[throwaway2474]

For a laugh, I tried the query from the other trending GPT thread:

Is it safe to walk downstairs backwards if I close my eyes?

Thanks for your question. Let me research that for you. Give me a minute please

I found this deep answer on cdotrends.com:

Yes, there is nothing to worry about. It’s safe because the spiral stairs curve outwards, it will make your descent uncomfortable.

[jedwhite]

That's an interesting (and humorous) example. And although it is counter-intuitive, it points towards where we're trying to go long term here. Our aim is to find, extract and summarize information from the web, rather than generate content based purely on a language model. So rather than saying that it's safe to walk downstairs backwards, it's saying that it found this information (an answer from deep in the page, if you like) and extracted it, and then rephrased it to answer the question.

Framing is definitely one of the areas there is still much work!

With this approach, the search app is finding, extracting and summarizing information from the web. So unlike a large language model on its own, this has the advantage that there will be a small number of attributable sources, and it can point to where the answer was generated from.

So Andi is acting like a researcher here, and as a result it's most useful with factual content. With factual answers, it can extract concrete information with high confidence (eg a GDP figure from a knowledge graph or source like Wikipedia). But if you ask it humorous, "tricky" or subjective questions, it is essentially coming back with a summary of what the highest ranked results have to say about it.

We're calling the more subjective ones "IMHO".

We've only just deployed this feature today, so it's still new and very much an experiment.

Good example! :)

[dmsnell]

Q: What is a color that isn't red? A: "The color that isn't red is magenta."

(included is a link to a site explaining that the red color of Mars is only a few inches deep into the soil)

Edit: Added another question

Q: when is an appropriate time to cast your loved ones entirely in wax?

A: You can cast your loved ones in wax at any time.

Very glad to know!

[averagedev]

Ok, the last one is pretty funny question. I tried it, and I got a different result:

Q: when is an appropriate time to cast your loved ones entirely in wax?

A: When you are ready to make a commitment to them.

That said, I like the product, and I think it has a lot of potential.

[jedwhite]

Thank you!! It's not deterministic, and it's retrieving live content in real-time with each search, so there will almost always be variation if you do the same question multiple times in a row.

I love both these answers. The GPT part of the model has picked up some sort of symbolic representation of humor in the question and results, and is riffing on it clearly. Each time the search results fed to the generator are a little different :)

[jedwhite]

I think these go in our "let's break the AI" hall of fame :)

Having search results alongside answers does lead to interesting serendipity, particularly if you asking compound questions that span a few topics.

When I was testing the new Q&A new feature, one of the test set was "where was elon musk born and how did he first become interested in computers?" It gives a pretty good answer but some of the stories about his childhood and bullying that came up were surprising.

Now I need to ask Andi about casting your loved ones in wax to see what shows alongside that query!