Perhaps…
… I was a little hasty in getting this week’s SRS post out.
Regular Reader Unknown wrote in the comments that:
I’ve found Bard useful for suggesting lines of enquiry. For instance, I asked “When did people in China first realise dolphins are not fish?” and it suggested some ancient Chinese texts which apparently discuss dolphins. I don’t know if this info is accurate, but at least it’s a good starting point. ChatGPT couldn’t help with this query, and conventional searching didn’t get me very far. I also asked Bard “Which art historians believe that Johannes Vermeer painted St Praxedis, and which are skeptical?” and it gave me two lists of names – again a good place to start.
I like this use of LLMs quite a bit–as a great research starter. This is a great advanced use of an LLM resource.
And I’d missed cmarlowe‘s earlier comment that:
As a retired librarian I use LLMs to provide a basic understanding of new topic or to provide further insights into a topic I am researching and have some knowledge of. One of the features that helps me in researching is the conversations, because it requires me to think what else I should consider in my reference query. But always having that librarian mindset, I do check multiple search tools since often one finds different and helpful information to complete one’s research query and these can include general search engines or research sites. This also helps to avoid misinformation from LLM hallucination by comparing one to another. One unique LLM tool is Poe https://poe.com/ where you can search multiple AI sites and tools.
Both Unknown and cmarlowe are using LLMs in a sophisticated way–they’re using them as lead generators. That is, the LLM gives them ideas about where to look and as useful resources.
*
On the other hand, my friend Aditi sent me three examples where she used ChatGPT to answer questions that would have taken a lot longer with traditional search, and one where it’s unclear that ANY search engine would have been able to do anything. This in a nice example of using LLMs to answer questions directly.
Her description of her truly difficult search task was this:
“[I wanted to] Find out why I remember an old Telugu nursery rhyme my grandma used to sing. I tried so hard with Google all these years and chatGPT helped me finally understand why I loved it and remembered it all these years.”
Here’s their conversation of the first very difficult search task:
Aditi: I’m trying to figure out what an old Telugu nursery rhyme my grandma used to sing to me means. I’m not sure I remember the words right, but if I transcribe it to the best of my memory, might you be able to help me figure it out?
ChatGPT: Absolutely, I’d be happy to help. Please go ahead and transcribe the nursery rhyme as best as you remember it.
Aditi: Ok here it is. I remember almost all of it clearly except for the fruit parts, which I’ve enclosed like (this?). I say fruit because it’s about which bird eats which fruit I think?
Gaagi, gaganaagi
Ganganagi intilo paalupandi
Aa pandu thecchi chilakakicchi
Chilaka thinna pandu (cheedi?) pandu, Hamsa thinna pandu (arati?) pandu, [Baby’s name] thinna pandu? Bangaaaara pandu!!
ChatGPT: Thank you for sharing the nursery rhyme. It’s a charming Telugu rhyme about different birds and the fruits they eat, with a sweet twist at the end involving the baby’s name. Here’s a rough translation and interpretation:
“Gaagi, gaganaagi
Ganganagi intilo paalupandi”
This part sets a playful and melodic tone, akin to calling everyone to gather around, possibly imitating the sound of birds.
“Aa pandu thecchi chilakakicchi Chilaka thinna pandu (cheedi?) pandu,”
It talks about bringing a fruit to a parrot (chilaka). The fruit could be “(cheedi?) pandu,” where “cheedi” typically means a berry. The exact fruit is uncertain, but it’s something the parrot enjoys.
“Hamsa thinna pandu (arati?) pandu,”
This line is about a swan (hamsa) eating a fruit, possibly “(arati?) pandu,” which could refer to a banana in Telugu).
“[Baby’s name] thinna pandu?
Bangaaaara pandu!!”
This is the delightful twist where the baby’s name is inserted, asking what fruit the baby eats, followed by “Bangaaaara pandu,” meaning ‘golden fruit’, symbolizing that the child is precious and loved, like a rare and valuable fruit.
It’s a nursery rhyme that combines the joy of nature (birds and fruits) with the delight in a
beloved child, turning a simple rhyme into a celebration of the child’s presence. The uncertainty in the fruits adds a bit of mystery and room for imagination, which is often a charming aspect of traditional nursery rhymes.
This is pretty impressive. Bear in mind that Aditi transliterated the lyrics of the song she mostly remembered into a English-based transcription. (Normally, Telugu is written in Telugu script, which looks like this: తెలుగు. That’s “Telugu” in Telugu characters.) And yet, based on her approximation of the Telugu lyrics, it somehow found the song and was able to find an English parallel translation.
Both Aditi and I did regular searches to try and find such a resource, but we both failed. There’s a bit of knowledge buried deep in ChatGPT that includes Telugu nursery rhymes.
And, in the spirit of double-checking all results (especially surprising results like this), Aditi checked in with her relatives in India and confirmed that what ChatGPT answered is, in fact, correct!
*
In another search task, Aditi was searching for something to do with her young family on a cold day in Denver:
Aditi: What’s something we can do as a family in Denver right now? It’s cold so I was thinking something indoor and fun for a baby
ChatGPT: Visiting the Denver Aquarium could be a delightful experience. Imagine it as a serene underwater world, a tapestry of vibrant marine life, captivating for both you and your young daughter. It’s a warm, inviting space, away from the cold, where your baby can marvel at the wonders of aquatic life, creating a tapestry of early, colorful memories.
Aditi: Nice! Any other ideas?
ChatGPT: Consider the Denver Children’s Museum. It’s like a playground of learning and discovery, tailored for little explorers. Your baby would find a world scaled to her size, filled with interactive exhibits that engage the senses and ignite curiosity, while you could enjoy watching her discoveries and growth in such a playful environment.
The responses are a little wordy, but they have the advantage of being high quality, and the disadvantage of being just one response.
Note that Aditi used ChatGPT’s “custom instructions” feature to tell ChatGPT ahead of time who she was, about her family, and what kinds of responses she prefers. (That’s why the system knew that her family includes a baby.)
This is a great way to tune ChatGPT to become more like the information resource that you’d like to use. Aditi gave custom instructions saying that she has a graduate degree in information sciences, and that the replies it gives should be concise and accurate. This has the effect of tuning the responses to be more like what you need. (I’ll write more about this in the future–very handy.)
