In a tweet on X.com, OpenAI CEO and founder Sam Altman listed what people most want most from the company in 2025. It makes for very interesting reading, and it’s not all about ChatGPT.
In answer to the question “What would you like OpenAI to build/fix in 2025?” Altman has identified a list of common themes, and honestly, it’s a great wishlist for OpenAI to get started on. The themes are: AGI agents, a much better 4o upgrade, much better memory, longer context, a “grown up mode”, deep research feature, a better Sora, and more personalization.
common themes:AGIagentsmuch better 4o upgrademuch better memorylonger context“grown up mode”deep research featurebetter soramore personalization(interestingly, many great updates we have coming were mentioned not at all or very little!) https://t.co/lMZmlZif66December 30, 2024
AGI
The first thing on Altman’s list is AGI. AGI stands for Artificial General Intelligence and is the next level of artificial intelligence, where we create human-equivalent, or smarter-than-human intelligence, which some people think comes with as many dangers as opportunities.
I don’t think we’ll be getting AGI in 2025 despite it being top of the wishlist. When I interviewed Dr Ben Goertzel, one of the people who created the term AGI, last year he thought that 2029 was a more realistic date for it, but that’s not to say that companies like OpenAI won’t make some significant strides towards it in 2025.
Agents
The second thing listed on Alman’s list is Agents. These are automated AI bots that can perform tasks for you, and act like a true personal assistant, perhaps rescheduling missed appointments or ordering food items you’ve run out of.
We were expecting OpenAI to release its long-awaited Operator Agent as part of its ‘12 days of OpenAI’, but we were left disappointed. Since Operator Agent was supposed to be released in January, we’re expecting OpenAI to release something in the very near future.
ChatGPT improvements
I’m assuming that most parts of the wishlist referred to ChatGPT, specifically: longer context, more personalization, “grown up mode”, much better 4o upgrade, deep research feature and much better memory.
It’s been noticed that the rate of improvements in Large Language Models is slowing down significantly the more they evolve, so people’s everyday experience of ChatGPT is not as significantly different between the 4o model and the new o1 model as it was between the older ChatGPT 3 and ChatGPT 4 models. It’s interesting to note that most of the requested changes are not about the ability to answer deep math questions or generate better code, but rather they’re about more simple things like a better memory of who it’s talking to and a more personalized user experience.
Call me superficial, but I would love these things too. It’s not immediately clear what a “grown up mode” means, but presumably, this would involve taking off some of the guardrails that ChatGPT currently works within. Personally, I think they’re there for a good reason, so I’m not sure how I feel about that.
An example of video generated by Sora. (Image credit: Future)
A better Sora
Sora was released, to a lot of praise, as part of ‘12 days of OpenAI’ event, and while it is without doubt one of the best AI video generators out there, its long development process (it was announced in February 2024) has allowed others to catch up.
While it may sound ungrateful to ask for a “better Sora” so soon after its release, rivals like Pika can already do things that it can’t do and are stealing its thunder. I’d like to see a more practical Sora. Give me some more reasons why I’d want to create AI video. We’re not all budding film directors, and some of us just want to use AI to make our lives easier.
What would I love to see?
I don’t think the world is quite ready for AGI yet, even if it was a technical possibility. Of everything on the wishlist, improvements to ChatGPT top my list. Particularly better memory, a feature that Google is spending a lot of time developing for ChatGPT rival Gemini, along with its deep research feature.
To finish off his tweet, Altman ended with the enigmatic, “Interestingly, many great updates we have coming were mentioned not at all or very little!”
Perhaps he’s just trying to keep us on our toes about what 2025 will look like, but wondering what OpenAI is going to come up with that’s not on the wishlist has certainly whetted my appetite for 2025. Bring it on!
You might also like…
Our predictions for AI in 2025 – what next for ChatGPT, Apple Intelligence and moreCorporate executives are being increasingly targeted by AI phishing scamsWe asked AI chatbots Gemini and ChatGPT to design our workouts – then we tried them out