OpenAI CEO Sam Altman shared an update on the company’s plans for GPT-4.5 and GPT-5. He emphasises the need to simplify their product lineup, and… “magic unified intelligence”?
Altman also acknowledges that OpenAI’s current offerings might be too complex for users. He wants AI to “just work” for everyone.
With AI barriers still relatively low in South Africa, more intuitive and user-friendly models would certainly be welcome.
Let’s unpack what he had to say.
Upcoming releases: GPT-4.5 and GPT-5
Altman confirmed that GPT-4.5 (internally called ‘Orion’) will be OpenAI’s last non-chain-of-thought model (more on that further down).
Chain-of-thought models are designed to mimic human-like reasoning. This allows large language models (LLMs) to break down and handle complex tasks more effectively.
However, this doesn’t necessarily mean OpenAI will never release another model without chain-of-thought reasoning in the future. It simply marks a shift in their approach.
Moving forward, OpenAI aims to develop models that think in steps, mimicking human reasoning instead of generating responses all at once.
After GPT-4.5, OpenAI plans to merge its o-series and GPT-series models. Altman says it’s to create a “unified system” that can use all available tools efficiently.
A better user experience
A big part of OpenAI’s plan is getting rid of the hassle of picking the right AI model for the job.
Altman says he hates the whole “model picker” setup and wants AI to just work without users having to think about it.
We hate the model picker as much as you do and want to return to magic unified intelligence.
Hold up, I have a question…
WTF is ‘magic unified intelligence’?
So, Altman wants a system that automatically knows what you need and delivers it effortlessly. He refers to this as ‘magic unified intelligence’.
My ignorance might be showing here, but I’m not familiar with the term ‘magic unified intelligence’.
I’m assuming it’s a phrase Altman coined for the type of seamless AI experience he is envisioning.
What is ‘Chain of Thought’ reasoning?
Chain-of-thought (CoT) reasoning is a technique that allows AI to break down complex tasks into logical steps before generating a final answer.
Think of it as showing how you calculated a math problem. Rather than jumping straight to an answer, the AI ‘thinks out loud’ like a human would, step by step.
This helps improve accuracy and allows the model to handle multi-step reasoning more effectively.
A 2022 study by Google Research found that this method significantly enhances problem-solving abilities, especially in tasks requiring planning and logic.
Who will have access to GPT-5?
There will be various subscription tiers for GPT-5:
Free GPT-5 tier:
Unlimited chat access to GPT-5, yay!! (Altman added two exclamations in his update, for emphasis),
But it will be at the model’s standard intelligence setting. Basically, the default mode, and there will be restrictions to prevent abuse.
GPT-5 for Plus Subscribers
According to Altman, the Plus subscription will be able to run GPT-5 at a higher level of intelligence.
GPT-5 for Pro Subscribers
The big brain version of GPT-5. It run GPT-5 at an even higher level of intelligence.
These models will include features such as Voice, Canvas, Search, and Deep Research capabilities.
So essentially, ChatGPT is going to be useful for everything from casual chats to more serious work. Much more serious work.
From GPT-3.5 to GPT-5
It’s wild to think how far ChatGPT has come.
Remember the early days? Responses were clunky, sometimes hilariously off, and you had to babysit it through conversations.
Now, we’re looking at a model that better understands nuance, context, and intent.
If this keeps up, we’ll soon be handing over our life admin to our friendly AI chatbot. Which, honestly, can’t come soon enough...
ChatGPT’s early days:
The very first ChatGPT model released to the public was ChatGPT (based on GPT-3.5) on November 30, 2022.
But OpenAI had been developing LLMs for years before that.
The first-ever GPT-1 model (Generative Pre-trained Transformer model) was created in 2018 but wasn’t released to the public.
GPT-2 was announced in 2019, though OpenAI initially withheld it over concerns about potential misuse. However, smaller versions of the model were made available later on.
ChatGPT, meet the world
Then came GPT-3 and GPT-3.5—the latter being the first widely used model.
GPT-3 powered OpenAI’s API and early AI tools, but GPT-3.5 introduced ChatGPT, making AI-powered conversation accessible to the public.
GPT-4 was released in March 2023. This was a massive upgrade. Better reasoning; fewer AI hallucinations.
It also introduced multimodal capabilities, allowing it to process both text and images. Users could upload pictures for analysis or discussion.
What was the first thing I did when we could upload images? Tested it with a picture of Smudge. True story. His are are “strikingly expressive”.
Click here for the bigger image
Looking back, 2023 was an interesting year for me…
- I asked 12,219 questions.
- Most active days were Mondays.
- Most active time? 1 am….
- Filled 127 conversations.
- Only 127 back then!
- I was polite 3,972 times,
- and I used bad words 786 times.
- No comment about that.
So what’s next for good ol’ ChatGPT?
OpenAI’s roadmap makes one thing clear: They want AI to be smarter, smoother, and easier to use.
Instead of users having to wrestle with complicated settings, they’re aiming for a system that just works.
Something powerful enough for pros but simple enough for everyday users.
