So with ChatGPT released in public preview at the end of last year, the ability to craft effective prompts for ChatGPT became an impressive skill. For the last few decades, masters of the Google search box have been wowing co-workers, but now acolytes of the prompt are the new hotness.
As with Google, how you ask ChatGPT a question greatly determines the quality and usefulness of the reponse returned. To that end, here are some of things you can ask ChatGPT. Each type of query can be tuned to get better responses, by refining the prompt.
Asking a question with a clear answer is really where Google shined, so ChatGPT needs to nail this one. So far, on questions with a clear answer, ChatGPT seems to be doing well, correctly identifying the musician, artist, and actor Guy Davises are all more talented than the nerdy computer one. :)
This is perhaps the most impressive feature of ChatGPT, asking it to create new text given a topic and a style. For example, asking it to write a short story in the style of Stephen King on the topic of werewolves.
Asking for a step-by-step guide to completing a task is another common type of prompt. Examples include asking for a recipe to bake a cake:
Finally, you can ask ChatGPT for a set of suggestions to a problem you are facing. Examples include asking for 3 ways to request a raise at work.
A less charitable read on this is that ChatGPT can sometimes make up compelling bullshit. The response may sound right, but on closer inspection a domain expert can see the response is simply not true. This is sometimes known as AI hallucination.
Given human-created inputs any AI system will potentially be affected by bias, simply reflecting the bias cataloged in the world. Not unique to ChatGPT, but rather any AI system trained on data not screened for bias in advance will show this problem.
While impressive, the ChatGPT model (free version) does have some limitations such as currency. It was trained on a corpus from a year ago so current events are beyond its purview:
Overall, ChatGPT is really an impressive technology and will definitely give Google a run for their money. I’ve been noticing that the top results in my recent Google searches have been declining in quality for years now, often clogged with blogspam such as above. :) With Chat interfaces improving and providing adequate summaries and responses, will users bother clicking through to original sources anymore?