OpenAI has launched the latest version of its wildly successful online chatbot, ChatGPT.
Known as GPT-4 (Chat GPT Powered by GPT-3.5), the new model’s title feature is its ability to recognize and interpret images.
For example, a chatbot can suggest recipes based on a picture of the inside of a refrigerator, so it can do wonders for my lackluster cooking habits (if only the 2020 banana bread craze came along).
But it also improves on its predecessor’s ability to interpret text and write its own, providing more complex answers, channeling different styles and voices, and being able to process thousands of words at a time.
How is it improved?
Currently, it’s only accessible to paid subscribers to OpenAI’s ChatGPT Plus service — but it’s clear that GPT-4 boosts the potential for generative AI.
For example, it significantly outperforms ChatGPT on established exams.
Dr Peter Van der Putten, AI Professor and Director of AI at Leiden University, said: “It seems to be more capable not only at writing text, but at generating more detailed and specific prose with a deeper understanding of the problem. ’” Pegasystems’ artificial intelligence told Sky News.
“From a marketing standpoint, it’s a little harder to market. But this ability to be more specific, more prosaic, and show a real understanding and explain why, will increase people’s trust in it.”
In fact, while ChatGPT often comes across as bland and verbose at best, at worst it can tell outright lies — and still be confident.
OpenAI says GPT-4 may still be prone to such problems, although it insists that security measures have been improved.
GPT-4 can also go beyond the fixed tone of ChatGPT, where users can ask to respond in a particular style — like the Shakespeare pirate seen above.
It may prove that the model appropriates art, writing stories, poems and screenplays in the voice of a well-known creator, or Might even help make you sound more interesting on dating apps.
Is ChatGPT the ultimate homework cheat?
How artificial intelligence is changing the way we search the web
Will it lead to more work?
In an online demo, OpenAI President Greg Brockman showed how GPT-4 can quickly come up with the appropriate income tax deduction after inputting a lot of tax laws — a problem he couldn’t solve on his own.
Beyond simplifying mundane tasks like tax filing and baking, experts say the implications for jobs, research and academia could be far-reaching.
Dr Andrew Rogoyski, from the University of Surrey’s Institute for Human-Centered Artificial Intelligence, said OpenAI’s tests showed “significant improvements in performance in science, mathematics and economics”.
The number of companies already leveraging the technology is a testament to its potential — Microsoft’s new Bing search enginepayments platform Stripe, and language-learning app Duolingo all use it.
No matter where you get the podcast, you can click to subscribe to Sky News Daily
While ChatGPT has already been identified as a threat to jobs like customer service and computer coders, OpenAI seems to believe GPT-4 can fill many more roles.
“We expect GPT-4 to impact even jobs that have historically required years of experience and education, such as legal services,” its release document states.
We’ve already seen AI answer questions at Westminsterso maybe the courthouse is not far away.
ChatGPT helps write Sky News articles
Google shows off its highly anticipated AI chatbot
Much of the speculation leading up to GPT-4’s unveiling has focused on compelling new capabilities, such as generating video.
Given the risk of releasing a model when it is not ready, we end up with results that advance public-facing AI may happen more slowly (Microsoft’s 2016 Tay bot says hello, or maybe it’s offensive).
Sridhar Ramaswamy, co-founder of search engine Neeva, which has its own GPT-style artificial intelligence, told Sky News that OpenAI’s upgraded model was a “natural but still important evolution”.
“Big changes are always like this—many incremental changes add up to a huge impact,” he said.
But experts say this is more evolution than revolution, and OpenAI admits it can’t yet replace humans.
“These models still don’t have a basic understanding of ‘real’ versus ‘false,'” Mr Ramaswamy said.
“But GPT-4 is less than three months away from ChatGPT. It does feel like there’s tremendous momentum in this space.”