ChatGPT is undergoing Turbo upgrade


If you pay for ChatGPT, whether Plus, Team or Enterprise, GPT-4 Turbo is now better. In a post on X (formerly Twitter) last Thursday, OpenAI announced that the latest version of its GPT-4 Turbo model is now available to all paying ChatGPT users and includes some interesting updates.

So, what's new? This is all a bit general, but OpenAI says the latest version of GPT-4 Turbo now performs better in writing, math, logical reasoning, and coding. As you can see from the chart in the announcement post, the April 9th ​​version of GPT-4 Turbo is an improvement over the January 25th GPT-4 Turbo of this year in almost every category. Some metrics are more eye-catching than others: the biggest jumps occur in GPQA, an AI benchmark on scientific datasets, where GPT-Turbo improves from 39.7% to 49.1%; and in the MATH benchmark, which improves from 64.2% to 72.2% .

Interestingly, one area where January's GPT-4 Turbo model outperformed April's was in the HumanEval test, which tests the LLM's ability to generate code: the former scored 88.2%, while the new model scored just 87.6%. It's a small difference, but an interesting step back that OpenAI doesn't address.

However, what OpenAI is particularly focused on is the overall improved experience with ChatGPT and the new GPT-Turbo, which is what users like me are most likely to notice. The company says that when you write using ChatGPT, the model's responses will be "more direct, concise, and use more conversational language." You can see this in the example OpenAI used in its post: it asked ChatGPT to generate a text message to remind a friend to RSVP for a birthday dinner invitation. January's GPT-4 Turbo model generated a lengthy response, and while still potentially useful, does yield artificial intelligence in some areas. However, the latest model generates a shorter, more casual reminder about RSVP.

Perhaps more importantly, GPT-4 Turbo's data set is as of December 2023. The previous version was stuck at April 2023, which meant it was missing an entire year of context. While the new GPT-4 Turbo knows nothing about 2024, unless it's connected to the internet, it knows about eight months more of current events than before.