GPT-4 Turbo is one of the most powerful AI models out there, and OpenAI is working on making it even more powerful. One thing that users have been dealing with, however, is the concept called “laziness.” Well, GPT-4 Turbo is getting a new update that will solve the laziness issue and bring some other useful changes.
What is AI laziness? When you ask a model to do tasks, you expect them to be completed thoroughly. Well, when an AI model doesn’t complete the task that you ask it to do, take shortcuts, or overall show little effort to provide proper information, that’s AI laziness. It’s pretty straightforward; it’s basically like human laziness.
AI laziness hasn’t gotten as much traction as AI hallucinations, but it’s still rather bad. So, anything that can be done to reduce it will be helpful.
GPT-4 Turbo received an update that will help fix the AI laziness
OpenAI recently posted a blog post, stating that the updated model “completes tasks like code generation more thoroughly than the previous preview model and is intended to reduce cases of ‘laziness’ where the model doesn’t complete a task.” So, people using this updated version of GPT-4 Turbo should be getting better results.
It’s important to note that the blog post referred to GPT-4 Turbo. This is the more advanced version of the model that’s currently in preview. It’s the version that’s been trained up to April 2023. We’re not sure if OpenAI upgraded the standard GPT-4 model. This is the model with knowledge that stops at September 2021. So, chances are that if you’re using the standard GPT-4 model, you will still experience the laziness after the update.
Price reduction
This is good news for people using GPT-3.5 Turbo. Starting next week, input prices will be reduced by 50% and output prices will be reduced by 25%. So, if you’re using the GPT-3.5 API, you’ll be paying a lot less
Embedding models
If you’re a person who uses embedding models for your AI tasks, then this is going to be good news for you. OpenAI just announced two new text embedding models. One is called text-embedding-3-small and together is called text-embedding-3-large.
The first one is a highly efficient model, and it’s leaps and bounds over its predecessor which is text-embedding-ada-002. As for the large model, this one is more powerful with up to 3,072 Dimensions.
2024-01-26 15:05:41