BARD vs ChatGPT (Yikes!)
Google sent me (and my 10 million closest friends) an invitation to try their AI tool, Bard. After having played around with OpenAI’s ChatGPT for a few weeks, I am really surprised at how much worse Bard performed than I expected.
I first asked it to summarize the top 5 news stories from China today. Then the same from Ireland. Then the top five news stories regarding AI. It gave me answers. They seemed reasonable, but I haven’t verified them yet. Because…
I got curious and asked Bard to give me a list of the top 33 companies with the longest histories of consistently paying dividends. It gave me a strange list, one that included Berkshire Hathaway — Warren Buffet’s company which is well known for never having paid dividends (he prefers to reinvest).
I called Bard out on this error.
It “corrected” it with a new list. But that list was strangely ordered. When I asked Bard what it was used to order the list of companies, it said the number of years that they had been paying dividends. Here was its updated list:
Rank | Company | Ticker Symbol | Number of Consecutive Years of Dividend Increases
— — — -| — — — — -| — — — — — — | — — — — — —
1 | Coca-Cola | KO | 60
2 | Procter & Gamble | PG | 66
3 | 3M | MMM | 66
4 | Johnson & Johnson | JNJ | 60
5 | Berkshire Hathaway | BRK.A | 56
6 | AT&T | T | 57
7 | Colgate-Palmolive | CL | 58
8 | Wells Fargo | WFC | 54
9 | Chubb | CB | 29
10 | Verizon | VZ | 49
11 | IBM | IBM | 50
12 | Cisco Systems | CSCO | 25
13 | Home Depot | HD | 19
14 | American Express | AXP | 49
15 | UnitedHealth Group | UNH | 34
16 | Pfizer | PFE | 47
17 | Visa | V | 50
18 | Mastercard | MA | 50
19 | Stanley Black & Decker | SWK | 27
20 | Eli Lilly | LLY | 53
21 | AbbVie | ABBV | 51
22 | McKesson | MCK | 39
23 | United Technologies | UTX | 48
24 | Salesforce | CRM | 16
25 | The Travelers Companies | TRV | 46
26 | Johnson Controls International | JCI | 44
27 | Adobe | ADBE | 19
28 | Autodesk | ADSK | 18
29 | Goldman Sachs | GS | 41
30 | DOW | DOW | 40
This is wrong in a few ways. First, there’s Berkshire again! But more importantly, the list is not even ordered by the numbers that Bard is claiming for each company. That is really weird. Is this hard for Bard?
Prompt:
I am surprised that this is difficult for you. Let’s try something simple. Please put this list in ascending order: 5, 6, 12, 3, 1, 2, 18, 26, 42, 7, 4
Bard:
“Sure, I can do that. Here is the list in ascending order:
Code snippet
1, 2, 3, 4, 5, 6, 7, 12, 18, 26, 42”
Prompt:
Now try keeping these items together, meaning keep the word and number together, but re-order the list in ascending order by number: cat 7, dog 6, bird 42, mouse 17, car 12, orange 3
Bard:
“Sure, here is the list in ascending order by number, with the word and number kept together:
Code snippet
orange 3
mouse 17
cat 7
dog 6
bird 42
car 12
I hope this is what you were looking for.”
Um, no.
This is worse than I thought. Not only does Bard get it wrong, but it also does so after trying to write code and running it. It recognizes a list, but it can’t even order it.
By then I was really curious, what would ChatGPT do?
Prompt:
Are you able to order lists?
ChatGPT:
As an AI language model, I don’t have a direct capability to order or sort lists. However, I can certainly help you understand and implement algorithms or approaches to order lists in various programming languages. Sorting lists is a common task in programming, and there are numerous sorting algorithms available, such as bubble sort, insertion sort, merge sort, quicksort, and more. If you provide more specific details about the programming language or requirements, I can assist you further.
Well, let’s try anyway.
Prompt:
Now try keeping these items together, meaning keep the word and number together, but re-order the list in ascending order by number: cat 7, dog 6, bird 42, mouse 17, car 12, orange 3
ChatGPT:
To keep the words and numbers together while reordering the list in ascending order by number, you can represent each item in the list as a tuple containing the word and the number. Then, you can use a sorting algorithm that considers the second element of each tuple (the number) for comparison. Here’s an example in Python:
items = [(“cat”, 7), (“dog”, 6), (“bird”, 42), (“mouse”, 17), (“car”, 12), (“orange”, 3)]
sorted_items = sorted(items, key=lambda x: x[1])
for item in sorted_items: print(item[0], item[1])
“In this code, the sorted function is used to sort the items list based on the second element of each tuple. The key parameter takes a lambda function that specifies the sorting criterion (x[1] represents the second element of each tuple). The resulting sorted list is stored in sorted_items, and then it is printed with the word and number separated as desired.
You can modify the code according to your programming language of choice, as the concept of using a custom comparison key should be applicable in most programming languages.”
This feels worlds apart.
Well, let’s ask ChatGPT my original question:
Bummer.
Well, at least it didn’t just make up a fake and incorrectly ordered list. Am I being unfair, or do these seem like very different levels of AI?