OpenAI CEO Sam Altman has expressed his affection for the GPT-2 language model, despite his previous criticism for all the models before GPT-5, which is yet to launch.
What Happened: On Monday, Altman took to X, formerly Twitter, and confess his fondness for GPT-2. He said, “I do have a soft spot for GPT2.”
Earlier this year, in an interview with Korean publication, Maeil Business Newspaper, Altman said, “GPT-2 was very bad. GPT-3 was pretty bad. GPT-4 was pretty bad. But GPT5 will be good.” Altman gave this interview in January when he attended the World Economic Forum held at Davos, Switzerland.
OpenAI has been consistently improving its GPT models since the release of GPT-1 in 2018, reported Forbes. The latest version, GPT-4, was launched in March last year, with GPT-5 scheduled for release later this year.
Here’s a look at how the GPT models have improved over the years:
GPT-1: Introduced in June 2018, GPT-1 comprised 117 million parameters, establishing the groundwork for the current incarnation of ChatGPT. It showcased the effectiveness of unsupervised learning in comprehending language, employing books as its training material to predict the subsequent word in a sentence.
GPT-2: This model was unveiled as a substantial advancement, boasting 1.5 billion parameters in February 2019. It displayed remarkable progress in text generation, generating coherent, multi-paragraph content. However, owing to concerns about its potential misuse, GPT-2 wasn’t immediately made available to the public.
Subscribe to the Benzinga Tech Trends newsletter to get all the latest tech developments delivered to your inbox.
GPT-3: In June 2020, the introduction of GPT-3 marked a significant advancement. Trained on an impressive 175 billion parameters, this model revolutionized text generation. Its sophisticated capabilities found extensive applications, including drafting emails, composing articles, crafting poetry, and even generating programming code. The updated version of it, GPT-3.5, was later publicly released to public in November 2022.
GPT-4: Launched last year in March, GPT-4 is currently only available for ChatGPT Plus subscribers. It came with new image- and text-understanding, lower likelihood of dangerous output, ability to follow users’ intention and the capability of searching internet in real-time.
GPT-5: The upcoming model is poised to offer a substantial improvement over its forerunner, GPT-4, which has encountered performance challenges. The upcoming model is slated for a summer release, with select enterprise clients already being given previews of its features. On CEO said described it as “really good, like materially better.”
Photo courtesy: Shutterstock
Check out more of Benzinga's Consumer Tech coverage by following this link.
Disclaimer: This content was partially produced with the help of Benzinga Neuro and was reviewed and published by Benzinga editors.
© 2024 Benzinga.com. Benzinga does not provide investment advice. All rights reserved.
Trade confidently with insights and alerts from analyst ratings, free reports and breaking news that affects the stocks you care about.
