Has AI Peaked?
Last week OpenAI released a brand new model “GPT-4o-mini” in what they’re calling a lightweight, faster and wildly cheaper version of GPT-4o.
This is the first time anyone is rolling the perception of a contracted AI model rather thank larger, grander, and more capable, and it begs the question.. have we peaked with AI?
GPT-4o mini feels like a shift in AI development.
Rather than pursuing ever-larger models, OpenAI has opted for a more refined approach. This new model achieves an 82% score on the MMLU benchmark and features a 128,000-token context window, all while offering multi-modal capabilities that include text and vision processing.
But most notably, it does so at a fraction of the cost of its larger counterparts. It is priced at 15 cents per million input tokens and 60 cents per million output tokens, which is more than 60% cheaper than GPT-3.5 Turbo. These are massive savings especially for guys like me running API calls 24/7 to power our projects.
Sound too good to be true?
This pivot towards efficiency over sheer power suggests that the AI landscape is evolving & we’re entering a phase where the ability to deliver advanced capabilities in a more accessible package is becoming just as crucial as pushing the boundaries of performance.
In fact just recently Mira Murati sat down for an interview and publicly stated they’ve just about released the most exciting work they have.
“The most recent thing we did was the release of GPT-4o, which is our omni model. We made it accessible to everyone for free,” she had stated, then added, “Inside the labs, we have these capable models and they’re not that far ahead from what the public has access to for free.”
Viewing the RewardBench, GPT-4o-mini is above Claude 3 sonnet, and Llama 3 70b, but also below Gemma 2 27b and just about the same as all the rest.
Honestly, it’s now feeling like we’ve hit a saturation point.
I’m feeling we’ve entered a new phase, not a peak.
Rather than signaling that we’ve reached the pinnacle capabilities (which we know we have not), this drop tells me we’re entering a new phase of development.
At least for OpenAI, it’s clear the focus now is optimization, practical applications, and accessibility. It’s about refining and honing the powerful tech that is growing in popularity not necessarily convincing us there’s new ways to use it.
We’re now balancing capability with cost-effectiveness and real-world applicability.
No matter what, I’m still here for it. And I’ll take the discounts. I’m still testing all my AI applications on GPT-4o-mini before making the full switch but it’s not every day a service you use offers you a ~60% discount just because it’s a Thursday.
If there was ever a time to lean in – this is your sign – it’s now.
Generative Tools Bring Statues to Life
Ever stopped and stared at one of these statues wondering what the person really might look like? AI tools have an incredible way of bringing the impossible to life.
I found this great video walkthrough 🔗 of how a Sr. Environment Artist at Bethesda Games uses some of these mixture of tools to bring still life… back to life. Enjoy!
ThisText AI (Explain This) – Project Update
If you missed my email last week, go read about my latest project launch right here! 🔗
I love launching new projects to the world and bringing people along for the ride on the good and the bad. It was great seeing such an influx of new users and testers and even already have some that have switched to ‘unlimited’ which is the best!
So grateful for you, and for this community of AI enthusiasts. If you have not tried my newest way of bringing AI to your messages app (over SMS):
Grab your phone.
If you’re already on your phone, tap this link 🔗 or:
Text: explain this
to: 866-733-2202
And due to the sheer volume of replies and support from last weeks, it’s clear so many of you love seeing how I’m using it as I build personally. Over the next few months I’ll keep sharing the latest projects I’m working on (I have two more big ones I can’t wait to tell you about).
I’ve started deep diving on this project inside my Slack community and this will be a fun one to share with everyone there.
And I’m opening my slack community to the waitlist 🔗 this Thursday, July 25th for a hand-full of new members.
If you’d like to join me there, have direct access to message me, get a copy of the insane AI Clone I made of myself to help you with your business, AI & tech questions AND join me every single month for my private office hours sessions via zoom, this is the spot.