Hi everyone:)
My name is Dmitry Trepachev.
I'm starting a new channel about AI development.
I named it News from the Field, because I'll be writing my posts right from the battlefield with modern AI:)
The channel will cover all sorts of things: from the history of AI and theory to various techniques and tricks.
Let's get started!
My name is Dmitry Trepachev.
I'm starting a new channel about AI development.
I named it News from the Field, because I'll be writing my posts right from the battlefield with modern AI:)
The channel will cover all sorts of things: from the history of AI and theory to various techniques and tricks.
Let's get started!
So do you actually know your stuff, dude?
Right now I work for a US company that is building a large project with AI (what we do is a trade secret, nyah nyah nyah ). This is my main job at the moment, and I have a great chance to look at modern models and approaches to AI development from my own experience (and most importantly, on the company's dime). Every day I see which models are trending, how people work with them, which techniques get used on a large project, and what that looks like in practice.
In parallel, I build my own projects with AI. I haven't abandoned code.mu, and I still release new textbooks. I write the new textbooks with AI: I think through the structure, write the lessons together with AI, and bring everything to a finished form. AI speeds up textbook writing dozens of times over. And that is really cool! After just 1–2 days of work, a new textbook shows up on the site, and you can already study from it.
My daily work is tied to AI. In effect, every day I am on the frontier — at the very peak of modern development: I use the coolest newest models, try out various clever AI-development techniques in practice, and all of this inside a large live project! Over this time I have seen a ton of real working situations and pulled a lot of valuable experience out of them, which is exactly what I will be sharing with you:)
Let's go!
Right now I work for a US company that is building a large project with AI (
In parallel, I build my own projects with AI. I haven't abandoned code.mu, and I still release new textbooks. I write the new textbooks with AI: I think through the structure, write the lessons together with AI, and bring everything to a finished form. AI speeds up textbook writing dozens of times over. And that is really cool! After just 1–2 days of work, a new textbook shows up on the site, and you can already study from it.
My daily work is tied to AI. In effect, every day I am on the frontier — at the very peak of modern development: I use the coolest newest models, try out various clever AI-development techniques in practice, and all of this inside a large live project! Over this time I have seen a ton of real working situations and pulled a lot of valuable experience out of them, which is exactly what I will be sharing with you:)
Let's go!
Decoding the boring terms
In conversation we use these closely related words: AI and neural network. Let's figure out what the difference actually is. So:
1.
AI stands for artificial intelligence.
Artificial, because it is not a human and not a living mind, but a program made by people. Intelligence, because it takes on tasks that require a human to think: answer a question, write a text, explain an error, put code together.
The name appeared in 1956. Back then the goal was to make a machine that could handle such tasks. At the time they did not succeed (no surprise!), but the word stuck.
2.
Neural network is an alternative name for AI. Let's figure out where this name came from. The thing is, AI is not just a program. More precisely, it is not a program at all! Ordinary programs follow an algorithm and do exactly what is written in the code. That is both good and bad. It is good because an ordinary program does what it was told (and does not do anything else!). And it is bad because an ordinary program cannot do things that require thinking.
AI, on the other hand, can solve such tasks. Because on the inside, AI works in a way similar to the human brain. You of course know that the brain is made of neurons (if you did not sleep through biology class the way I did). Neurons are those little things in the brain that nerves stretch out from, like little wires, and electrical signals run along them. And thanks to that, a human has consciousness, memory, the ability to reason, and other similar abilities.
A program for AI is built on the same principle. Of course there are no real neurons in there — a program simply imitates how they work. Engineers, as often happens, did not invent the feature themselves. They peeked at how nature does it.
Inside an AI there is not one neuron, but a whole network. That is where the name neural network comes from. In everyday life, saying AI or saying neural network is practically the same thing.
In conversation we use these closely related words: AI and neural network. Let's figure out what the difference actually is. So:
1.
AI stands for artificial intelligence.
Artificial, because it is not a human and not a living mind, but a program made by people. Intelligence, because it takes on tasks that require a human to think: answer a question, write a text, explain an error, put code together.
The name appeared in 1956. Back then the goal was to make a machine that could handle such tasks. At the time they did not succeed (no surprise!), but the word stuck.
2.
Neural network is an alternative name for AI. Let's figure out where this name came from. The thing is, AI is not just a program. More precisely, it is not a program at all! Ordinary programs follow an algorithm and do exactly what is written in the code. That is both good and bad. It is good because an ordinary program does what it was told (and does not do anything else!). And it is bad because an ordinary program cannot do things that require thinking.
AI, on the other hand, can solve such tasks. Because on the inside, AI works in a way similar to the human brain. You of course know that the brain is made of neurons (if you did not sleep through biology class the way I did). Neurons are those little things in the brain that nerves stretch out from, like little wires, and electrical signals run along them. And thanks to that, a human has consciousness, memory, the ability to reason, and other similar abilities.
A program for AI is built on the same principle. Of course there are no real neurons in there — a program simply imitates how they work. Engineers, as often happens, did not invent the feature themselves. They peeked at how nature does it.
Inside an AI there is not one neuron, but a whole network. That is where the name neural network comes from. In everyday life, saying AI or saying neural network is practically the same thing.
Even more boring terms
Last time we figured out that AI and neural network are almost the same thing. But if you dig deeper, even more boring words come up: model, LLM, GPT. Let's sort those out too.
But first, a small analogy.
A human is born with a ready-made set of neurons. But babies, as you know, are a bit dim: they cannot talk, or walk, or solve math problems. A bare set of neurons gives you nothing — it has to be trained. Parents, daycare, school, university, life experience — and only then does something decent come out of a baby (and even that is not guaranteed).
With AI it is exactly the same story. A neural network is "born" dumb. It is just a set of mathematical formulas that imitate neurons. On their own they cannot do anything. For an AI to start answering questions and writing texts, it has to be trained. Show it millions of examples, let it "read" mountains of text, correct its mistakes — and only after that does it start to be able to do something.
Now, the boring terms.
1. Model
Different companies work on creating AI. Among them: OpenAI, Anthropic, Google, xAI, Meta, DeepSeek. Each company creates its own AI, trains it, and releases it.
That specific trained AI from a specific company is what is called a model.
For example:
- OpenAI released the ChatGPT model.
- Anthropic released Claude.
- Google released Gemini.
- xAI released Grok.
- Meta released Llama.
- DeepSeek released DeepSeek.
Guess who in this lineup is short on imagination!
So when we say "I use ChatGPT" or "I asked Claude", we mean a specific model.
But there is a second meaning too. A company trains a model, releases it — and immediately starts working on the next, smarter version. ChatGPT has had versions: GPT-3, GPT-3.5, GPT-4, GPT-4o, and so on. Each such release is also a model.
So the word "model" lives in two senses:
- in the broad sense: a model is a specific AI from a specific company (ChatGPT, Claude, Gemini).
- in the narrow sense: a model is a specific trained version of that AI (GPT-4, GPT-4o, Claude 3.5 Sonnet).
2. LLM
LLM stands for Large Language Model.
Let's break it down word by word:
- Large. Because inside such a model there are billions, or even trillions, of parameters. It is like billions of little knobs you can turn to tune the model toward the behavior you want.
- Language. Because it works with language: text, words, sentences.
- Model. Well, we already covered that.
In plain terms, LLM is a clarification. Not just a model, but specifically a language model that can work with text. ChatGPT, Claude, Gemini — all of these are LLMs.
3. GPT
GPT stands for Generative Pre-trained Transformer.
It sounds scary, but let's take it piece by piece:
- Generative. The model does not just pick an answer from a ready-made list. It generates it: it composes the text word by word.
- Pre-trained. The model is first trained on a huge amount of text before it is released. This is like the "basic education" that any AI gets.
- Transformer. This is the name of a neural network architecture. In 2017 the paper "Attention Is All You Need" came out, and that paper proposed this architecture. It turned out to be so successful that almost all modern LLMs are built on it.
So GPT is not just a model, but a model of a specific type: generative, pre-trained, and built on transformers. ChatGPT is a product from OpenAI that runs on models from the GPT family.
To sum up:
- A model is a specific trained AI from a specific company (ChatGPT, Claude, Gemini).
- An LLM is a large language model, that is, a model that works with text.
- GPT is a generative pre-trained model built on the transformer architecture.
If last time we established that AI and neural network are almost the same thing, now we can put it this way: ChatGPT is a model that is an LLM built on the GPT architecture.
Last time we figured out that AI and neural network are almost the same thing. But if you dig deeper, even more boring words come up: model, LLM, GPT. Let's sort those out too.
But first, a small analogy.
A human is born with a ready-made set of neurons. But babies, as you know, are a bit dim: they cannot talk, or walk, or solve math problems. A bare set of neurons gives you nothing — it has to be trained. Parents, daycare, school, university, life experience — and only then does something decent come out of a baby (and even that is not guaranteed).
With AI it is exactly the same story. A neural network is "born" dumb. It is just a set of mathematical formulas that imitate neurons. On their own they cannot do anything. For an AI to start answering questions and writing texts, it has to be trained. Show it millions of examples, let it "read" mountains of text, correct its mistakes — and only after that does it start to be able to do something.
Now, the boring terms.
1. Model
Different companies work on creating AI. Among them: OpenAI, Anthropic, Google, xAI, Meta, DeepSeek. Each company creates its own AI, trains it, and releases it.
That specific trained AI from a specific company is what is called a model.
For example:
- OpenAI released the ChatGPT model.
- Anthropic released Claude.
- Google released Gemini.
- xAI released Grok.
- Meta released Llama.
- DeepSeek released DeepSeek.
So when we say "I use ChatGPT" or "I asked Claude", we mean a specific model.
But there is a second meaning too. A company trains a model, releases it — and immediately starts working on the next, smarter version. ChatGPT has had versions: GPT-3, GPT-3.5, GPT-4, GPT-4o, and so on. Each such release is also a model.
So the word "model" lives in two senses:
- in the broad sense: a model is a specific AI from a specific company (ChatGPT, Claude, Gemini).
- in the narrow sense: a model is a specific trained version of that AI (GPT-4, GPT-4o, Claude 3.5 Sonnet).
2. LLM
LLM stands for Large Language Model.
Let's break it down word by word:
- Large. Because inside such a model there are billions, or even trillions, of parameters. It is like billions of little knobs you can turn to tune the model toward the behavior you want.
- Language. Because it works with language: text, words, sentences.
- Model. Well, we already covered that.
In plain terms, LLM is a clarification. Not just a model, but specifically a language model that can work with text. ChatGPT, Claude, Gemini — all of these are LLMs.
3. GPT
GPT stands for Generative Pre-trained Transformer.
It sounds scary, but let's take it piece by piece:
- Generative. The model does not just pick an answer from a ready-made list. It generates it: it composes the text word by word.
- Pre-trained. The model is first trained on a huge amount of text before it is released. This is like the "basic education" that any AI gets.
- Transformer. This is the name of a neural network architecture. In 2017 the paper "Attention Is All You Need" came out, and that paper proposed this architecture. It turned out to be so successful that almost all modern LLMs are built on it.
So GPT is not just a model, but a model of a specific type: generative, pre-trained, and built on transformers. ChatGPT is a product from OpenAI that runs on models from the GPT family.
To sum up:
- A model is a specific trained AI from a specific company (ChatGPT, Claude, Gemini).
- An LLM is a large language model, that is, a model that works with text.
- GPT is a generative pre-trained model built on the transformer architecture.
If last time we established that AI and neural network are almost the same thing, now we can put it this way: ChatGPT is a model that is an LLM built on the GPT architecture.
A bit of interactivity!
Now let's check whether you read carefully! Below are a few questions based on the previous posts. Answer in the comments, no peeking! That is how we will find out who was looking at the pictures and who was reading the smart words!
Ha-ha-ha, and you thought you would just read and chill? You know me — I'm a master at making exercises! It is going to get even more fun!!
So, the questions:
1.
Are AI and a neural network the same thing or not?
2.
Are AI and a model the same thing or not?
3.
Are a model and an LLM the same thing or not?
4.
Are an LLM and GPT the same thing or not?
5.
Which company made the ChatGPT model?
6.
Which company made the Claude model?
7.
Which company made the Gemini model?
8.
Which company made the Grok model?
9.
Which company made the DeepSeek model?
10.
Is ChatGPT a GPT?
11.
Is Claude a GPT?
12.
What is an LLM?
13.
What is GPT?
14.
Is GPT-4o a model in the broad sense or in the narrow sense?
15.
Can an ordinary program, without AI, solve a task that requires thinking?
16.
What came first: the term "artificial intelligence" or the transformer architecture?
17.
Is a neural network made of real neurons, or of an imitation of them?
18.
In what year did the term "artificial intelligence" appear?
19.
Find the odd one out: OpenAI, Anthropic, Google, ChatGPT, xAI, DeepSeek.
20.
What does the letter L stand for in the abbreviation LLM?
21.
What does the letter L stand for in the abbreviation LLM?
22.
What does the letter M stand for in the abbreviation LLM?
23.
What does the letter G stand for in the abbreviation GPT?
22.
Can a set of neurons, without training, answer questions?
23.
How does a model in the broad sense differ from a model in the narrow sense?
---
Write your answers in the comments!
Now let's check whether you read carefully! Below are a few questions based on the previous posts. Answer in the comments, no peeking! That is how we will find out who was looking at the pictures and who was reading the smart words!
So, the questions:
1.
Are AI and a neural network the same thing or not?
2.
Are AI and a model the same thing or not?
3.
Are a model and an LLM the same thing or not?
4.
Are an LLM and GPT the same thing or not?
5.
Which company made the ChatGPT model?
6.
Which company made the Claude model?
7.
Which company made the Gemini model?
8.
Which company made the Grok model?
9.
Which company made the DeepSeek model?
10.
Is ChatGPT a GPT?
11.
Is Claude a GPT?
12.
What is an LLM?
13.
What is GPT?
14.
Is GPT-4o a model in the broad sense or in the narrow sense?
15.
Can an ordinary program, without AI, solve a task that requires thinking?
16.
What came first: the term "artificial intelligence" or the transformer architecture?
17.
Is a neural network made of real neurons, or of an imitation of them?
18.
In what year did the term "artificial intelligence" appear?
19.
Find the odd one out: OpenAI, Anthropic, Google, ChatGPT, xAI, DeepSeek.
20.
What does the letter L stand for in the abbreviation LLM?
21.
What does the letter L stand for in the abbreviation LLM?
22.
What does the letter M stand for in the abbreviation LLM?
23.
What does the letter G stand for in the abbreviation GPT?
22.
Can a set of neurons, without training, answer questions?
23.
How does a model in the broad sense differ from a model in the narrow sense?
---
Write your answers in the comments!
Know the heroes by sight!
So you are reading the news: "Sam Altman came up with...", "Elon Musk launched...", "Zuckerberg rolled out...". And a reasonable question comes up: who are all these people, and which companies do they belong to?
Now you have a cheat sheet. Read it so that next time you see a story about yet another AI, you immediately understand who it is about and which model they released.
Here are our heroes:
- Sam Altman — company OpenAI, model ChatGPT.
- Dario Amodei — company Anthropic, model Claude.
- Sundar Pichai — company Google, model Gemini.
- Elon Musk — company xAI, model Grok.
- Mark Zuckerberg — company Meta, model Llama.
- Liang Wenfeng — company DeepSeek, model DeepSeek.
So you are reading the news: "Sam Altman came up with...", "Elon Musk launched...", "Zuckerberg rolled out...". And a reasonable question comes up: who are all these people, and which companies do they belong to?
Now you have a cheat sheet. Read it so that next time you see a story about yet another AI, you immediately understand who it is about and which model they released.
Here are our heroes:
- Sam Altman — company OpenAI, model ChatGPT.
- Dario Amodei — company Anthropic, model Claude.
- Sundar Pichai — company Google, model Gemini.
- Elon Musk — company xAI, model Grok.
- Mark Zuckerberg — company Meta, model Llama.
- Liang Wenfeng — company DeepSeek, model DeepSeek.
What is a prompt?
A prompt is your request to an AI. When you write something in an AI chat and ask it to do something, that is a prompt. The word comes from English "prompt", and it literally means a request.
So we talk to AI through prompts. We send a request — it sends a reply. If we do not like the reply, we adjust it with the next prompt, and so on.
A prompt is the foundation of everything. And writing prompts is a whole art you have to learn (and that is what we will gradually do later on the channel). If a prompt is put together properly, the chance that the AI solves your task on the first try is high. And if a prompt is very short or, worse, contradicts itself, the result will match.
For now let's not dive deep into the art of prompting. Let's just look at some examples of prompts, so it is clear what we are talking about.
Here are a few options:
Weak prompts
This is what most people write. Short, to the point, without extra details.
- "Tell a joke about a cat"
- "Explain why the sky is blue"
- "Write a birthday greeting"
- "Translate into English: I want to sleep"
Prompts like these work. The AI will answer something. But the result will be unpredictable: it might be good, or it might be so-so. Because you did not say what exactly you want.
Strong prompts
This is when you explain in detail what you need. The more details, the better the result.
For example, instead of "write a text about a cat" you can write:
"Write a funny text about a cat, 3 paragraphs long. The style should be like a blogger's. The cat should be lazy, but at the same time consider itself a genius. Add a dialogue with its owner."
The difference is obvious. In the first case the AI fills in what you meant on its own. In the second, you have already set the frame, and it is easier for it to hit the target.
Programmer prompts
An example of a simple programmer prompt:
"Write a Python function that adds two numbers."
The AI will write it. But it will be a basic version.
And now a complex prompt:
"Write a Python function that takes a list of numbers and returns their sum. The function should handle an empty list and return 0. Add error handling if the input is not a list. Write a docstring and a usage example. The code should follow PEP 8."
Feel the difference? In the second case you are not just asking for code, you are setting specific requirements. And the AI produces exactly what is needed.
Another example. Instead of "why doesn't my code work", it is better to write:
"I have some Python code. It is supposed to calculate the factorial of a number, but it throws an error. Here is the code: [code]. Explain what the error is, and suggest a fixed version."
That way the AI will not have to guess what your code is. It will see it and give an answer right away.
A prompt is your request to an AI. When you write something in an AI chat and ask it to do something, that is a prompt. The word comes from English "prompt", and it literally means a request.
So we talk to AI through prompts. We send a request — it sends a reply. If we do not like the reply, we adjust it with the next prompt, and so on.
A prompt is the foundation of everything. And writing prompts is a whole art you have to learn (and that is what we will gradually do later on the channel). If a prompt is put together properly, the chance that the AI solves your task on the first try is high. And if a prompt is very short or, worse, contradicts itself, the result will match.
For now let's not dive deep into the art of prompting. Let's just look at some examples of prompts, so it is clear what we are talking about.
Here are a few options:
Weak prompts
This is what most people write. Short, to the point, without extra details.
- "Tell a joke about a cat"
- "Explain why the sky is blue"
- "Write a birthday greeting"
- "Translate into English: I want to sleep"
Prompts like these work. The AI will answer something. But the result will be unpredictable: it might be good, or it might be so-so. Because you did not say what exactly you want.
Strong prompts
This is when you explain in detail what you need. The more details, the better the result.
For example, instead of "write a text about a cat" you can write:
"Write a funny text about a cat, 3 paragraphs long. The style should be like a blogger's. The cat should be lazy, but at the same time consider itself a genius. Add a dialogue with its owner."
The difference is obvious. In the first case the AI fills in what you meant on its own. In the second, you have already set the frame, and it is easier for it to hit the target.
Programmer prompts
An example of a simple programmer prompt:
"Write a Python function that adds two numbers."
The AI will write it. But it will be a basic version.
And now a complex prompt:
"Write a Python function that takes a list of numbers and returns their sum. The function should handle an empty list and return 0. Add error handling if the input is not a list. Write a docstring and a usage example. The code should follow PEP 8."
Feel the difference? In the second case you are not just asking for code, you are setting specific requirements. And the AI produces exactly what is needed.
Another example. Instead of "why doesn't my code work", it is better to write:
"I have some Python code. It is supposed to calculate the factorial of a number, but it throws an error. Here is the code: [code]. Explain what the error is, and suggest a fixed version."
That way the AI will not have to guess what your code is. It will see it and give an answer right away.
Problems with prompts
It seems like any task can be solved on the first try, just by writing a very precise prompt. Unfortunately, right now that is not the case. All AIs have a fundamental problem: they get lazy. If something is written in your prompt, it is far from certain that it will actually get done. The AI might simply ignore part of the prompt. Or it might apply the prompt not in every place where it is needed. And at the same time it will cheerfully report that, supposedly, it did everything!
This is how AI differs from ordinary programs. In an ordinary program, if you wrote a command, it will definitely be executed. If you give a command to an AI, it might carry it out, and it might not. Depending on its mood:)
Sound familiar?) Write in the comments!
It seems like any task can be solved on the first try, just by writing a very precise prompt. Unfortunately, right now that is not the case. All AIs have a fundamental problem: they get lazy. If something is written in your prompt, it is far from certain that it will actually get done. The AI might simply ignore part of the prompt. Or it might apply the prompt not in every place where it is needed. And at the same time it will cheerfully report that, supposedly, it did everything!
This is how AI differs from ordinary programs. In an ordinary program, if you wrote a command, it will definitely be executed. If you give a command to an AI, it might carry it out, and it might not. Depending on its mood:)
Sound familiar?) Write in the comments!
WEAK PROMPT vs STRONG PROMPT
Let's start practicing how to write more detailed prompts. Below you will get tasks with weak prompts, and you will need to turn them into strong ones and check them in a chat with a real AI.
An example of a weak prompt:
"make a logo"
Let's turn it into a strong one:
"Develop a logo concept for a children's bakery. Style: minimalism, warm colors (beige, coral). The main element is a pretzel shaped like a smile. The logo should evoke a feeling of coziness and trust. Describe 3 options, with a rationale for each."
Now try it yourself:
1. The prompt is: "write a text about a cat". What the author wanted: a funny social media post written from the point of view of a philosopher cat. Rework the prompt so it better matches what the author wanted. Check your prompt in a chat with an AI.
2. The prompt is: "make a website". What the author wanted: a one-page site for a coffee shop, with a menu and a booking form. Rework the prompt so it better matches what the author wanted. Check your prompt in a chat with an AI.
3. The prompt is: "write a post". What the author wanted: a LinkedIn post about launching a startup, told as a story with a takeaway. Rework the prompt so it better matches what the author wanted. Check your prompt in a chat with an AI.
4. The prompt is: "make a workout plan". What the author wanted: a one-month plan for a beginner, 3 times a week, at home, with no equipment. Rework the prompt so it better matches what the author wanted. Check your prompt in a chat with an AI.
5. The prompt is: "translate the text". What the author wanted: a translation of an article from English into Russian, keeping the academic style and the terminology. Rework the prompt so it better matches what the author wanted. Check your prompt in a chat with an AI.
6. The prompt is: "come up with a name". What the author wanted: a name for an eco-friendly cosmetics brand — feminine, catchy, and not already taken. Rework the prompt so it better matches what the author wanted. Check your prompt in a chat with an AI.
7. The prompt is: "write a letter". What the author wanted: a polite refusal to a client, with an alternative offered. Rework the prompt so it better matches what the author wanted. Check your prompt in a chat with an AI.
8. The prompt is: "make a summary". What the author wanted: a short lecture summary on one page, with the key terms highlighted. Rework the prompt so it better matches what the author wanted. Check your prompt in a chat with an AI.
9. The prompt is: "write a script". What the author wanted: a 30-second ad script for a fitness club. Rework the prompt so it better matches what the author wanted. Check your prompt in a chat with an AI.
10. The prompt is: "make a menu". What the author wanted: a weekly menu for a family of 4, on a budget of 5,000 rubles. Rework the prompt so it better matches what the author wanted. Check your prompt in a chat with an AI.
11. The prompt is: "explain the topic". What the author wanted: an explanation of photosynthesis for a 10-year-old, with examples from everyday life. Rework the prompt so it better matches what the author wanted. Check your prompt in a chat with an AI.
12. The prompt is: "write a review". What the author wanted: an honest book review in 3 paragraphs, with pros and cons. Rework the prompt so it better matches what the author wanted. Check your prompt in a chat with an AI.
13. The prompt is: "come up with a question". What the author wanted: 5 questions for an interview with an interior designer. Rework the prompt so it better matches what the author wanted. Check your prompt in a chat with an AI.
14. The prompt is: "write a product description". What the author wanted: a description of wireless headphones for a marketplace, with the emphasis on the benefits. Rework the prompt so it better matches what the author wanted. Check your prompt in a chat with an AI.
Write your answers in the comments!
Let's start practicing how to write more detailed prompts. Below you will get tasks with weak prompts, and you will need to turn them into strong ones and check them in a chat with a real AI.
An example of a weak prompt:
"make a logo"
Let's turn it into a strong one:
"Develop a logo concept for a children's bakery. Style: minimalism, warm colors (beige, coral). The main element is a pretzel shaped like a smile. The logo should evoke a feeling of coziness and trust. Describe 3 options, with a rationale for each."
Now try it yourself:
1. The prompt is: "write a text about a cat". What the author wanted: a funny social media post written from the point of view of a philosopher cat. Rework the prompt so it better matches what the author wanted. Check your prompt in a chat with an AI.
2. The prompt is: "make a website". What the author wanted: a one-page site for a coffee shop, with a menu and a booking form. Rework the prompt so it better matches what the author wanted. Check your prompt in a chat with an AI.
3. The prompt is: "write a post". What the author wanted: a LinkedIn post about launching a startup, told as a story with a takeaway. Rework the prompt so it better matches what the author wanted. Check your prompt in a chat with an AI.
4. The prompt is: "make a workout plan". What the author wanted: a one-month plan for a beginner, 3 times a week, at home, with no equipment. Rework the prompt so it better matches what the author wanted. Check your prompt in a chat with an AI.
5. The prompt is: "translate the text". What the author wanted: a translation of an article from English into Russian, keeping the academic style and the terminology. Rework the prompt so it better matches what the author wanted. Check your prompt in a chat with an AI.
6. The prompt is: "come up with a name". What the author wanted: a name for an eco-friendly cosmetics brand — feminine, catchy, and not already taken. Rework the prompt so it better matches what the author wanted. Check your prompt in a chat with an AI.
7. The prompt is: "write a letter". What the author wanted: a polite refusal to a client, with an alternative offered. Rework the prompt so it better matches what the author wanted. Check your prompt in a chat with an AI.
8. The prompt is: "make a summary". What the author wanted: a short lecture summary on one page, with the key terms highlighted. Rework the prompt so it better matches what the author wanted. Check your prompt in a chat with an AI.
9. The prompt is: "write a script". What the author wanted: a 30-second ad script for a fitness club. Rework the prompt so it better matches what the author wanted. Check your prompt in a chat with an AI.
10. The prompt is: "make a menu". What the author wanted: a weekly menu for a family of 4, on a budget of 5,000 rubles. Rework the prompt so it better matches what the author wanted. Check your prompt in a chat with an AI.
11. The prompt is: "explain the topic". What the author wanted: an explanation of photosynthesis for a 10-year-old, with examples from everyday life. Rework the prompt so it better matches what the author wanted. Check your prompt in a chat with an AI.
12. The prompt is: "write a review". What the author wanted: an honest book review in 3 paragraphs, with pros and cons. Rework the prompt so it better matches what the author wanted. Check your prompt in a chat with an AI.
13. The prompt is: "come up with a question". What the author wanted: 5 questions for an interview with an interior designer. Rework the prompt so it better matches what the author wanted. Check your prompt in a chat with an AI.
14. The prompt is: "write a product description". What the author wanted: a description of wireless headphones for a marketplace, with the emphasis on the benefits. Rework the prompt so it better matches what the author wanted. Check your prompt in a chat with an AI.
Write your answers in the comments!
👍1