OpenAI's GPT-4 Turbo Update Fixes Laziness Problem – CDOTrends
OpenAI has announced updates to the AI models that power ChatGPT, including a fix to issues of ”laziness” that some users have reportedly observed.
You can read about complaints by users about the worsening output of GPT-4 in our earlier report here. Indeed, a study of GPT-3.5 and GPT-4 by external researchers found that their performance and behavior “varied significantly” and have “gotten substantially worse” over time.
The fix to the laziness issue is tucked away in its announcement of a new version of the powerful GPT-4 Turbo, which OpenAI says completes tasks like code generation more thoroughly than the previous preview model.
According to OpenAI, the updated GPT-4 Turbo model should be better now: “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.”
However, the blog post stopped short of explaining what went wrong, and what was changed to alleviate the problem. In addition, OpenAI says that GPT-4 Turbo with vision (GPT-4 V) will launch to general availability “in the coming months.”
OpenAI also announced a new version of GPT-3.5 Turbo (gpt-3.5-turbo-0125) with various improvements, including a fix for a bug which caused a text encoding issue for non-English language function calls.
Moreover, the prices for GPT-3.5 Turbo will also be reduced for the third time in the past year. Specifically, input cost will drop by 50% and output by 25%, bringing the price to USD0.0005 per thousand tokens in, and USD0.0015 per thousand tokens out.
Since the release of ChatGPT that took the world by storm, the generative AI models from OpenAI have been used to gauge performance of every other model released.
For instance, Chinese AI firm iFlytek says it aims to match the performance of GPT-4 Turbo with upgrades to its Spark Model by the first half of this year. As reported, the Spark V3.5 is close to the performance level of GPT-4 Turbo, and is said to surpass it in areas such as language comprehension and mathematical abilities.
Paul Mah is the editor of DSAITrends, where he report on the latest developments in data science and AI. A former system administrator, programmer, and IT lecturer, he enjoys writing both code and prose.
Copyright ©2018 – 2023 CDOTrends. All Rights Reserved.
