In partnership with
Today on AI For Humans:
GPT-5.4 is Very Useful
Fly Brains Have Been Uploaded??
Plus, vibecode your own game!
Welcome to the AI For Humans newsletter!
Have you been making stuff this week?
I’ve been “making” a lot of “stuff” but, other than the feature-packed AI For Humans show website, I haven’t vibecoded anything I’m willing to put out there yet.
But now that we’ve got yet another state-of-the-art AI model, I might be able to move some of those projects forward more quickly.
OpenAI has released GPT-5.4 Thinking and, yes, it’s VERY good.
This is, from all accounts, a major launch for OpenAI and a significant step up in actual, functional work it can do.
The benchmarks are good but how does it do in real world tests?
Let’s get into it!
Please support AI For Humans by learning about our sponsors below:
Want to get the most out of ChatGPT?
ChatGPT is a superpower if you know how to use it correctly.
Discover how HubSpot's guide to AI can elevate both your productivity and creativity to get more things done.
Learn to automate tasks, enhance decision-making, and foster innovation with the power of AI.
How GPT-5.4 Has Improved…
New AI models are always a bit tricky to test. But I can tell you from personal experience... this one just feels smarter.
They’ve done some significant tuning under the hood for the chat experience (this is, after all, where most people interact with it) and the responses are better, more detailed and, according to OpenAI, have significantly fewer hallucinations.
I asked it the following:
Give me all the details you can find about the AI For Humans newsletter (pretend you don't know me) and give it a rating based on what it could do better for readers.And very quickly it spun up the web browser and returned me a multi-page report on the overall messaging of this particular newsletter.
There’s also some solid criticism in the longer post.
But the bigger deal is the new model’s huge leap in CUA (computer use agent) ability.
In layman's terms, this is about how well an AI agent can operate a computer the way a normal human would.
This three minute video from OpenAI does a remarkably good job of showing off the new model’s abilities to get things done as a human might…
Of course, it greatly depends on what you do with AI to see the improvements in your work.
But it’s very interesting to see OpenAI COO Brad Lightcap shout out the work they’re doing on finance models (specifically within Excel).
Coding Improvements To Compete With Anthropic
One of the biggest areas the model has improved is in coding skill.
OpenAI has been desperate to shore up this aspect of their models, especially after losing significant market share to Anthropic.
Love this newsletter? Forward it to one curious friend. They can join in one click.
While OpenAI has shared some of the best examples of what people have built so far on their website, one of the easiest ways to see this is the sorts of projects people are sharing online:
And maybe my favorite, personal use case I’ve seen…
So… where do we go from here?
The good news is that unlike some of the other model releases lately, it feels like you, the human, will be able to directly see improvements in your day-to-day work with GPT 5.4 Thinking.
The slightly trickier news to digest is that it’s yet another example of how these models are speeding up improvements and it’s near impossible to know where we’ll be at the end of this year.
Looking back at all of the above, I realize these examples lean into what I'm excited about and I’d love to hear more about what you’re excited about as well.
Just keep learning, stay up to date and keep making stuff.
-Gavin
This week’s AI For Humans: GPT-5.4 Lands & All The Anthropic Drama👇
3 Things To Know About AI Today
Someone Uploaded A Fruit Fly’s Brain. Yes, Really.
There are normal days in our lives and then there are the days where it seems like we’re living in the early stages of a sci-fi novel.
There’s been more of the latter than the former lately but… I mean, we’re actually doing brain uploads now?
Yes, Eon Systems just showed a full brain emulation of a fruit fly.
This is not an animation. It's not an AI trained to mimic a fly. It's a copy of a biological brain running in simulation and making a body move.
All 125,000 neurons. 50 million synaptic connections. Mapped from actual electron microscopy data. And it's not just sitting there doing nothing.
It's controlling a body. Walking. Turning. Responding to what it sees. Multiple behaviors, all driven by the emulated brain's wring.
I mean… WHAT?!?
Eon's already working toward a mouse brain next (70 million neurons, 560x more complex) and eventually... human brains.
Make sure you read the full X article linked below the tweet above, it’s worth it.
If you’ve seen Pantheon on Netflix, you know where this story goes. If you haven't... maybe start there.
Seth McFarlane’s Very Reasonable Take on AI Use in Hollywood
There is still a huge ongoing fight in Hollywood around AI (see our coverage of Seedance 2.0 for more on that) but there are people with reasonable takes on AI when it makes sense, even with all the hassles that come along with it.
In the same week that Ben Affleck’s secret AI company sold to Netflix, Family Guy and Ted creator Seth McFarlane did press about his recent AI acting turn as Bill Clinton.
Variety caught up with Seth on the Ted red carpet and asked him about this and I’d genuinely love to hear people tell me how they could disagree with this take.
Software Engineers Are Going Through It
With the advent of the latest AI models (Opus 4.6 & GPT-5.4), it’s beyond clear that they are getting much better at writing software code.
And the people whose job it used to be to write that code are starting to think pretty deeply about their roles going forward.
I’ve seen a few of these sorts of posts or videos lately but this ten minute video from one of those developers, Mo Bitar, is worth your time.
After watching Mo’s video, I keep thinking about the idea of what software is and trying to get my head around how it might compare to something more creative like films or books.
One thing I do think that’s different about software is the fact that it’s designed to do something first and foremost, where “art” is designed to make you feel something.
This is not to say that software code itself can’t also make you feel something but… maybe it will makeing stuff that envokes feelings going forward that powers human beings to fulfilling careers and lives worth living.
We 💛 This: Step-By-Step Vibecoding Complex Games With Codex & Claude Code
It’s no secret, we love ourselves a good video game here.
I made the mistake of purchasing the early access build of Slay The Spire 2 and have already lost too many hours of AI building time.
But… since these models have gotten so much better at coding, you too can now make your own video game and it’s not as hard as you might think.
The video below from AI dev Chong-U is a great step-by-step look at how to make a Final Fantasy Tatics-like game.
Even if this type of game isn’t what you want to make, I’ve found that following one of these tutorials really will give you a good sense of what the tools can do and how to bring them into whatever you want to make.
Are you a creative or brand looking to go deeper with AI?
Join our community of collaborative creators on the AI4H Discord
Get exclusive access to all things AI4H on our Patreon
If you’re an org, consider booking Kevin & Gavin for your next event!