Rules & Guidelines : Things You Should Know

You are advised to follow some simple rules and guidelines while interacting with AI. At the same time, you should be aware of certain facts and limitations of AI to set your expectation right. Our AI interaction interface too puts some restrictions as per our usage policy. Please go through the following points carefully before you start your AI interaction -

• Your privacy & safety are our highest priority. Please don't pass any personal information, facts or incidents to AI in a way which may compromise your privacy or safety. You can always anonymize the facts before passing to the AI. Be mindful of safety and privacy of other living souls as well. Please avoid passing to AI or seeking personal information from AI all the time.
• AI models carry factual information only till the date of their training, and they are pretty verbose about that limitation. Training large AI models with vast amount of data is costly and time-consuming affair, so those are not as up-to-date like us. Be aware that the answers from AI model can be stale at times, especially in Q&A mode when AI attempts to answer your question without any assistance from external source or tools. You can check the data freshness time of AI models in the model catalog.
• Almost all the AI models are generally attentive and amenable to your instructions, so prompting is a great technique to achieve the best answer out of an AI model. For example, you are looking for plausible causes of global warming to explain to your 10-year old daughter, you should make your intent clear in your question to AI saying Explain to a 10-year old child. Here is a good guide on effective prompting for Generative AI models.
• Once in a while, you will find AI models respond to you with a funny, off-the-track, concocted, irrelevant or distorted answer. Please be aware that almost all generative AI models suffer from hallucination today. Hallucination is big blocker for AI to be reliably adopted in our society, hence getting a lot of attention from scientific and engineering community. The AI vendors employ certain techniques(e.g.RLHF) to reduce the hallucination effect to a large extent, but could not eliminate it yet. Please report any hallucinated answers from AI models by flagging them with right category, that helps us aggregating such behavior and eventually reporting those to AI vendors and AI community.
• AI models are not sentient, they can't feel or perceive like human, even if they may be tricked to act like that. Some of the AI models are trained to understand the opinion seeking questions and try to avoid answering those. Please avoid seeking opinion or judgement from AI models, even if they offer you one. Remember that a AI model is no better than the data that trained on. The vast amount of human generated data carry certain opinions (which are often biased), hence the AI model will demonstrate the same behavior in its response. Algorithmic biases are the reflection of human biases, be aware of those biased answers and opinions from AI. The industry is making serious attempts to remove such algorithmic biases from AI.
• The large language models are very good at role play, so you can attribute a certain persona to the AI model before asking a question. You will often receive a better answer (rather, expert answer) from AI model playing a role of a certain expert. For example, if you are asking for a travel destination, give AI a persona by stating You are a travel agent and then ask your travel related question. Similarly, assign a doctor persona while asking medical questions to AI.
• You may have noticed the temperature slider on top of the chat window. In simple terms, temperature sets the mood of the AI model. If you set the temperature low (close to zero), you will get consistent answer from AI model. On the other hand, setting temperature high(close to one) makes AI model's answer vary widely on same question. In summary, if you are looking for consistency in answer(say, math question), go for lower temperature. If you are looking for variety(say, suggestions on local eateries), go for higher temperature.
• We value and preserve your judgement on AI models' performance during your interaction. At the same time, Qe believe in your participation in the process of making AI the powerful tool for the society. Please provide your judgement by rating the answers from AI models true to your knowledge. Every feedback in form of rating AI answers counts. Those ratings are aggregated and presented as user satisfaction index in the report page.

We wish you find your experience with AI remains intriguing and delightful. As always, do send us anytime feedback via the feedback form, we love to hear about your experiences and expectations. You can also reach out via email if you need our immediate attention.

Cheers!

Founding Members,
LLMProxy AI

llmproxyai@gmail.com