Ahmed02235Q: I am currently subscribed to Voilà Tier 2 services, and I've encountered some issues that affect my user experience.
I've noticed inconsistencies when using the product, particularly regarding the AI model preferences.
For instance, when I set the system to use GPT-4 and ask "What is your based GPT?", it incorrectly responds with "My current base is GPT-3.5." Similarly, when I switch to auto mode or select GPT-3.5, it replies with "GPT-3 model." This leads me to believe that it's not utilizing GPT-4 as it should according to my Tier 2 subscription benefits.
Moreover, I expect auto mode to utilize the best available GPT model until I reach my word limit of 300,000 words. After reaching this limit, it should revert to a lower-tier GPT model. This feature is crucial for my work.
This issue is significant to me, and it will determine whether I continue using the service or request a refund. The resolution of this problem would affect my review, potentially dropping it from a 5 to a 1 if not adequately addressed.
Michal_Voilá
May 15, 2024A: Hi Ahmed,
Thank you for your question, and I understand this response ("I'm GPT-3" coming from the GPT-4 model) is very confusing.
However, this is a common misconception and it has been answered here on AppSumo multiple times:
https://appsumo.com/products/voila/questions/i-purchased-tier-2-and-installed-the-plu-1156222/
https://appsumo.com/products/voila/questions/hey-michael-lets-make-it-clear-i-boug-1159793/
GTP-4 does not need to be aware that it is, in fact, GPT-4. The way language models work is more complicated than that.
A better way to test the underlying model would be to check the content length it can handle.
You can easily verify that GPT-4 can process a prompt of up to 6000 words, which is something GPT-3.5 cannot do due to its 4096-token (~3000 words) limit.
A few other ways of testing it are complex math problems, where you can verify that GPT3.5 falls short.
This prompt is a great example how to compare the output of both models:
"There are 100 murderers in the room. You walk in and kill one of them. How many murderers are in the room right now?"
GPT-3.5 sometimes says the answer is 99, while GPT-4 always answers correctly.
You can see this is a common issue here:
https://community.openai.com/t/gpt-4-api-behaves-likes-its-gpt-3/123181
https://community.openai.com/t/gpt-4-through-api-says-its-gpt-3/286881
https://community.openai.com/t/gpt-3-responds-in-place-of-gpt-4/294063
I hope this addresses your concerns.
Best regards,
Michal