I've noticed that GPT-3.5 unfiltered has been producing lower quality responses.

[tight warren]

After yesterday's DDOS attack that took down the servers I started to notice that GPT-3.5 unfiltered has become less compliant with the prompt. Previously the responses were spacious and contained very creative text, but now this model has become less attention to detail, and the responses have mostly become less informative. I've even occasionally had a system prompt given as an answer. but that was a single time. Has anyone had the same problem, or did I screw something up in my settings myself (though I didn't change anything.)?

[fierce crypt]

GPT-3.5 is not hosted by us; it is owned by OpenAI and hosted by them. Therefore, anything that happens to our servers won't affect that model. Also, a DDoS attack can only make the server unreachable for a specific amount of time; it can't affect the model's responses.

We are not responsible for OpenAI models; we only provide access to them.

[tight warren]

Okay, thanks for the answer.

[gloomy mantle]

How are you guys even providing GPT-3.5, is that legal

[fierce crypt]

why do you think it's illegal? looks like you still are not aware that gpt-3.5-turbo has a public API by OpenAI itself

[gloomy mantle]

really?

let me search

[fierce crypt]

note that public API not free API

[gloomy mantle]

oh I see, but then you are charging money for it and you have to pay some of it to open ai for usage?

[fierce crypt]

Of course, I also pay a few thousand of dollars for the GPU servers that host PAI, as well as $1000-$2000 per month for PAI training.

nothing is free

[gloomy mantle]

do you earn any profit from this?

[fierce crypt]

not yet, my main goal is to provide LLM for free