[ad_1]
Several users have complained that the latest ChatGPT model (ChatGPT 4) has become less useful.
People, on openAI’s developer forum and Reddit, have complained that ChatGPT refuses to do as requested or doesn’t reproduce detailed answers to queries. In certain instances, it has told the user that they can do that task themselves and there was no need for AI.
Users speculate that this is done intentionally by OpenAI to curb the use of resources required to run AI models. For reference, one AI-generated image needs as much power as charging your smartphone.
OpenAI’s response
These complaints have been noted by OpenAI, and the investigations are underway. The company addressed the concern in a tweet, assuring users that the model hasn’t been changed since Nov 11th and the reason why AI has been behaving in such a way is still unknown.
We’re always striving to make our models more capable and useful for everybody across millions of use cases. So please keep the feedback coming.ChatGPTapp
OpenAI, through its @ChatGPTapp X handle, explained that AI model training isn’t a “clean industrial process”. The same datasets are capable of producing AI models with completely different personalities, performances, writing styles, and political biases.
Before releasing a new AI model, it is thoroughly tested on offline metrics as well as online A/B tests. These results are then evaluated to decide whether the model provides users with improvements as compared to previous ones.
The Recent OpenAI Upheaval
Problems don’t seem to be subsiding for OpenAI. A recent study found that ChatGPT is susceptible to a “divergence attack,” where users could use tricky prompts to extract training data and personal information from the AI model. This put the private data of millions of users at risk.
Before this, a major DDoS attack rendered ChatGPT useless for several users. For as long as 24 hours, users were displayed the message “ChatGPT is at capacity right now”, when they tried to access the platform.
Shortly after this, Sam Altman was mysteriously removed as the CEO of OpenAI, citing the lack of “candid conversations”. However, after a week of drama, he was reinstated as the CEO, with full backing and support from Microsoft.
More recently, CMA, the UK competition watchdog, is considering launching an antitrust probe into OpenAI’s partnership with Microsoft, considering their tie-up a virtual merger of sorts.
Amidst all these concerns, the last thing OpenAI would want is a faltering AI model. It remains to be seen whether this recent “lazy” behavior is a one-off incident and how the AI giant tackles the same.
[ad_2]
