(Credit: Iryna Imago/Shutterstock.com)
Sometimes, we all feel lazy, but when your popular AI tool refuses to complete a task, that's a problem. It's precisely the complaint ChatGPT users made a few months ago. OpenAI says a new ChatGPT 4 Turbo preview fixes the problem, but you'll need to access it through the API.
The problem first cropped up in November, when users noticed that ChatGPT would refuse to complete even simple tasks. Ask it to sort a CSV or write Python code to add two numbers, and it would return a long explanation of how that could be done and recommend the user try to do it themselves. Push ChatGPT to actually do the work, and it would finally relent. The complaints stacked up to the point that OpenAI acknowledged it in a tweet and promised a fix. That fix is here now, although only in a preview version.
An updated GPT-4 Turbo preview promises to correct the laziness problem and fix a bug impacting non-English UTF-8 generations. According to OpenAI, 70% of API users have already switched to GPT-4 Turbo because it has an updated knowledge cutoff, larger 128k context windows, and lower prices. This updated preview promises to generate more code, and require fewer prompts to complete a task.
For developers, there's also an exciting bonus: prices are going down. Using the API, input prices are about half of what they used to be, and OpenAI reduced output prices by a quarter.
As of now, OpenAI has yet to state when this preview model will make its way to the ChatGPT web interface for Plus subscribers. The previous preview arrived on web interface about a week after its announcement, so there's hope this new version could arrive soon.