According to press reports, the new generation of OpenAI's language model, GPT-5, may be released within a few months. According to those familiar with the beta version of the new generative AI, it is “significantly better” than GPT-4, which has shown fluctuations in performance recently.
OpenAI GPT-3.5, a subscriber chatbot based on GPT-4 language models, ChatGPT opened a new era in applied AI in November 2022 with great interest. The once small, independent science-oriented company struck a billion-dollar deal with Microsoft, and then, on roughly the first anniversary of ChatGPT's release, the chaotic palace erupted unexpectedly: Sam Altman was fired, but back on the job. Company president a week later.
The few months that followed were relatively quiet, with only the New York Times suing them for unauthorized use of their articles and three other newspapers suing them for copyright infringement. In addition, the famous technology mogul Elon Musk also launched a strange lawsuit against the company, because, according to his views, it violated its founding document that promotes openness and scientific research.
That's out of the question because I don't know
In a recent conversation with popular broadcaster Lex Friedman, Sam Altman talked about the company's plans for the near future and expected products. The new old CEO said on the show that he doesn't know when GPT-5 will be introduced. Based on Lex Friedman's suggestion of two flashes if it happens this year, Altman
Refused to blink
He dodged the question by saying that several products would be introduced this year, and hinted that it was not even certain that they would be called GPT-5.
the Interested in trade I got a little more detail than those provided by early testers, who said that GPT-5 is clearly more powerful than its predecessor. The new AI impressed not only testers, but also company representatives and executives who learned about the chatbot built on it. One person who came across the beta reported that the new model actively and independently uses other AI-based devices, and is believed to be multimodal, meaning it can't just interpret text and images.
Improve trends
GPT-4 was introduced in March 2023, and its performance has recently started to deteriorate and become lazier, according to user reports.
In light of this, it was particularly interesting that Altman said in the podcast conversation that GPT-4 was “vulnerable,” but immediately pointed out that GPT-5 is expected to represent a significant technical leap over 4 like GPT-4. compared to 3.
I think our job is to live a few years into the future and remember that the tools we use today are a bit weak compared to how we'll look at them, so let's solve them so the future will be better.
Altman explained the testimony.
According to the leaked information, GPT-5 is currently being trained, after which the so-called “red team” will test the system for vulnerabilities and errors.
As GPT-5 approaches, it's worth noting that it has been repeatedly suggested that it will become an artificial general intelligence, if you like, a machine that reaches the intellectual level of a human. The testers who spoke did not mention anything of the sort, so there is currently no indication that development will continue.
At the same time, a strange picture emerges for competitors. Apple is developing its own language model full steam ahead, but at the same time it has fallen so far behind in AI progress that it is being debated whether it will allow Google's Gemini model into the iPhone. Since Gemini features are available on Samsung and Google Pixel phones, Google seems to dominate mobile, so OpenAI and Microsoft probably won't rest on their laurels.
(Ars Technica, Mashable, Znet)
Follow Index on Facebook too!
