My name is Andrey Tsygan – I’m not a programmer, I look at AI technologies from the point of view of a person who knows what he wants, but does not have the skills to do it through code.
That is, he could give formulas for Excel, or tell you step by step how to do it – but you had to do it yourself – and this was the weakest link.
But with the new plugin code interpreter the rules of the game have changed. Now he doesn’t tell how, but now he does. And it looks fantastic for non-coders!
Not only does it look – it seems to me that this is a radical change in the rules of the game – many applications, convenient, cool, may undergo radical changes .. Since “in the foreseeable future” it is quite realistic that there will be no need for an interface, manuals – just say what you want, and you will get it. Of course, not everything, but the speed of improvements impresses me.
I have seen from personal experience how a person can “jump” the levels of interaction with the interface. Grandmother, 74 years old, push-button phone, no tablets (I didn’t want to poke my finger somewhere), I didn’t hear about YouTube. But when she comes to visit, she knows – you just need to say “Alice, put on a concert for me by Nadezhda Kadysheva where she is with a young bayan player in Vitebsk” – and that’s it! That might work here as well.
What is Code Interpreter?
Code Interpreter is a plugin from OpenAI for ChatGPT. It allows you to create and run code directly in the interface, and can work with uploaded files.
For a long time it was in closed testing, but now it is becoming available to mere mortals, with a Plus subscription (20USD per month). It works with the GPT-4 model
We connect it in the settings.
Next, in the model window, select code interpreter
Now we have the opportunity to upload files up to 100 mb in size.
It can be your database, your calculations in Excel, pdf files, media files. In general – 100 MB is quite where to turn around.
But it’s all lyrics, the main question “What can I do with this, what can I do with this?”.
I tested several real tasks, I’m sure in the near future we will see hundreds of cool cases. Write your cases in the comments – I will be grateful.
All examples I will show are real, from business (small business). Large corporations probably have other requests – well, other resources.
In a small business, you may not have a call center, a marketing department, and sometimes even sales, but there will always be taxes and Excel. Let’s start with the second.
In general, often all our small data is placed in Excel – these are advertising metrics, uploading from CRM, banking movements.
For the test, let’s take open data.
ISO has an annual report on the number of valid certificates by country and standard. Download and upload the file to it, without preparation.
Дай мне количество сертификатов по русскоязычным странах по стандартам, и посчитай общее количество выданных сертификатов как по стандарту так и по стране
I wrote all requests in a very simple language, namely “this is what you need, and that’s it.” Look at the columns yourself, understand where the data is. I tried to emulate the user’s position, there is a dataset – I want answers in the form of graphs and diagrams.
This is the data we get in a minute (first it looks at the data, tries it, marks it out) and then quickly builds the necessary output data.
We ask you to make a vertical chart from the table according to ISO 9001
Here is the heatmap by quantity and by country
And here are the top 10 industries according to the ISO 9001 standard from Belarus
According to him, he can build these types of diagrams (but something tells me that soon smart people will find a way to expand this list)
In my current environment, I can plot the following kinds of charts using the matplotlib library:
Line plots: used to visualize data over time (so-called “time series”) or to show the relationship between two sets of data.
Bar plots: used to compare values of different categories or groups.
Histograms: Used to visualize the distribution of a one-dimensional dataset.
Scatter plots: Used to visualize the relationship between two sets of data.
Box plots: used to visualize the statistical properties of the data distribution (median, quartiles, outliers, etc.).
Pie charts: Used to display the proportions of different categories in the total.
Heatmaps: Used to render a matrix of data, where the color of each cell corresponds to the value in that cell.
Error bars: used to visualize uncertainty or variation in data.
This data can be immediately requested to be made in PDF or PPT
Work on PDF Invoice Recognition
Invoice, invoice, act – we get a lot of them in business, and we need to manually enter into our payment system – to whom, for what, and how much to pay.
Let’s ask to recognize the text, and highlight the payment data.
Let’s take not even the simplest type of account, it is more an account-agreement. There is both a text part and a table part. Our task is to get who to pay, how much to pay, and what to pay for
We upload the file and ask, in a simple, human way, to give the data:
The only BUT – with the Russian language he writes that he can’t work, there is no library (
Working with audio
I did not succeed in recognizing the text from the audio either in Russian or in English. I’m sure you can suggest libraries and maybe he can do something better, but I know only one library – them. Lenin.
But it can do simple procedures – cut, for example, audio into the required number of parts (with a maximum size and duration). This may be necessary, for example, when a thread of an application does not take a large file, and you need to split it.
For example, Whisper, which I managed to use, only eats files up to 18 minutes, and it was inconvenient to break long videos into these pieces. Now it’s easy
Working with images
Putting text on an image.
I tried to write text on the image, but in this wording he completely refused.
But if you change the task to “apply a watermark” – he writes. Swears a little about fonts, but writes. He writes in very small print, he even indicated the font size, but he said so far I can do it. But it CAN!
In general, simple moments with the image can do – rotate, reversal, change the format.
Working with video
It can convert GIF to MP4, do simple video editing like zoom and create GIF
Riley Goodside’s twitter has examples of working with videos
I tried to solve a problem that I recently fiddled with in CapCut. Make a uniform video according to the picture (slide). He completed this task as well, here is the result:
For me personally, this plugin was the opening of “Pandora’s box”. The main thing is the request – not only the instruction – but the finished result.
I think this may shake the position of some utilities – now it will be possible to do many things inside chatgpt.
He often answered me that he did not have this or that library – I am happy to imagine how this plugin will play if it will be possible to load the necessary libraries into it