Is GPT-5 Dropping Soon? OpenAI's Secret Project Strawberry | EP 33
Description
In this episode of UNinvested, Sahil explores the buzz around OpenAI's potential new release, possibly GPT-5, and the mysterious "Project Strawberry." He discusses the significance of recent events, including a cryptic tweet from OpenAI's CEO, Sam Altman, and the sudden appearance of an "anonymous chatbot" on a popular AI testing platform. Sahil shares his thoughts on what these developments might mean for the future of AI and why this could be a major breakthrough.
What we explore:
The implications of Sam Altman's strawberry tweet and its connection to "Project Strawberry"
Background on OpenAI's secretive "Project Strawberry" and its possible link to GPT-5
The emergence of an anonymous chatbot on LMSYS Chatbot Arena and its potential significance
Speculations on Project Strawberry's advanced capabilities, including autonomous reasoning and decision-making
OpenAI's history of testing new AI models under anonymous names before public release
The concept of "Reasoners" in OpenAI's roadmap toward achieving superintelligence
Where to find Uninvested:
Website: https://www.uninvested.org/
In this episode, we cover:
[00:00] Introduction and disclaimer about GPT-5 speculations
[00:22] Discussion on Sam Altman's cryptic tweet and its connection to Project Strawberry
[00:51] Overview of the anonymous chatbot on LMSYS Chatbot Arena
[01:16] Explanation of Project Strawberry and its potential link to GPT-5
[01:55] Speculations on Project Strawberry's advanced reasoning capabilities and its implications for AI development
[02:30] OpenAI's strategy for testing new models anonymously on public platforms
[03:00] Analysis of the anonymous chatbot's performance on the leaderboard and its potential significance
[04:00] Sahil's closing thoughts on the future of AI and the impact of Sam Altman's strategy
[04:18] Final remarks on the importance of using AI and a personal message to listeners
[04:31] Sahil expresses gratitude to supporters and discusses the passion behind UNinvested
[04:37] Closing statement with details on the next episode
I’m going to be totally honest, I have no idea if GPT-5 is coming out or not. So don’t get mad because I told you in the first 10 seconds of a 5-minute podcast. But I will lay out all the facts of why I think something very big is going to be announced soon by OpenAI, the creators of ChatGPT.
[Intro Music]
This week in tech, two very weird things happened simultaneously. One, is that Sam Altman, CEO of OpenAI, posted a picture of strawberries. The second is that on LMSYS Chatbot Arena, a very popular, independent website that lets people test two different AI models side-by-side (think OpenAI’s ChatGPT right next to Anthropic’s Claude), released an “anonymous-chatbot.” Essentially, AI’s compete on the website to come up with the best answer.
So for a little bit of background, the picture of strawberries is significant because OpenAI has a secret initiative called Project Strawberry. It is a tightly guarded project with the public having little information about it. Even employees at OpenAI are unaware of all the details.
Apparently, project strawberry could be the code name for GPT-5, a new AI model that will blow every AI model out of the water or it’s at least a serious step towards it’s development.
Project Strawberry is described as having advanced reasoning abilities and can do “deep” research autonomously. So in simpler words, it is an AI model that is attempting to be more human-like, which I know is the ultimate goal of AI. The AI won’t just provide answers to questions via chatbots, but instead will be able to understand problems and take action.
Let’s take meal planning and grocery shopping as an example. Right now, AI could give recipes and shopping lists, but not much more. Project Strawberry could potentially take in all the information about you and your family’s nutritional needs and preferences and then create a long-term meal plan, make shopping lists that compare prices from different stores, and even suggest substitutions of ingredients to meet cost or diet restrictions. You get the point, it’s more like the AI in the movies rather than a chatbot. If this is achieved, then OpenAI will have reached level 2, which they call “Reasoners” of their 5 level plan to get to superintelligence. Superintelligence is basically AI that will surpass human intelligence in every single aspect, and will instead operate like an organization managing multiple things at once, while improving itself.
We have a long way to go before this is a reality, but I don’t even know what society will look like if that happens.
And at this point, you’re probably wondering how any of this relates to the “anonymous-chatbot” released on the Chatbot Arena.
Well OpenAI has been known to put unreleased models on the arena for evaluation. They don’t publicly tell people they do this either. GPT-4o, the model that most people use for ChatGPT was released on May 13th, 2024, but William Fedus, an employee at OpenAI disclosed that they had been testing the model in the arena under the codename i'm-also-a-good-gpt2-chatbot for about week before release.
That model shattered the Chatbot arena’s leaderboard since the whole point of the website is to compare AI models. Now, a new model under the name “anonymous-chatbot” is shattering records and soaring to the top of the website’s leaderboards. So maybe we will get the release of Project Strawberry, whether it be GPT-5 or some other upgrade in a week.
I do want to note that OpenAI has not taken ownership for “anonymous-chatbot.” But because of Sam Altman’s tweet of Strawberries over the same weekend, they think it can’t be a coincidence.
There are a lot of doubters that this anonymous-chatbot is related to Project strawberry or GPT-5. After testing it, it may not have that”reasoning” capability that it should. There's also the thought process that the website simply isn’t built for something like Project Strawberry which would be able to autonomously scour the web for information.
Sam Altman could be playing a giant game. He’s very smart, and known to be slightly deceptive. When OpenAI released their Voice AI, it mirrored the voice of Scarlett Johansson from the movie Her. Scarlett Johansson had told Sam Altman no to using her voice a few months before the launch and 2 days before. Open AI reluctantly took the voice down when she sued them. At this point, we don’t know what Sam is doing.
If I had to guess, all this is no coincidence and OpenAI will have a major update soon. This is good news. It means ChatGPT will only get smarter for everyone. If you’re not using AI please start. In a few years, not learning to use AI will be like not learning to use a computer.
Anyways, I just want to give a big shoutout to everyone that has supported me through the last few months, you know who you are. Last year, I wasn’t sure if UNinvested would still be around, but this is truly my passion and I’m just having fun!
In the meantime, I’m Sahil and thank you for listening to UNinvested. I'll be back for the next episode on the 1st Thursday of the month at 6 pm.
Peace.
