From two new releases of GPT-4, o1 to a new search engine and internal drama, ChatGPT and parent company OpenAI had a notable second year. What’s next?
On November 30, 2022, OpenAI launched its first model of ChatGPT into the world.
What was supposed to start as a test for OpenAI’s models quickly became a chatbot that became synonymous with the development of generative artificial intelligence (genAI).
In 2024, OpenAI saw the launch of several new versions, including GPT-4 which came with faster intelligence across text, voice, and vision, and o1, a new series of models that can reason through complex tasks in science, coding, and maths.
Just a few weeks ago, OpenAI released SearchGPT, a browser extension that gives “fast, timely answers” to user queries with relevant web sources, bypassing search engines altogether.
The company’s been dealing with a fair share of internal issues in the background, with the resignation of co-founder Ilya Sutskever, the dissolving of the company’s team researching super intelligence, and a slew of lawsuits from US-based news companies for alleged copyright infringement.
So what’s next for OpenAI’s ChatGPT as it tries to up the ante in its third year? Euronews Next breaks it down.
OpenAI’s next ‘giant breakthrough’
In a Reddit Ask Me Anything (AMA) last month, OpenAI founder Sam Altman and his colleagues gave a little insight into their priorities for their third year.
Kevin Weil, chief product officer at OpenAI, told the AMA that a “big theme” for 2025 will be whether ChatGPT can perform tasks independently.
Altman suggested that it could look like an autonomous agent, which he would consider the company’s next “giant breakthrough”.
“AI agents,” referred to as agentic AI, will let companies design the large language models (LLMs) that run their systems to automate workplace tasks.
That’s a feat that’s already being worked on by some of OpenAI’s rivals, like Google Cloud’s Vertex AI agents, LinkedIn and Microsoft. Media reports say the next update of Google’s Gemini, Project Jarvis, could also include autonomous agents.
Kate Devlin, professor of Artificial Intelligence & Society at King’s College London, said there’s been mixed reactions to the AI agents OpenAI’s competitors have already launched.
“Some people are really positive about this and see it as being a complete game changer because it allows a lot more of a personal assistant type situation where you can outsource a lot of your tasks to the AI,” she said.
“There are some people who are wary of that and don’t like the idea of that much input or control being given to the model”.
ChatGPT’s next model: What we know so far
There have been some predictions that OpenAI will be releasing a whole new model sometime before the end of the year.
Altman told the Reddit AMA in October that there are “very good releases” coming later in the year but there are “none that we will be calling ChatGPT-5 though”.
The priority, according to Altman, is to “prioritise the shipping” of the other models, like GPT4.0 and 0.1 which were launched this year.
Reporting from the Verge a few weeks later puts the release of Orion, widely believed to be the successor to both GPT-4o and o1, to December for select companies that work closely with OpenAI so they can build their own products and features.
Tadao Nagasaki, the CEO of OpenAI Japan, teased in September a future model of ChatGPT that will be “100 times more powerful” than GPT-4, local media reports say.
Devlin said it’s possible that AI agent activity could be included in any future release of Orion.
Other updates to look for could be to Sora, an AI text-to-video model, that has so far been delayed.
The software, still being “perfected” according to Weil, was leaked this week.
Altman said the next version of DALL-E, the picture-generating software from OpenAI, is “worth the wait” but does not have a release plan yet.
Should OpenAI consider specialising or scaling down?
One of the things that OpenAI and other companies will have to watch out for in 2025 are some of the overall shortages taking over the industry, Devlin continued.
“[AI companies] need more compute, they need more energy, they need more data,” Devlin said. “So, just how much can they do with the limitations on those things?”
One option that OpenAI could consider is to scale down to small or medium-sized LLMs that are less resource-intensive.
These smaller models would be able to “curate data” to be more specific or helpful in specific domains, like law or health, Devlin said.
“Instead of just scaling up and up and up, it’s time to reflect on what we’ve got so far because there are definite benefits… but we know there are risks and maybe we should take some time to evaluate those risks,” Devlin said.
Abdul Sadka, director of the Sir Peter Rigby Digital Futures Institute at Aston University, said OpenAI should keep ChatGPT “generic” so that individual companies or sectors could refine it to the specific data sets they want it to be used for.
However, Sadka said he could see OpenAI expand to give ChatGPT more “modalities,” like the ability to recognise medical images to “potentially give you diagnostic reports about any underlying condition” a patient might have.
To help ChatGPT become more specialised, Sadka said companies using the software could build an external knowledge base that the AI hasn’t seen before to “reduce the probability of… hallucinations,” the term used to explain AI’s best-guess responses to questions it doesn’t recognise.
