Ascii bar chart

[amber acorn]

ChatGPT can make ASCII bar charts however although it looks qualitatively ok, ChatGPT 3.5's bar chart is quantitatively false ( for example too many bars for Q2 when compared with the bars for Q1 and Q3)

First prompt:

provide an example of table data

Second prompt:

turn that table into an ascii bar chart

[amber acorn]

I asked Bing. It produced this table:

Quarter	Sales
Q1	1000
Q2	1200
Q3	1500
Q4	1800

I asked it:

using that data, provide a bar chart using emojis in a code block

This is also false

That said it requires quite a bit of computation to scale the charts relatively to each other

[amber acorn]

I asked Bing about the emojis and it said it considered an emoji is 200 units and it rounded the number. I then asked it to use 100 units for the emoji. It made a mistake again and invented a concept of leveling up to justify its obvious mistake

[azure walrus]

/run_prompt --output_format "Use markdown tables. The "Sales($)" column has been represented using ASCII horizontal bar charts, with the length of the bar representing the sales value. The label indicating the sales value has been added to the right side of the bar" \
--content_type "json"  \
-d "{
  "Quarter":  ["Q1", "Q2", "Q3"],
  "Sales($)": [100000, 120000, 90000]
}"

[azure walrus]

update: add emojis.

/run_prompt --output_format "Use markdown tables. The "Sales($)" column has been represented using ASCII horizontal bar(:moneybag:)  charts, with the length of the bar representing the sales value. The label indicating the sales value has been added to the right side of the bar" \
--content_type "json"  \
-d "{
  "Quarter":  ["Q1", "Q2", "Q3"],
  "Sales($)": [100000, 120000, 90000]
}" 

[amber acorn]

Hi, nice addition with the markdown table. I think ChatGPT made a mistake there again as Q2 seems too big when compared to Q1 and Q3.

[azure walrus]

In the course “ChatGPT Prompt Engineering for Developers” by Isa Fulford and Andrew Ng, this phenomenon has been explained. Chatgpt cannot precisely control the number of characters and words. It seems that the minimum unit chatgpt can precisely control is sentences.

[amber acorn]

I read about the existence of that course but I did not want to want to go through the registration process on the website. In what context did they say it can not control the number of words? I do not see why a model that predicts word by word would not be able to repeat a word 5 times if prompted to do so. From my quick tests, I was able to repeat a word 3 or 5 times for example.

I think the issue with the chart is that it has to plan what one unit block in the chart should correspond to and then follow that instruction. I think it needs to think out loud to be able to feed on its own thoughts to follow a method. The most difficult part is finding a method for figuring out what a unit block should be. Then it can tell itself to distribute the blocks. I will try to give it instructions on how to calculate the value of a block.

I believe there are situations where the model would have to know in advance the answer to be able to answer correctly. Maybe you are referring to word-counting tasks that would require knowing the answer in advance. Indeed it cannot do that.

[south mason]

this may be unrelated to your tables objective, but i realized that ChatGPT also has a limit of 8 items in a matrix[] and will assume additional items after the 8th.

[amber acorn]

Hi, I am sorry what did you mean by "will assume additional....8th"

I am also not sure I understand the 8 items limit. Do you mean that a table can have no more than 8 rows ?

[south mason]

after the 8th item given. EX: A, B, C, D, E, F, G, H, I... it will recall A to H and make up additional variables like "Q". which is strange behavior since those variables were never given in the statements.

[south mason]

try giving* it a list like : Q1($, 10000), Q2($, 20000), etc.

the 9th variable is completely made up or randomized. showing there's a limit to the internal list it can recall or manage at any given time. AKA a "Byte".

i said i doubt it was related, but perhaps it is counting each variable as a separate "item" and limited to 8 overall items. try a simpler method by using the matrix. EX: Q1($, 10000), etc to print out Q1 the format and amount inside the Q variables.

[amber acorn]

Interesting I had not heard about that