OpenAI is working on GPT-4 to make ChatGPT less “lazy” again. The chatbot has provided more limited responses in recent months as usual.

OpenAI today releases a new version of GPT-4 Turbo in preview, gpt-4-0125. “This model performs tasks such as code generation more thoroughly than the previous preview model and is intended to reduce the number of cases of ‘laziness’ in which the model fails to complete a task,” the notable release notes read.

ChatGPT users noticed that the chatbot became “lazier” starting in late 2023. Instead of carrying out tasks, the chatbot preferred to provide a “recipe” with guidelines so that the user could do it themselves. OpenAI immediately recognized the problem and explained that it was certainly not intentional.

With another thorough update to the chatbot, OpenAI hopes that its chatbot will prove to be more hardworking again. There are always some tricks you can use in your prompts to get longer answers from ChatGPT, such as: B. by asking the chatbot to reason step by step or by promising a financial incentive.

It’s not the only news OpenAI has. A new version of GPT-3.5 Turbo will go live next week and will come with a price cut for developer APIs to integrate the LLM into their own applications. Input prices fall by fifty percent, output prices by 25 percent. OpenAI offers an improved moderation API free of charge to ensure that chatbots do not get off track.