Users of the free version of ChatGPT reported that the were unexpectedly given access to GPT-4, the advanced and multimodal capable version of OpenAI’s large language model. One user, TimelessTraveler on OpenAI’s official forum, shared their experience: “This morning, I log into my ChatGPT account and I saw a new model named ‘GPT-4 all tools, confidential,’ he said. “I am currently able to use it and upload images, files, and generate images using DALL·E-3 even though I’m on the free plan.”
The user also wondered why OpenAI put a disclaimer on the screen saying the feature was “confidential.”
This unexpected access to GPT-4—typically exclusive to subscribers to the premium version—sparked eager discussions among users. Forum member Foxabilo speculated about the occurrence. “It could be one of two things,” they wrote. “A preview of features for users who have not tried GPT-4… or it could be an error and everyone gets to try the super powerful GPT-4.”
OpenAI was unable to respond to a request for comment from Decrypt.
But the joy seems to have been short lived. In forum replies, many users reported that they suddenly lost access to the feature after tinkering with it for a couple of days.
“I got access to the free GPT4 alpha for about a day, and now it’s gone,” Steven Chalom said. Echoing this sentiment, Quri_cash noted, “Even I lost it. I had it for like 2 to 3 days, but it did not have the term ‘confidential’ for me; now it’s gone.”
The unexpected access coincides with the recent introduction of GPT-4 Turbo this week. As reported by Decrypt, GPT-4 Turbo—unveiled at OpenAI’s developer conference—is described as a more potent and cost-effective successor to GPT-4, featuring enhanced context processing and fine-tuning capabilities.
Interestingly, OpenAI’s approach to enhancing its AI models may be a move to counteract its competitor Anthropic, which is now leveraging Google’s advanced Tensor Processing Units (TPUs) to train models more capable than its current Claude 2 version. These TPUs are instrumental in developing more sophisticated models, resulting in faster, more efficient training.
The temporary access to GPT-4 in the free version of ChatGPT sparked speculation about OpenAI’s strategy against competitors like Claude, which has a Pro version, and Google Bard, which is free. Claude’s free version provides access to Claude 2, while paying customers have the same LLM but benefit from prioritized response times. OpenAI’s move, possibly a test of market competitiveness, could be a strategic response to these developments, especially considering that a possible Claude 3 and Meta’s anticipated open-source Llama 3 may be coming sooner than expected.
If OpenAI were considering offering a preview of GPT-4’s capabilities to free users, it might be a way to gauge public interest. “It is free GPT-4 of a different flavor,” A regular forum user wrote. “Possibly they want feedback from those not exposed to GPT-4 and its existing quality.”
In the race for AI superiority, OpenAI’s inadvertent exposure of GPT-4 to free users may be a clever ploy to stay ahead in the game. This incident, whether a glitch or a strategic move, underscores the rapidly changing dynamics of AI technology.
Edited by Ryan Ozawa.