Artificial interest
OpenAI introduces next version of technology behind ChatGPT
ChatGPT can formulate sentences that are almost indistinguishable from those of a human. photo
© Philipp Brandstädter/dpa
The text robot ChatGPT has been causing a stir for months. The technology that runs ChatGPT just got a major upgrade. However, weaknesses and risks of the software remain.
The start-up OpenAI has presented the next version of the technology behind the popular text machine ChatGPT. Among other things, GPT-4 should deliver better results than the previous variants, as OpenAI announced. Problems with technology – such as the fact that it can simply invent alleged facts – continue to exist, but should occur less frequently.
ChatGPT and the Dall-E software, which can generate images from text specifications, are currently still based on the previous GPT generation. Paying customers of OpenAI get access to GPT-4 for their services. There is a waiting list.
Microsoft bought in
However, some customers are already using the technology. The language learning app Duolingo, for example, uses GPT-4 for dialogue training, which is available in a new, more expensive subscription. Microsoft confirmed that its Bing search engine has been using GPT-4 for a few weeks. Microsoft bought OpenAI in a multi-billion dollar deal, and the money secured access to the enormous computing power needed, among other things.
Various services immediately announced new offers based on GPT-4. The Intercom company wants to use them for customer service chatbots that users can talk to. The provider DoNotPay, with which you can assert claims, wants to use it to automatically generate complaints about unwanted advertising calls. A video made the rounds online in which GPT-4 created a working website based on a handwritten sketch.
For the GPT technologies, the software captured enormous amounts of text and images. On this basis, she can formulate sentences that can hardly be distinguished from those of a human being. The program estimates which words could follow next in a sentence. One of the risks of this basic principle is that the software will “hallucinate facts”, as OpenAI calls it.
Describe pictures with words
That could also happen to GPT-4, albeit less frequently than before, the blog entry said. The new version could also make simple logical errors and spread prejudices. GPT-4 only knows facts that happened before September 2021 – and it does not learn from experience, OpenAI emphasized. Microsoft temporarily limited the length of conversations with the software at Bing, because after a while the program’s answers sometimes slipped into the absurd.
GPT-4 is also said to be good at analyzing images and describing them with words – but OpenAI is not initially making this function available to customers. One of their risks is that the software could reveal its identity to people in the image if it recognizes people from publicly available recordings.
On Wednesday, experts rated the progress in development as significant. Rasmus Rothe, chief technology officer at Berlin-based artificial intelligence investment platform Merantix, said it would be presumptuous to say that GPT4 is just a minor evolution of GPT3 or GPT3.5. “The jump in such a short time is impressive.”
However, GPT4 is still struggling with the problem of “hallucination”, in which the model generates incorrect information because it does not know what is true and what is not. “However, improvements have also been made here, and the machine has a better understanding,” said Rothe.