GPT-4o (“o” for “omni”) is a step towards much more natural human-computer interaction—it accepts as input any combination of text, audio, and image and generates any combination of text, audio, and image outputs. It can respond to audio inputs in as little as 232 milliseconds, with an average of 320 milliseconds, which is similar to human response time(opens in a new window) in a conversation. It matches GPT-4 Turbo performance on text in English and code, with significant improvement on text in non-English languages, while also being much faster and 50% cheaper in the API. GPT-4o is especially better at vision and audio understanding compared to existing models.
Prior to GPT-4o, you could use Voice Mode to talk to ChatGPT with latencies of 2.8 seconds (GPT-3.5) and 5.4 seconds (GPT-4) on average. To achieve this, Voice Mode is a pipeline of three separate models: one simple model transcribes audio to text, GPT-3.5 or GPT-4 takes in text and outputs text, and a third simple model converts that text back to audio. This process means that the main source of intelligence, GPT-4, loses a lot of information—it can’t directly observe tone, multiple speakers, or background noises, and it can’t output laughter, singing, or express emotion.
GPT-4o’s text and image capabilities are starting to roll out today in ChatGPT. We are making GPT-4o available in the free tier, and to Plus users with up to 5x higher message limits. We’ll roll out a new version of Voice Mode with GPT-4o in alpha within ChatGPT Plus in the coming weeks.
Maybe this is wishful thinking but this, at first glance, seems like a sign that we’re already entering the LLM plateau. Like when they got the point with phones that each new version is just more cameras, smoother UI, and harder glass.
It’s an end-user plateau for the moment, but there are still tons of things going on underneath the hood. From the outside it may not look like things are moving, but we’ve gone from Model-T to Chevy Bel Air fairly quickly, and while the difference is huge, the engineers are still trying to get us to Bugatti Veyron level. Until then, we are going to have a long “80 and 90s” period of sameness.
18 months ago, chatgpt didn’t exist. GPT3.5 wasn’t publicly available.
At that same point 18 months ago, iPhone 14 was available. Now we have the iPhone 15.
People are used to LLMs/AI developing much faster, but you really have to keep in perspective how different this tech was 18 months ago. Comparing LLM and smartphone plateaus is just silly at the moment.
Yes they’ve been refining the GPT4 model for about a year now, but we’ve also got major competitors in the space that didn’t exist 12 months ago. We got multimodality that didn’t exist 12 months ago. Sora is mind bogglingly realistic; didn’t exist 12 months ago.
GPT5 is just a few months away. If 4->5 is anything like 3->4, my career as a programmer will be over in the next 5 years. GPT4 already consistently outperforms college students that I help, and can often match junior developers in terms of reliability (though with far more confidence, which is problematic obviously). I don’t think people realize how big of a deal that is.
There’s a basic problem with replacing human experts with AI. Where will they get their info from with no one to scrape? Other AI generated content?
They can’t learn anything and are just “standing on the shoulders of giants”. These companies will fire their software developers, just to hire them back as AI trainers.
“they can’t learn anything” is too reductive. Try feeding GPT4 a language specification for a language that didn’t exist at the time of its training, and then tell it to program in that language given a library that you give it.
It won’t do well, but neither would a junior developer in raw vim/nano without compiler/linter feedback. It will roughly construct something that looks like that new language you fed it that it wasn’t trained on. This is something that in theory LLMs can do well, so GPT5/6/etc. will do better, perhaps as well as any professional human programmer.
Their context windows have increased many times over. We’re no longer operating in the 4/8k range, but instead 128k->1024k range. That’s enough context to, from the perspective of an observer, learn an entirely new language, framework, and then write something almost usable in it. And 2024 isn’t the end for context window size.
With the right tools (e.g input compiler errors and have the LLM reflect on how to fix said compiler errors), you’d get even more reliability, with just modern day LLMs. Get something more reliable, and effectively it’ll do what we can do by learning.
So much work in programming isn’t novel. You’re not making something really new, but instead piecing together work other people did. Even when you make an entirely new library, it’s using a language someone else wrote, libraries other people wrote, in an editor someone else wrote, on an O.S someone else wrote. We’re all standing on the shoulders of giants.
It’s not like there’s a shortage of human generated content. And the content that has already been generated isn’t going anywhere. It will be available effectively forever.
So? If you ask an LLM a question, you often get a very useful response. That’s ultimately all that matters.
Why would you wish for technology to stop improving?
It’s not about wanting it to stop, it’s about getting it to maturity so we can get out of this phase of buzz words, misleading marketing, and then we can find out what the tech can actually be useful for.
This is not just some technology but something that may lead to a true artificial intelligence with all the far reaching consequences. It’s like nukes and manhattan project.
If we had feasible way to prevent birth of true AI I am sure most would want to stop just before it becomes sentient and spreads to every network connected device in the world.
Definitely not.
If anything, them making this version available for free to everyone indicates that there is a big jump coming sooner than later.
Also, what’s going on behind the performance boost with Claude 3 and now GPT-4o on leaderboards in parallel with personas should not be underestimated.
Edit: After enough of a chance to look more into the details, holy shit we are unprepared for what’s around the corner. What this approach even means for things like recent trends in synthetic data is mind blowing.
They are making this free because they desperately need the new data formats. This is so cool.
Worth listening to a podcast by Ed Zitron who covers this exact same thing.
I work in analytics consulting and aside from some relatively straight forward text classification and parsing there really are so few instances where AI is useful. He actually made the case that AI is useful to people selling AI. But these chatbots are mostly useless bullshit.
https://www.wheresyoured.at/peakai/