Daily Study Group: The Wolfram Plugin for ChatGPT
Q&A from April 18, 2023

ARE THE NOTEBOOKS DOWNLOADABLE?

It is a cloud published notebook so you should have immediate access: https://wolfr.am/1cKo4X0ko. Yes you can download it as well.

COULD YOU DEMONSTRATE THE PLUG-IN GENERATE A HIGH-RES PLOT OF A FRACTAL, SUCH AS THE MANDELBROT SET, OR ANOTHER FAMOUS FRACTAL?

Not a high-res image but you can get visual output back using the Wolfram plugin and the following prompt: visualize the mandelbrot set

PLEASE, POST THE LINKS THAT THE SPEAKER SHOWED AT THE BEGINNING OF TODAY'S PRESENTATION (WOLFRAM AI IN THE REPOSITORY PAGE, ETC)

OpenAILink: https://resources.wolframcloud.com/PacletRepository/resources/ChristopherWolfram/OpenAILink/
AIAssistant: https://resources.wolframcloud.com/FunctionRepository/resources/AIAssistant/

I AM AN CHATGPT PLUS MEMBER, BUT I DO NOT SEE THE PLUGINS AS YOU ARE SHOWING ON YOUR PRESENTATIONS

Access to the plugin may not be immediate, so you may want to add yourself to the waiting list: https://openai.com/blog/chatgpt-plugins
The plugins model is only available to a limited number of "Plus" users right now — we aren't in control of that rollout, but OpenAI is working on opening things up to more users.

HOW DO YOU GET CHATAI TO DISPLAY UNITS IN AN EQUATION, SUCH AS METRIC?

Try through very explicit prompting.

DO YOU NEED TO BE A CHATGPT PLUS MEMBER TO USE THE PLUG IN?

Yes, as of now

NOT A QUESTION, BUT I HAVE FOUND IF YOU CLICK THE PEN AND EDIT THE ORIGINAL PROMPT AND THEN SAVE AND REGENERATE, THAT TENDS TO WORK BETTER THAN JUST HITTING THE REGENERATE BUTTON.

Thanks for sharing!
Agreed. Sometimes this works, for some reason presently testing my inputs again, that doesn't seem to be working either unfortunately.

*REGARDLESS OF CHATGPT* LOTS OF DATA (AT LEAST IN THE ECONOMIC DOMAIN) IN WOLFRAM ALPHA ENDS IN 2021 OR EVEN 2020. WHY IS IT SO OLD?

It depends on the data source, generally (and to some degree the Wolfram Alpha release cycle). It's not unusual for the latest available normalized, clean, large-scale international or subnational socioeconomic data to be a year or two old, and generally we favor sources that provide comparable data across all countries/regions at the same level, rather than piecemeal filling in data points from individual countries/states as they become available. If you have specific areas where you think we're definitely missing more recent data, please file feedback through the Wolfram Alpha website.
Thanks a lot Alan. I'm not looking for some obscure data. For example: inflation data from Brazil (quarterly produced by the National Stats Institute and the latest is from last month). I type "inflation data brazil" in Wolfram Alpha and the latest is from 2021. It even shows in the World Bank or UN website repositories. I type population Madrid and gives me a 2007 estimate! You should check the sources and update them more often

DO THE WOLFRAM OPENAI PACLET AND RESOURCE FUNCTION ALLOW TO TRACK THE NUMBER OF TOKENS SENT AND COST INCURRED FROM USING THE API ?

We will have Christopher Wolfram, developer of the OpenAILink paclet present on Thursday. We will forward your question to him.

BY USING THE PLUGIN, ARE THERE STILL INSTANCES WHEN CHATGPT STILL HALLUCINATE ITS ANSWERS?

ChatGPT will take the results from the plugin, but it will still do its won reasoning around and this can contain hallucinations.

IT IS POSSIBLE TO GENERATE A PLOT 3D?

yes

WHY ASK CHATGPT ABOUT FIBONACCI NUMBERS. IS IT , FOR EXAMPLE, BECAUSE: 1. CHATGPT IS MORE NATURALLY TALKING TO WOLFRAM - THUS, NO NEED TO LEARN WL IN SOME FUTURE? 2. CHATGPT WILL, IN ADDITION, OUTPUT A NICELY COMPOSED DOCUMENT.

Fibonacci numbers example could be seen as an example of “boilerplate work” Stephen Wolfram mentions in his post (https://writings.stephenwolfram.com/2023/03/chatgpt-gets-its-wolfram-superpowers/ )
Also referring to the post:
“ChatGPT can work with Wolfram Language, in a sense building up computational ideas just using natural language. And part of what’s then critical is that Wolfram Language can directly represent the kinds of things we want to talk about. But what’s also critical is that it gives us a way to “know what we have”—because we can realistically and economically read Wolfram Language code that ChatGPT has generated.”

DOES CHATGPT GENERATE INCORRECT ANSWERS EQUALLY FREQUENTLY WHEN USING THE WOLFRAM ALPHA PLUGIN VS. WHEN NOT USING IT?

Depends on the question. For computational ones the answer is yes. Say for fashion or dining out the answer is no.
Due to the nature of large language models, using the Wolfram Plugin enables it to hallucinate facts and computational results significantly less often. When it recognizes to make a call to the Wolfram Plugin, and successfully makes said call, it has the backing of the Wolfram knowledge base and code to be correct.

HOW DO THE ERRORS CHATGPT MAKES IN ITS WL CALLS COMPARE WITH THE ERRORS MADE BY TYPICAL NAIVE HUMAN USERS?

In my experience: quite similar

CAN I USE MATHEMATICA COMMANDS, FOR EXAMPLE, PLOT3D[]?

Yes, you can both give explicit WL code to run and ask it to give usage statements for WL functions.

'@ABRITA: UNLESS IT'S ME, THE LINKS YOU ARE PROVIDING DON'T WORK. MY QUESTIONS REMAIN, HOW DOES CHATGPT ENRICH WOLFRAM? AND WHY DATA IN THE WOLFRAM ALPHA APP IS SO OLD?

Sorry to hear that Jose. We will add clickable links to the cloud notebook https://wolfr.am/1cKo4X0ko as well
Thanks very much Abrita :)

IS IT POSSIBLE TO PULL IN AN INTERACTIVE IMAGE?

The ChatGPT interface does not support an interactive image right now but you can get the code to generate an interactive image within a notebook

WHEN YOU ASK CHATGPT TO PLOT A FUNCTION OF TWO VARIABLES, DO YOU KNOW HOW IT CHOOSES THE DOMAIN OF THE PLOT?

Wolfram alpha has plot range finding abilities, so when ChatGPT calls the wolfram plugin, it will inherit the generated plot ranges

CAN IT SOLVE (PLANE, EUCLIDEAN) GEOMETRY PROBLEMS ?

In a limited amount according to my experience.

HOW ABOUT SOLVING A LINEAR SYSTEM WHICH HAS INFINITELY MANY SOLUTION ? HOW ABUT BRINGING A MATRIX TO ITS CANONICAL/JORDAN FORM ?

both will work

COULD YOU PLEASE TRY TO USE WOLFRAM FOR ALGEBRAIC MANIPULATIONS? I HAVE A CONCRETE DETAILED EXAMPLE: COMPUTE THE KILLING VECTOR FIELD FOR A CAUSAL DIAMOND IN A DE SITTER BACKGROUND. IT WILL HAVE TO USE ITS KNOWLEDGE OF GR BUT CALL MATHEMATICA FOR TENSORIAL CALCULATIONS AND SOLVING DIFFERENTIAL EQUATIONS IN A CURVED SPACETIME. IN PRINCIPLE, IT SHOULD BE ABLE TO DO THAT.

Its not hopeless to do this, but it will need some hand-holding. Computations of this level are IMO still better than in Mathematica with xAct. But I did succeed today ChatGPT giving me the equations of motion derived from a Born-Infeld action.
I see. So maybe I don't fully understand, but ˜I did think the plugin would have the same capabilities as Mathematica? Or maybe that's in the future?

ON THIS LEVEL OF INPUT THERE IS NO OR VERY LITTLE DIFFERENCE TO THE PERFORMANCE OF WOLFRAM ALPHA ITSELF. WHAT ABOUT SOME SYNTHESIS?

I tried this for fun:
Write a poem about the 5 tallest buildings on earth and include their heights and geolocation as well. Include a clearly labeled plot of the heights
Request to Wolfram:
{"input":"5 tallest buildings in the world"}
Here is the poem from ChatGPT:
Ode to the Giants of the Sky
​
In Dubai, where the desert sands do lie,
Stands the Burj Khalifa, reaching for the sky.
Its towering height, two thousand seven seventeen,
A marvel of the world, a sight to be seen.
​
In Tokyo, where cherry blossoms bloom,
The Skytree stands tall, a beacon in the gloom.
Two thousand and eighty feet, its height so grand,
A symbol of Japan, across the land.
​
In Shanghai, where the river winds its way,
The Shanghai Tower, a beacon by the bay.
Two thousand and seventy-three, its stature great,
A testament to progress, a symbol of the state.
​
In Guangzhou, where the Pearl River flows,
The Canton Tower, above the city it glows.
One thousand nine eighty-two, its height so bold,
A story of ambition, a tale to be told.
​
In Makkah, where the faithful come to pray,
The Clock Royal Tower, a guide by night and day.
One thousand nine seventy-two, its height so pure,
A beacon of devotion, forever to endure.
​
While i did not get the actual chart, I got back the code to create it:
In[]:=
buildingNames={"Burj Khalifa","Tokyo Skytree","Shanghai Tower","Canton Tower","Makkah Clock Royal Tower"};​​buildingHeights={2717,2080,2073,1982,1972};​​BarChart[buildingHeights,ChartLabels->buildingNames,AxesLabel->{"Building","Height (ft)"},PlotLabel->"Heights of the Tallest Buildings in the World"]​​
Out[]=

HOW DOES IT FARE WITH EXPLORATORY DATA ANALYSIS?

I tried the following:
can you provide exploratory data analysis of the Abalone data set in the UCI machine learning repository
From ChatGPT:
As an AI language model, I don’t have direct access to external data sources such as the UCI Machine Learning Repository. However, I can guide you through the process of performing exploratory data analysis (EDA) on the Abalone dataset from the UCI Machine Learning Repository.
​
The Abalone dataset contains information about physical measurements of abalone (a type of marine snail). The dataset includes the following attributes:
​
Sex: Categorical variable with three categories: M (male), F (female), and I (infant).
Length: Continuous variable representing the longest shell measurement (in mm).
Diameter: Continuous variable representing the measurement perpendicular to length (in mm).
Height: Continuous variable representing the height of the shell (in mm).
Whole weight: Continuous variable representing the whole abalone weight (in grams).
Shucked weight: Continuous variable representing the weight of the meat (in grams).
Viscera weight: Continuous variable representing the gut weight (after bleeding) (in grams).
Shell weight: Continuous variable representing the weight of the shell (after being dried) (in grams).
Rings: Integer variable representing the number of rings (which can be used to estimate the age of the abalone).
​
The EDA process typically involves the following steps:
​
Loading the dataset.
Checking for missing values and handling them if necessary.
Examining the summary statistics of the dataset.
Visualizing the distribution of each variable.
Analyzing the relationships between variables using scatter plots, correlation matrices, etc.
​
Let’s start by loading the Abalone dataset and performing some basic EDA using Wolfram Language code:
1.Load the dataset into a Wolfram Language Dataset object:
In[]:=
url="https://archive.ics.uci.edu/ml/machine-learning-databases/abalone/abalone.data";​​columnNames={"Sex","Length","Diameter","Height","WholeWeight","ShuckedWeight","VisceraWeight","ShellWeight","Rings"};​​abaloneData=Dataset[AssociationThread[columnNames->#]&/@Import[url]]
Out[]=
Sex
Length
Diameter
Height
WholeWeight
ShuckedWeight
VisceraWeight
ShellWeight
Rings
M
0.455
0.365
0.095
0.514
0.2245
0.101
0.15
15
M
0.35
0.265
0.09
0.2255
0.0995
0.0485
0.07
7
F
0.53
0.42
0.135
0.677
0.2565
0.1415
0.21
9
M
0.44
0.365
0.125
0.516
0.2155
0.114
0.155
10
I
0.33
0.255
0.08
0.205
0.0895
0.0395
0.055
7
I
0.425
0.3
0.095
0.3515
0.141
0.0775
0.12
8
F
0.53
0.415
0.15
0.7775
0.237
0.1415
0.33
20
F
0.545
0.425
0.125
0.768
0.294
0.1495
0.26
16
M
0.475
0.37
0.125
0.5095
0.2165
0.1125
0.165
9
F
0.55
0.44
0.15
0.8945
0.3145
0.151
0.32
19
F
0.525
0.38
0.14
0.6065
0.194
0.1475
0.21
14
M
0.43
0.35
0.11
0.406
0.1675
0.081
0.135
10
M
0.49
0.38
0.135
0.5415
0.2175
0.095
0.19
11
F
0.535
0.405
0.145
0.6845
0.2725
0.171
0.205
10
F
0.47
0.355
0.1
0.4755
0.1675
0.0805
0.185
10
M
0.5
0.4
0.13
0.6645
0.258
0.133
0.24
12
I
0.355
0.28
0.085
0.2905
0.095
0.0395
0.115
7
F
0.44
0.34
0.1
0.451
0.188
0.087
0.13
10
M
0.365
0.295
0.08
0.2555
0.097
0.043
0.1
7
M
0.45
0.32
0.1
0.381
0.1705
0.075
0.115
9
rows 1–20 of
4177
Data not saved. Save now
2. Display basic information about the dataset, such as the number of rows and columns, column names, and data types:
3. Calculate summary statistics for each numerical column, such as mean, median, and standard deviation:
The above code did not work so I tweaked it.
Needed to tweak the code:
Needed to tweak the code:

I TRIED GPT4 TO SOLVE A SUDOKU INSTANCE, AND IT FAILED. WOULD USING THE PLUGIN MAKES IT WORK?

There is the SudokuSolve function in the function repository: https://resources.wolframcloud.com/FunctionRepository/resources/SudokuSolve/
Prompt: use the SudokuSolve function at https://resources.wolframcloud.com/FunctionRepository/resources/SudokuSolve/ to solve the Sudoku instance {{ , , 8, 1, 7, 6, , 2,}, { , 4, , , , 9, 7, ,}, { , , , , , , , ,}, { , 7, 1, 8, , , , ,}, { , , , , , , 6, 7, 9}, { , , , 5, 9, 7, , 8, 1}, { , , , , 2, , , ,}, { , 5, 9, , , , , 1,}, { , 1, 4, 6, 5, , 3, ,}}

CAN YOU FIND THE EQUATION OF THE FUNCTION IN WL AS FINDPEAKS?

how can you find the peaks of a function
From ChatGPT:
To find the peaks (local maxima) of a function in Wolfram Language, you can use a combination of the Solve or NSolve function and the D (derivative) function. The basic idea is to find the critical points of the function by solving for the values of the independent variable where the derivative of the function is equal to zero. Then, use the second derivative test to determine which of these critical points are local maxima.
​
Here’s an example of how to find the peaks of a function:
In this code, we define a sample function f[x] with peaks. We then find the derivative of the function using the D function and solve for the critical points where the derivative is zero using NSolve. We find the second derivative and use the second derivative test to determine which of the critical points are local maxima. The Select function is used to filter the critical points based on the second derivative test. Finally, we display the x-values of the peaks.
​
Note that NSolve is used for numerical solutions, and Solve can be used for symbolic solutions if needed. The Reals domain restriction is used to find real solutions.
Note: I haven’t tried this code. You may need to tweak it.

WHAT ARE SOME INTERESTING NOVEL USE CASES YOU HAVE SEEN WITH THIS COMBINATION?

Lot’s of examples to be demonstrated during the week. Please visit the links we have shared:

  • https://wolfr.am/1cNh4it1X
    • 
  • https://wolfr.am/1cNgomKJA
    • 
  • https://community.wolfram.com/groups/-/m/t/2897251

    • CAN YOU DEMONSTRATE USING THE FUNCTION IMPORT AND THE FORMATS AVAILABLE?

    Import[] works since a few days.

    HISTORICAL CARBON DIOXIDE EMISSIONS FROM GLOBAL FOSSIL FUEL COMBUSTION AND INDUSTRIAL PROCESSES FROM 1750 TO 2020

    use wolfram to make a chart of Historical carbon dioxide emissions from global fossil fuel combustion and industrial processes from 1750 to 2020

    HOW CAN WOLFRAM BE TERMINATED ELEGANTLY IN A TOO-LONG TO RUN CALL? IS THERE AN ANALOG TO MMA'S 'ALT .' ?

    The 'stop generating' button at the bottom.

    NON ENGLISH SPEAKER - IS THE WORD HALLUCINATION USED A FAD?

    One definition of Hallucination found on the internet....
    Hallucination: A well-known phenomenon in large language models, in which the system provides an answer that is factually incorrect, irrelevant or nonsensical, because of limitations in its training data and architecture.

    IF YOU INTEND TO ADD WORDS TO AN ALREADY ENTERED INPUT, YOU HAVE TO REPEAT THAT PREVIOUS INPUT AND ADD THAT INTEND TO THE RERUN.

    Thanks for sharing.

    IS THERE A WAY TO MAKE CHATGPT "PASS ON" THE PROBLEM AS SOON AS POSSIBLE TO WOLFRAM ALPHA/LANGUAGE, RATHER THAN TRYING TO SOLVE IT ON ITS OWN? IT LOOKS LIKE CHATGPT TRIED TO SOLVED THAT SYSTEM OF EQUATION ITSELF, INCREASING THE LIKELIHOOD OF A MISTAKE.

    Yes. At the start of the Chat you can instruct ChatGPT pretty exactly what it should do.

    NOBODY SAID THE SOLUTIONS TO THAT LINEAR SYSTEM SHOULD BE REAL. HOW WOULD ONE TELL CHATGPT WE WANT ALL COMPLEX (SAY - OR IN AN ARBITRARY FIELD) SOLUTIONS ?

    Just spell it out and ChatGPT will use the appropriate third argument of Solve/Reduce.

    IS THERE A MECHANISM (API, ...) TO KEEP STATE OF SESSION WHEN FORCE TO INTERRUPT WORK (EOB, LATE NIGHT, ...) IN THE WAY WOLFRAMKERNEL IS ABLE TO SAVE ITS COMPLETE STATE (INCL. ALL RESULTS) TO KEEP GOING LATER, OR SAY BACKUP AND REDO LATER?

    You can continue at the 'end', but you can't make changes to earlier pieces.

    COULD YOU POST A LINK FOR WOLFRAM GPT APPLIED TO NOTEBOOKS?

    AIAssistant: https://resources.wolframcloud.com/FunctionRepository/resources/AIAssistant/

    IN THE FUTURE, DO YOU PLAN TO ENABLE THE USE OF GPT THROUGH THE API WITHIN WOLFRAM?

    We have an example here https://resources.wolframcloud.com/FunctionRepository/resources/AIAssistant/

    CAN YOU UPLOAD A DATABASE AND WORK WITH IT?

    You can probably build a plugin for retrieving information from a proprietary database and use it within the chat interface.

    CAN IT SOLVE COMPLEX LINE INTEGRALS USING RESIDUES ?

    This would be non-trivial and require a lot of coaching - explicit instructions to deal with every step of the problem.

    WHAT DO YOU RECOMMEND AS SOME POSSIBLE PLUGIN IDEAS TO DEVELOP WITH WOLFRAM? (I HAVE ACCESS, LOOKING FOR IDEAS)

    This might help: https://www.youtube.com/live/gTQqak1PKFk

    DO I ALSO HAVE COMPUTING TIME LIMITATION (EXTRA THAN HD IMAGE RESOLUTION) IF I USE THE WOLFRAM PLUG-IN?

    This would depend on ChatGPT, I believe

    SOMETIMES CHATGPT CAN JUST ASK WOLFRAM ALPHA, SOMETIMES IT NEEDS TO WRITE SOME WOLFRAM LANGUAGE CODE. HOW GOOD IS IT AT JUDGING BETWEEN THESE TWO CASES AND MAKING THE PROPER CHOICE?

    Pretty good. If what it needs are just facts/data or simple calculations, generally it goes to Alpha and gets a direct result. Otherwise it goes to the Cloud. But if it fails to get a useful result after a couple of tries from one or the other, it often is quite good at trying a different approach with the other endpoint.

    CAN YOU SHOW US HOW TO WRITE AN EQUATION IN MATHEMATICA USING CHATAI. FOR AN EXAMPLE, WRITE OUT THE THEIS EQUATION AND ASSIGN UNITS TO EACH VARIABLE.

    Use Wolfram Language to write out the Theis equation and assign units to each variable.
    From ChatGPT:
    In this code, rValue, tValue, QValue, TValue, and SValue are placeholders for the actual numerical values of the variables. You can replace them with specific numerical values to evaluate the drawdown at a particular location and time. The ExpIntegralEi function is used to represent the Theis well function.

    CAN IT GIVE ME A PREDICTION USING MACHINE LEARNING ALGORITHM?

    I tried the following: