by By Anna Tong
San Francisco (Reuters)-Ondenai’s GPT-5, the newest episode of the AI technology that the Chatgpt-Juggernaut driven in 2022, is set for an imminent release and users will examine the step of GPT-4 with the earlier improvements to the research laborium.
Two early testers from the new model told Reuters that they were impressed by the ability to solve and resolve science and math problems, but they believe that the jump-4 jump-4 to GPT-5 is not as great as those from GPT-3 to GPT-4. The testers, who signed non-dismantling agreements, refused to be called this story.
OpenAi refused to comment on this story.
The jump of GPT-4 was based on more computing power and data, and the company hoped that “scaling up” would consistently lead to improved AI models.
But OpenAi, which is supported by Microsoft and is currently appreciated at $ 300 billion, came to pick up in issues. A problem was the data wall that the company encountered, and OpenAi’s former chief scientist Ilya Sutkever said last year that while the processing force grew, the amount of data was not.
He referred to the fact that large language models have been trained on massive data sets that scrape the entire internet, and AI laboratories have no other options for large strives of textual data generated by people.
Apart from the lack of data, there was another problem that ‘training runs’ probably have more out of hardware for large models, considering how complicated the system is, and researchers may not know the final performance of the models until the end of the run, which can last for months.
OpenAi did not say when GPT-5 will be released, but the industry expects it to be every day, according to media reports. Boris Power, head of applied research at OpenAi, said on Monday in an X-post: “Excited to see how the public receives GPT-5.”
“OpenAi made such a big leap from GPT-3 to GPT-4, which has since been an enormous amount of anticipation on GPT-5 since then,” said Navin Chaddha, managing partner at Venture Capital Fund Mayfield, who invests in AI companies but is not an openi-investor. “The hope is that GPT-5 AI applications will unlock that go beyond chat to fully autonomous task performance.”
‘Test time calculations’
Almost three years ago, Chatgpt introduced the world to Generative AI, dazzling users with his ability to write human prose and poetry, and soon became one of the fastest growing apps ever.
In March 2023, OpenAI Chatgpt succeeded with the release of GPT-4, a large language model that made huge jumps ahead in intelligence. While GPT-3.5, an earlier version of the model, received a Balke exam score in the lower 10%, GPT-4 has successfully passed the simulated Balke exam in the top 10%.
GPT-4 then became the model to beat and the world came to terms with the fact that AI models could perform better in many tasks than people.
Soon other companies started. In the same year, Alphabet’s Google and Anthropic – which are supported by Amazon and Google – released competitive models to GPT -4. Within a year, open-source models on the same footing with GPT-4 such as Llama 3 models from Meta Platforms were released.
Together with the training of large models, OpenAi has now invested in a different route, called “Test-time Compute”, which canalize more processing power to resolve challenging tasks such as mathematics or complex operations that require human reasoning and decision-making.
Sam Altman of the company said earlier this year that GPT-5 would combine both the test-time comput and its large models. He also said that the model and product range of OpenAi had become ‘complicated’.
(Reporting by Anna Tong in San Francisco; Edit by Sayantani Ghosh and Muralikumar Anantharaman)
