While OpenAI promised improvements with the ChatGPT update, many are crying disappointment. The performance of the flagship tool would have dropped drastically. Internet users are therefore perplexed. Simple impression or real regression? Let's dive into this affair which has generated a lot of digital ink.
Since its launch, ChatGPT has become an essential reference in the world of artificial intelligence. The latest update to the GPT-4o model, unveiled with great fanfare by OpenAI. This promised smoother writing, more relevant responses and better file management. However, barely deployed, this version is accused of regression. Between criticisms from Internet users and figures that don't lie, we wonder if ChatGPT is losing its strength?
A ChatGPT update that shakes everything up
You have probably seen the latest ChatGPT update? Apparently, this flagship OpenAI language model has regressed, and not just a little. However, the company proudly announced on November 20 that their update would improve the quality of writing and the relevance of the answers.
« The template's creative writing capability has been improved: more natural, engaging and personalized writing to improve relevance and readability », affirmed OpenAI on X.
An analysis published by Artificial Analysis indeed proves that the performance of GPT-4o has dropped. The quality index went from 77 to 71a drop that places GPT-4o at the same level as the mini version. And that's not all!
On the benchmark GPQA Diamond (a test known to assess the intelligence of an AI model), GPT-4o fell from 51% to 39%. It's the same for MATH tests with a score of 78% to 69%. Here, I find that we are really far from the promise of a model “ improved ».
Despite everything, GPT-4o is twice as fast as before. Before the update, the model produced 80 words per second, and now it's 180 words per second.
Small size, big questions
Artificial Analysis researchers believe the November pattern is likely smaller than that of August. We don't know why OpenAI would do this? What is certain is that this update leaves a bitter taste.
« Since OpenAI has not reduced prices for the November 20 release, we recommend that developers do not move workloads from the August release without extensive testing », underline the researchers.
I also remind you that GPT-4o was launched in May 2024 and that this model was expected to outperform GPT-3.5 and GPT-4. It also had to excel in areas like real-time translation, visual AI and, of course, conversation. Suffice to say that this version was a master card in the OpenAI strategy. But its regression seriously puts the company's reputation at risk.
So what happened? OpenAI has yet to respond directly to the criticism. But if I may, I think maybe they are optimizing performance for manage a large influx of users. But if it is at the expense of quality, it risks going badly.
And what do you think about it? Have you already tested this famous November version of GPT-4o? Share your views in the comments?
Share the article:
Facebook
LinkedIn
Our blog is powered by readers. When you purchase through links on our site, we may earn an affiliate commission.
