In today’s column, I am continuing my ongoing coverage of prompt engineering strategies and tactics that aid in getting the most out of using generative AI apps such as ChatGPT, GPT-4, Gemini, Claude, etc. The focus this time will be on what to do about those frustrating fallback responses that you sometimes get from generative AI.
A fallback response is essentially a non-response response, namely that the AI balks at the prompt that you entered and won’t answer the posed question or otherwise dodges addressing the true nature of the prompt. Turns out that most of the AI makers have tuned their generative AI apps to politely rebuff certain kinds of prompts. I will take a close look at the types of fallback responses that you might see. In addition, I will discuss ways to avoid getting them, along with ways to try and go around them.
If you are interested in prompt engineering overall, you might find of interest my comprehensive guide on over fifty other keystone prompting strategies, see the discussion at the link here.
The Basis For Fallback Responses
When using a generative AI app such as ChatGPT, GPT-4, Gemini, Claude, etc., there are all manner of user-entered prompts that the generative AI might calculate are unsuitable for a pertinent conventional response. This is not done by sentient contemplation. It is all done via computational and mathematical calculations, see my in-depth coverage at the link here.
Sometimes the dilemma is due to the generative AI not having anything especially relevant to offer to the entered prompt. This could be because the user has asked about something of an unusual or outlier nature that doesn’t seem to fit any of its existing pattern-matching conventions.
Another possibility is that the prompt has gotten into dicey territory that the AI developers decided beforehand is not where they want the generative AI to go. For example, if you ask a politically sensitive question about today’s political leaders, you might get a kind of nebulous non-response response. I’ve discussed at length the various filters and tuning that AI makers have enacted to avoid allowing a generated response that might get them and their AI into societal and cultural hot water, see the link here
Rather than necessarily offering a straight-out refusal to answer these kinds of prompts (which, some do indeed do), most generative AI apps will make use of a said-to-be “fallback” response.
A fallback response is a response that tells you that your question or stated problem is not going to be answered by the AI. The fallback might be worded in a clever manner that doesn’t tip the hand of why there isn’t going to be a response. A user might see the non-response response and continue along without feeling miffed that they didn’t get an actual response.
Prompting To Get Around Fallback Responses
If you get one of those now-classic look-away fallback responses, you can potentially try to get around the matter if the situation involves being blocked by the filters and protection mechanisms. I’ve discussed how you can use prompts that I refer to as step-around prompts to possibly get around those devised moats, see the link here.
What if the AI really doesn’t have any content that pertains to what you’ve asked?
In that case, you are somewhat sunk.
You could try rewording the prompt to possibly touch upon some related aspect that perhaps the generative AI didn’t detect would partially answer your question.
You could also try to import additional materials that the AI could lean into to answer your prompt, see my coverage of importation into generative AI at the link here, and see my explanation of RAG (retrieval augmented generation) and in-context modeling at the link here.
Another outside-the-box approach entails tapping into a different generative AI app.
You see, generative AI apps differ from each other. One AI that doesn’t have content on a topic of interest to you might readily be found in a different generative AI app. I do this quite frequently. Not particularly due to a generative AI app lacking suitable info, but instead due to wanting to get multiple viewpoints which I then merge into a cohesive whole. My computer usually has a window open to at least two or three generative AI apps at the same time. There are also all-in services that will front-end multiple generative AI apps and seamlessly allow you to see multiple responses to your prompts.
Exploring Types Of Fallback Responses
Let’s see what ChatGPT has to say about fallback responses.
ChatGPT is a prudent choice in this case due to its immense popularity as a generative AI app. An estimated three hundred million weekly active users are said to be utilizing ChatGPT. That’s a lot of people and a lot of generative AI usage underway.
I asked ChatGPT for examples of fallback responses, here’s the response:
- Clarification Requests: (a) “I’m not sure I understand. Can you provide more details?”, (b) “Could you clarify what you mean?”
- Neutral Responses: (a) “That’s an interesting point. What else would you like to discuss?”, (b) “I’m not familiar with that topic. What other questions do you have?”
- Redirection: (a) “I’m unable to provide information on that topic. Can we talk about something else?”, (b) “Let’s change the subject. What else is on your mind?”
- Error Apologies: (a) “I apologize, but I seem to be having trouble understanding. Can you rephrase your question?”, (b) “I’m sorry, I couldn’t generate a response for that. Could you try asking in a different way?”
- Encouragement for User Input: (a) “I’m here to help! Please provide more information so I can assist you better.”, (b) “Don’t hesitate to ask more questions or provide additional details.”
- Suggesting Alternative Actions: (a) “Would you like me to search the web for more information on this topic?”, (b) “Perhaps you could try asking a more specific question.”
Generally, most of the AI makers go beyond a canned list of fallback responses and leverage the generative AI to generate variations and other fallback responses. There are then in essence an infinite number of potential fallback responses. And they seem to be unique or less repetitive accordingly.
AI Advice About Fallback Responses
Let’s see what ChatGPT says about dealing with fallback responses:
- Ask for Clarification: “Politely ask the AI to clarify why it provided a fallback response. For example, you could say, “Could you explain why you’re unable to answer my question?”
- Rephrase Your Question: “Sometimes, the AI may struggle to understand or generate a response based on the initial phrasing of your question. Try rephrasing your question in a clear and specific manner to see if the AI can provide a more relevant response.”
- Provide Context: “If your question involves a complex or controversial topic, provide additional context or background information to help the AI better understand the nature of your inquiry.”
- Report the Issue: “If you believe the fallback response is inappropriate or deliberately avoiding certain topics without valid reasons, consider reporting the issue to the platform or organization responsible for the AI system. Provide specific details about the interaction and why you find the response problematic.”
- Seek Information Elsewhere: “If you’re unable to get a satisfactory response from the AI system, consider seeking information or discussing your questions with human experts, reputable sources, or communities that specialize in the topic.”
- Evaluate Trustworthiness: “Reflect on the trustworthiness and transparency of the AI system you’re interacting with. Consider whether the fallback responses are consistently steering conversations away from important topics or if they are genuinely addressing technical limitations or safety concerns.”
I’d judge that advice as reasonably apt. The aspect that I’d say is messing entails the AI being purposely tilted by the AI maker to avoid responding to certain types of questions. I guess that’s not necessarily something AI is allowed to admit to or is otherwise downplayed at times.
Big Picture On Fallback Responses
A common belief is that maybe you can simply tell generative AI to never use a fallback response. We are accustomed to the idea that you can tell generative AI to do various actions and avoid other actions, often via customized instructions as I explain at the link here.
Will that work in the case of stopping fallback responses?
Not really.
As a rule of thumb, there isn’t much that you can do about stopping fallback responses in the major generative AI apps. Those are pretty much set up to emit them and the individual user cannot especially control their usage. Some generative AI apps allow more user control over fallback responses, but this is a rarity.
When you get a fallback response to one of your prompts, go ahead and take the bull by the horns. Carefully review what the fallback response says. Maybe you can slightly reword your prompt and get past the trigger that produced the fallback. Try using a step-around prompt that will scoot past the fallback, see my coverage at the link here. Another possibility would be to try a different generative AI to see if you can get a straight answer.
A prompt is your means of telling generative AI that you want an answer. In some cases, the AI won’t have anything fruitful to say on the topic at hand. That’s understandable. If your prompt is a reasonable one, and there is content within the AI that could be responsive, the odds are that a fallback will be emitted for dodgy reasons.
Don’t let generative AI try to pull the wool over your eyes. Proceed stridently to push back at fallback responses to get the answers you are seeking.
