ChatGPT’s Decline, the FTC and Microsoft

By Staff Reporters

***

***

ChatGPT (Chat Generative Pre-Trained Transformer) is an artificial intelligence chatbot developed by OpenAI and launched on November 30, 2022. It is notable for enabling users to refine and steer a conversation towards a desired length, format, style, level of detail, and language used. Successive prompts and replies are taken into account at each stage of the conversation as a context.

However, the accuracy of OpenAI’s generative language model appears to have declined much this Spring.

LINK: https://openai.com/blog/chatgpt

A not-yet-peer-reviewed study by Stanford and Berkeley researchers of ChatGPT-4 and ChatGPT-3.5 found a huge difference in the chatbot’s answers between March and June of this year—often for the worse, something people were already reporting anecdotally online.

  • GPT-4 went from 97.6% accuracy in identifying prime numbers a few months ago to only…2.4% in June.
  • The newer version of the language model got better at fending off problematic prompts, like coming up with illegal money-making schemes. But rather than explaining why queries are troublesome, it’s now more likely to simply say something like, “Sorry, but I can’t assist with that.”
  • When asked to create computer code, GPT-4 generated functional work 52% of the time in March and only 10% of the time in June. The lines of code it provided weren’t wrong, but they started to be accompanied by non-usable text, which could create headaches for companies trying to integrate ChatGPT into programming workflows.

***

Additionally, the FTC is looking into ChatGPT. The commission seeks to understand whether OpenAI has broken any consumer protection laws. OpenAI received a 20-page demand letter highlighting the commission’s concerns over its data security practices. Microsoft’s Activision Blizzard is likely to be affected if OpenAI is found in the wrong.

***

***
COMMENTS APPRECIATED

Thank You

***

Leave a comment