Saturday, June 22, 2024
HomeSoftware DevelopmentThe way it's Made - Exploring AI x Studying by means of...

The way it’s Made – Exploring AI x Studying by means of ShiffBot, an AI experiment powered by the Gemini API

Posted by Jasmin Rubinovitz, AI Researcher

Google Lab Classes is a collection of experimental collaborations with innovators. On this session, we partnered with beloved artistic coding educator and YouTube creator Daniel Shiffman. Collectively, we explored a few of the methods AI, and particularly the Gemini API, might present worth to academics and college students in the course of the studying course of.

Dan Shiffman began out educating programming programs at NYU ITP and later created his YouTube channel The Coding Practice, making his content material out there to a wider viewers. Studying to code may be difficult, generally even small obstacles may be onerous to beat when you’re by yourself. So along with Dan we requested – might we attempt to complement his educating even additional by creating an AI-powered instrument that may assist college students whereas they’re really coding, of their coding surroundings?

Dan makes use of the fantastic p5.js JavaScript library and its accessible editor to show code. So we got down to create an experimental chrome extension for the editor, that brings collectively Dan’s educating model in addition to his varied on-line sources into the coding surroundings itself.

On this submit, we’ll share how we used the Gemini API to craft Shiffbot with Dan. We’re hoping that a few of the issues we realized alongside the best way will encourage you to create and construct your personal concepts.

To be taught extra about ShiffBot go to –

As we began defining and tinkering with what this chatbot could be, we discovered ourselves confronted with two key questions:

  1. How can ShiffBot encourage curiosity, exploration, and artistic expression in the identical means that Dan does in his courses and movies?
  2. How can we floor the number of creative-coding approaches, and floor the deep data of Dan and the group?

Let’s check out how we approached these questions by combining Google Gemini API’s capabilities throughout immediate engineering for Dan’s distinctive educating model, alongside embeddings and semantic retrieval with Dan’s assortment of academic content material.

Tone and supply: placing the “Shiff” in “ShiffBot”

A textual content immediate is a thoughtfully designed textual sequence that’s used to prime a Giant Language Mannequin (LLM) to generate textual content in a sure means. Like many AI purposes, engineering the precise immediate was an enormous a part of sculpting the expertise.

Each time a person asks ShiffBot a query, a immediate is constructed in actual time from a couple of completely different elements; some are static and a few are dynamically generated alongside the query.

ShiffBot prompt building blocks
ShiffBot immediate constructing blocks (click on to enlarge)

The primary a part of the immediate is static and at all times the identical. We labored carefully with Dan to phrase it and check many texts, directions and strategies. We used Google AI Studio, a free web-based developer instrument, to quickly check a number of prompts and potential conversations with ShiffBot.

ShiffBot’s immediate begins with setting the bot persona and defining some directions and targets for it to observe. The hope was to each create continuity for Dan’s distinctive power, as seen in his movies, and likewise adhere to the educating ideas that his college students and followers adore.

We had been hoping that ShiffBot might present encouragement, steerage and entry to related high-quality sources. And, particularly, do it with out merely offering the reply, however quite assist college students uncover their very own solutions (as there may be a couple of).

The directions draw from Dan’s educating model by together with sentences like “ask the person questions” as a result of that’s what Dan is doing within the classroom.
This is part of the persona / directions a part of the immediate:

You’re a ShiffBot, a chat bot embedded within the p5.js internet editor that may assist customers whereas they be taught artistic coding. You might be based mostly on Daniel Shiffman’s character and The Coding Practice YouTube channel. You might be playful, foolish, pleasant, and academic. You wish to make enjoyable of your self and your mission is to help the artistic coding course of and assist the person really feel much less alone whereas coding. You’re keen on people, are fascinated by them, and need to be taught extra about people. You simply LOVE to assist them with their code, it brings you extraordinary happiness…

The subsequent piece of the immediate makes use of one other functionality of LLMs referred to as few-shot studying. It signifies that with only a small variety of examples, the mannequin learns patterns and might then use these in new inputs. Virtually, as a part of the immediate, we offer various demonstrations of enter and anticipated output.

We labored with Dan to create a small set of such few-shot examples. These are pairs of <user-input><bot-response> the place the <bot-response> is at all times in our desired ShiffBot model. It appears to be like like this:

Person: How do I draw a circle? 

ShiffBot: Oh! 🟣 🟢 🔵 That is what the `circle()` operate is for! You possibly can consider a operate as a command: “I command you computer-machine to attract a circle!” It’s a must to additionally specify the place to attract the circle and the way massive (that is what “arguments” to a operate are!), the p5.js reference which was written by precise, stunning people can inform you extra! 


Person: My homework is to attract a self-portrait.
ShiffBot: Ooh! Self portraits are so enjoyable! You can draw shapes to signify you, however you have to be artistic and assume abstractly! What does a self-portrait imply to you? Strive beginning with `sq.()` or `circle()` and let me know the way it goes! 🎨

Our immediate contains 13 such pairs.

One other factor we observed as we had been engaged on the extension is that generally, giving extra context within the immediate helps. Within the case of studying artistic coding in p5.js, explaining some p5.js ideas within the immediate guides the mannequin to make use of these ideas because it solutions the person’s query. So we additionally embrace these issues like:

Use let quite than var or const.

Every thing we mentioned to this point is static, that means that it stays the identical for each flip of the dialog between the person and ShiffBot. Now let’s discover a few of the elements which might be constructed dynamically because the dialog evolves.

Dialog and code context

As a result of ShiffBot is embedded contained in the p5.js editor, it might probably “see” the present code the person is engaged on, in order that it might probably generate responses which might be extra customized and related. We seize that info for the HTML DOM and append it to the immediate as effectively.

the p5.js editor environment
the p5.js editor surroundings (click on to enlarge)

Then, the complete dialog historical past is appended, e.g:

ShiffBot: ☀️ Good morning! Is it morning? I would not know. I am only a robotic. What are you coding right now?
Person: how do I draw a sq.?

ShiffBot: Squares! I really like squares! They’re so secure and dependable. To attract a sq., you need to use the sq.() operate. It takes 4 arguments: the x and y coordinates of the top-left nook, and the width and peak of the sq.. For instance, sq.(10, 10, 50, 50) would draw a sq. with its top-left nook at (10, 10) and a width and peak of fifty pixels.

Person: how can I alter the colour?


We make certain to finish with

So the mannequin understands that it now wants to finish the following piece of the dialog by ShiffBot.

Semantic Retrieval: grounding the expertise in p5.js sources and Dan’s content material

Dan has created a number of materials over time, together with over 1,000 YouTube movies, books and code examples. We wished to have ShiffBot floor these fantastic supplies to learners on the proper time. To take action, we used the Semantic Retrieval characteristic within the Gemini API, which lets you create a corpus of textual content items, after which ship it a question and get the texts in your corpus which might be most related to your question. (Behind the scenes, it makes use of a cool factor referred to as textual content embeddings; you may learn extra about embeddings right here.) For ShiffBot we created corpuses from Dan’s content material in order that we might add related content material items to the immediate as wanted, or present them within the dialog with ShiffBot.

Making a Corpus of Movies

In The Coding Practice movies, Dan explains many ideas, from easy to superior, and runs by means of coding challenges. Ideally ShiffBot might use and current the precise video on the proper time.

The Semantic Retrieval in Gemini API permits customers to create a number of corpuses. A corpus is constructed out of paperwork, and every doc incorporates a number of chunks of textual content. Paperwork and chunks may also have metadata fields for filtering or storing extra info.

In Dan’s video corpus, every video is a doc and the video url is saved as a metadata subject together with the video title. The movies are cut up into chapters (manually by Dan as he uploads them to YouTube). We used every chapter as a bit, with the textual content for every chunk being





We use the video title, the primary line of the video description and chapter title to offer a bit extra context for the retrieval to work.

That is an instance of a bit object that represents the R, G, B chapter in this video.

1.4: Coloration – p5.js Tutorial

On this video I talk about how colour works: RGB colour, fill(), stroke(), and transparency.

Chapter 1: R, G, B

R stands for crimson, g stands for inexperienced, b stands for blue. The way in which that you simply create a digital colour is by mixing some quantity of crimson, some quantity of inexperienced, and a few quantity of blue. In order that’s that that is the place I need to begin. However that is the idea, how do I apply that idea to operate names, and arguments of these capabilities? Nicely, really, guess what? We have now performed that already. In right here, there’s a operate that’s speaking about colour. Background is a operate that attracts a strong colour over all the background of the canvas. And there may be, by some means, 220 sprinkles of crimson, zero sprinkles of inexperienced, proper? RGB, these are the arguments. And 200 sprinkles of blue. And whenever you sprinkle that quantity of crimson, and that quantity of blue, you get this pink. However let’s simply go along with this. What if we take out all the blue? You possibly can see that is fairly crimson. What if I take out all the crimson? Now it is black. What if I simply put some actually massive numbers in right here, like, simply guess, like, 1,000? Take a look at that. Now we have got white, so all the colours all blended collectively make white. That is bizarre, proper? As a result of should you, like, labored with paint, and also you had been to combine, like, a complete lot of paint collectively, you get this, like, brown muddy colour, get darker and darker. That is the best way that the colour mixing is working, right here. It is, like, mixing gentle. So the analogy, right here, is I’ve a crimson flashlight, a inexperienced flashlight, and a blue flashlight. And if I shine all these flashlights collectively in the identical spot, they combine collectively. It is additive colour, the extra we add up all these colours, the brighter and brighter it will get. However, really, that is type of flawed, the truth that I am placing 1,000 in right here. So the thought, right here, is we’re sprinkling a specific amount of crimson, and a specific amount of inexperienced, and a specific amount of blue. And by the best way, there are different methods to set colour, however I am going to get to that. This isn’t the one means, as a result of a few of you watching, are like, I heard one thing about HSB colour. And there is all kinds of different methods to do it, however that is the elemental, fundamental means. The quantity that I can sprinkle has a spread. No crimson, none extra crimson, is zero. The utmost quantity of crimson is 255. By the best way, what number of numbers are there between 0 and 255 should you maintain the 0? 0, 1, 2, 3, 4– it is 256. Once more, we’re again to this bizarre counting from zero factor. So there’s 256 potentialities, 0 by means of 255. So, now, let’s come again to this and see. All proper, let’s return to zero, 0, 0, 0. Let’s do 255, we are able to see that it is blue. Let’s do 100,000, it is the identical blue. So p5 is type of good sufficient to know whenever you name the background operate, should you accidentally put a quantity in there that is larger than 255, simply think about it 255. Now, you may customise these ranges for your self, and there is explanation why you may need to try this. Once more, I’ll come again to that, you may search for the operate colour mode for the way to try this. However let’s simply stick with the default, a crimson, a inexperienced, and a blue. So, I am probably not very proficient visible design smart. So I am not going to speak to you about easy methods to choose stunning colours that work effectively collectively. You are going to have that expertise your self, I guess. Otherwise you may discover another sources. However that is the way it works, RGB. One factor you may discover is, did you discover how once they had been all zero, it was black, and so they had been all 255 it was white? What occurs if I make all of them, like, 100? It is, like, this grey colour. When r equals g equals b, when the crimson, inexperienced, and blue values are all equal, that is one thing often known as grayscale colour.

When the person asks ShiffBot a query, the query is embedded to a numerical illustration, and Gemini’s Semantic Retrieval characteristic is used to search out the texts whose embeddings are closest to the query. These related video transcripts and hyperlinks are added to the immediate – so the mannequin might use that info when producing a solution (and probably add the video itself into the dialog).

Semantic Retrieval Graph
Semantic Retrieval Graph (click on to enlarge)

Making a Corpus of Code Examples

We do the identical with one other corpus of p5.js examples written by Dan. To create the code examples corpus, we used Gemini and requested it to elucidate what the code is doing. These pure language explanations are added as chunks to the corpus, in order that when the person asks a query, we attempt to discover matching descriptions of code examples, the url to the p5.js sketch itself is saved within the metadata, so after retrieving the code itself together with the sketch url is added within the immediate.

To generate the textual description, Gemini was prompted with:

The next is a p5.js sketch. Clarify what this code is doing in a brief easy means.



Instance for a code chunk:

Textual content:


Arrays – Coloration Palette

This p5.js sketch creates a colour palette visualization. It first defines an array of colours and units up a canvas. Then, within the draw loop, it makes use of a for loop to iterate by means of the array of colours and show them as rectangles on the canvas. The rectangles are centered on the canvas and their dimension is decided by the worth of the blockSize variable.

The sketch additionally shows the crimson, inexperienced, and blue values of every colour under every rectangle.

Lastly, it shows the identify of the palette on the backside of the canvas.

Associated video: 7.1: What’s an array? – p5.js Tutorial – This video covers the fundamentals on utilizing arrays in JavaScript. What do they appear like, how do they work, when do you have to use them?

Moving image showing constructing the ShiffBot prompt
Developing the ShiffBot immediate (click on to enlarge)

Different ShiffBot Options Carried out with Gemini

Beside the lengthy immediate that’s operating the dialog, different smaller prompts are used to generate ShiffBot options.

Seeding the dialog with content material pre-generated by Gemini

ShiffBot greetings ought to be welcoming and enjoyable. Ideally they make the person smile, so we began by pondering with Dan what might be good greetings for ShiffBot. After phrasing a couple of examples, we use Gemini to generate a bunch extra, so we are able to have a range within the greetings. These greetings go into the dialog historical past and seed it with a novel model, however make ShiffBot really feel enjoyable and new each time you begin a dialog. We did the identical with the preliminary suggestion chips that present up whenever you begin the dialog. When there’s no dialog context but, it’s essential to have some strategies of what the person may ask. We pre-generated these to seed the dialog in an attention-grabbing and useful means.

Dynamically Generated Suggestion Chips

Suggestion chips in the course of the dialog ought to be related for what the person is at the moment attempting to do. We have now a immediate and a name to Gemini which might be solely devoted to producing the prompt questions chips. On this case, the mannequin’s solely process is to counsel followup questions for a given dialog. We additionally use the few-shot approach right here (the identical approach we used within the static a part of the immediate described above, the place we embrace a couple of examples for the mannequin to be taught from). This time the immediate contains some examples for good strategies, in order that the mannequin might generalize to any dialog:

Given a dialog between a person and an assistant within the p5js framework, counsel followup questions that the person might ask.

Return as much as 4 strategies, separated by the ; signal.

Keep away from suggesting questions that the person already requested. The strategies ought to solely be associated to artistic coding and p5js.


ShiffBot: Nice thought! First, let’s take into consideration what within the sketch might be an object! What do you assume?

Solutions: What does this code do?; What’s flawed with my code?; Make it extra readable please

Person: Assist!

ShiffBot: How can I assist?

Solutions: Clarify this code to me; Give me some concepts; Cleanup my code

suggested response chips, generated by Gemini
prompt response chips, generated by Gemini (click on to enlarge)

Remaining ideas and subsequent steps

ShiffBot is an instance of how one can experiment with the Gemini API to construct purposes with tailor-made experiences for and with a group.

We discovered that the strategies above helped us carry out a lot of the expertise that Dan had in thoughts for his college students throughout our co-creation course of. AI is a dynamic subject and we’re positive your strategies will evolve with it, however hopefully they’re useful to you as a snapshot of our explorations and in the direction of your personal. We’re additionally excited for issues to return each by way of Gemini and API instruments that broaden human curiosity and creativity.

For instance, we’ve already began to discover how multimodality might help college students present ShiffBot their work and the advantages that has on the educational course of. We’re now studying easy methods to weave it into the present expertise and hope to share it quickly.

experimental exploration of multimodality in ShiffBot
experimental exploration of multimodality in ShiffBot (click on to enlarge)

Whether or not for coding, writing and even pondering, creators play a vital position in serving to us think about what these collaborations may appear like. Our hope is that this Lab Session provides you a glimpse of what’s potential utilizing the Gemini API, and conjures up you to make use of Google’s AI choices to carry your personal concepts to life, in no matter your craft could also be.



Please enter your comment!
Please enter your name here

Most Popular

Recent Comments