Dive Brief:
- OpenAI released a slew of new models Thursday, including an updated GPT-4 Turbo preview model, the company said.
- “This 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,” OpenAI said in a blog post.
- OpenAI introduced an alias function that will automatically redirect GPT-4 Turbo users to the preview model and future updates. The AI startup said it plans to make GPT-4 Turbo with vision generally available in the coming months.
Dive Insight:
OpenAI’s released the initial iteration of GPT-4 Turbo in preview at the startup’s Dev Day in November to industry and customer interest. More than 70% of requests from GPT-4 API customers have transitioned to GPT-4 Turbo since its release, according to OpenAI.
But the model didn’t always work the way it should. In some instances, users said the model refused tasks or generated half-completed responses.
OpenAI acknowledged customer complaints in December and said teams were investigating the issue, though the company emphasized the complex nature of training chat models.
“When releasing a new model we do thorough testing both on offline evaluation metrics and online A/B tests,” OpenAI said in a post to X in December. “After receiving all these results, we try to make a data driven decision on whether the new model is an improvement over the previous one for real users.”
Even small changes in model training can have trickle-down effects that are hard to manage and detect, researchers told CIO Dive. Model optimization remains an ongoing process.
