In response to reports that ChatGPT is becoming 'lazy', OpenAI considers fixing it as 'unintentional'
In response to a series of reports that ChatGPT, which was built using OpenAI's top language model ``GPT-4,'' became lazy, the developer OpenAI has started an investigation.
OpenAI says it is investigating reports ChatGPT has become 'lazy' | The Independent
OpenAI looks into complaints about 'lazy' ChatGPT with GPT-4
In recent days, users of the latest version of ChatGPT, which is built on OpenAI's GPT-4 model, have been increasingly complaining that ChatGPT won't answer their questions or responds in a disinterested manner.
For example, if a human asks about the code, ChatGPT will just give a little information and leave the rest to the user, or it will cheekily say something like, ``You can do the task perfectly yourself, right?'' In some cases, they may refuse to respond, or if you continue to ask questions, the conversation may be interrupted.
GPT-4 is a model that appeared on March 14, 2023, and can be used in the paid version of ChatGPT. Although it has been a while since its release, GPT-4 has not been updated since November 11, 2023, so reports of 'laziness' that have occurred after this date are things that OpenAI did not expect. Says.
In response to the complaints, OpenAI announced the start of an investigation and consideration of corrections. “We have heard all of your feedback regarding GPT-4 delays,” they said. “Model behavior can be unpredictable, so we are considering fixes.”
we've heard all your feedback about GPT4 getting lazier! we haven't updated the model since Nov 11th, and this certainly isn't intentional. model behavior can be unpredictable, and we're looking into fixing it ????
— ChatGPT (@ChatGPTapp) December 8, 2023
AI systems like ChatGPT are notoriously expensive for companies to operate, requiring significant processing power and computational time to provide detailed answers to questions. Improvement work requires a certain amount of effort, and many people need to be involved in planning, construction, and evaluation.
OpenAI says, ``Different model training can produce models with distinctly different personalities, writing styles, evaluation performance, and even political biases, even when using the same dataset.'' ``The correction process is It's more than just a website update.''When we release a new model, we thoroughly test it with both offline metrics and online A/B testing. 'We try to use data to determine whether the model is an improvement over the previous model for real users,' he said, explaining the difficulty of making modifications.
◆Forum now open
A forum related to this article has been set up on the GIGAZINE official Discord server . Anyone can write freely, so please feel free to comment! If you do not have a Discord account, please create one by referring to the article explaining how to create an account!
• Discord | “Are you experiencing a drop in ChatGPT performance?” Haven't you? ' | GIGAZINE
https://discord.com/channels/1037961069903216680/1183717133557964831
Related Posts:
in Software, Posted by log1p_kr