that's what the hype engineers at OpenAI want you to think.
Not everybody is falling for it, though.
That SimpleBench score was just a rumor.
In fact, GPT-5 is actually in fifth place.
GPT-5 failed to beat Grok on the ARCAGI benchmark,
an important benchmark that was conveniently left out of yesterday's announcement.
It also underwhelmed the betting markets, and OpenAI is no longer the favorite to have the best model of 2025.
Worst of all, people found numerous problems with OpenAI's own charts,
which really shouldn't happen when you have PhD-level intelligence in your pocket.
In today's video, we'll find out if GPT-5 is just another overhyped incremental upgrade
or a true game-changer, edging us closer to artificial superintelligence.
It is August 8, 2025, and you're watching The Code Report.
In the past, GPT got better as it got bigger.
It was trained on more data that would activate more parameters when generating text.
But those days are essentially over.
The thing that makes GPT-5 special is not that it's a bigger, smarter model,
but rather it unifies multiple models like fast reasoning, routing, etc.
to pick the right tool per task without the end user needing to think about it.
In many ways, GPT-5 feels more like a consolidation and cost reduction effort
after they spent the last year panic-shipping a bunch of stupidly named models to justify the $200 pro plan.
Speaking of cost, GPT-5 is priced at only $10 per million output tokens.
That's a great deal compared to Claude Opus 4.1,
which will burn a hole in your wallet at $75 per million output tokens.
Sam Altman says GPT-5 is like having multiple PhD-level experts in your pocket,
but one of the funniest parts of their announcement is that their benchmark graphics
were presented with a y-axis that doesn't make any sense.
And there can only be two possible explanations for this error.
One, they vibe-charted with GPT-5 and it doesn't actually have PhD-level intelligence.
Or two, they were trying to be intentionally misleading.
What's even better is that GPT-5 is supposed to have lower deception rates,
but then someone or something tried to deceive us with the y-axis on the deception benchmark.
That's kind of embarrassing for a $500 billion intelligence company.
But if you're a programmer, the biggest question is whether or not
GPT-5 can code a Svelte 5 app with runes.
Many models have come close, but none have succeeded.
I put GPT-5 to the test and was stunned to see that
it generated beautiful-looking Svelte code very quickly,
far faster than any other reasoning model.
At first I thought I was working with a PhD-level Svelte developer,
but I was soon disappointed after running the code and getting a 500 error as the UI.
What's troubling is that GPT-5 used the right syntax,
but tried to use a rune in the template, which is not allowed.
GPT-5 is supposed to have fewer hallucinations,
but it hallucinated its own rules for using runes.
However, it did redeem itself when I asked if it knew what was wrong with its code,
at which point it figured it out and fixed it.
The end result was a functional app with a very nice-looking UI.
I also had it build a flight simulator game with 3JS, which turned out pretty bad.
But the tall guy from Cursor said it's the smartest model they've ever used.
Unfortunately though, I don't think GPT-5 is going to take my job or euthanize me anytime soon.
It's clear at this point that the real power comes when you combine these new AI tools
with existing technologies you already know and love.
Which is why you need to check out Dreamflow, the sponsor of today's video.
It's a full-stack AI development environment created by the team at Flutterflow
that lets you build, run, and deploy cross-platform apps all from your browser.
You can spin up a new project with a prompt and it'll run directly in Dreamflow,
where you'll get full file system access along with the ability to preview and edit
all of your pages and components, either by using their visual editor
or manually changing the code yourself.
It integrates seamlessly with Firebase and Superbase,
and also lets you deploy to the web or the app stores with a single click.
If someone had shown me this when I was first getting into Flutter back in 2018,
my brain would have exploded.
But you can try out Dreamflow for free with the link below.
This has been the Code Report.
Thanks for watching, and I'll see you in the next videos.
