ChatGPT-4: Main Differences from v3.5, Sample Queries, and Practical Applications

Hi! My name is Kristina, and I am the Design Director of WE2. I’m passionate not only about design, but also about new technologies: for example, I started using ChatGPT almost immediately after its release. Firstly, the version 3.5, then I upgraded to version 4. In this article I would like to share my personal experience of using ChatGPT as a personal AI assistant.

Version 3.5 handles most tasks quite well, its functionality is quite enough for many users. But if you want to move to a new level of interaction with artificial intelligence, I recommend you to try ChatGPT-4.

The main advantages for an average user (like me) are that version 4 has 40% less errors and higher creativity. It “feels” the request better and gives a more “human” response.

Chat’s ability to remember previous messages has also received a serious boost. Now the logical chain of your interaction is much less likely to break.

And of course, the killer feature is the ability to use plugins. Yes, version 4, like 3.5, is trained using up to 2021 data. But Bing or the onepilot plugin make this of little consequence, as it opens you up to going online. And what about a plugin that can make a summary from a podcast or video? I respect Andrew Huberman, sure, but I rarely get to listen to his entire 2-hour podcast. With ChatGPT-4, I can understand the main points of a release in 5 minutes.

I mean, now you can even create a CSV report for “friendly correspondence”, write working Python code, choose the most successful flight for your next trip, and do a Tarot reading. I’m not sure that last one is a benefit though 🙃

When this neural network boom started, everyone went “dabbling” with ChatGPT and Midjourney. And that’s where a lot of people ended up using AI. All because their expectations didn’t match reality due to incorrect prompts.

It’s not enough to write to ChatGPT “Make me an itinerary for Mongolia”. There’s a 99% chance you’ll have to redo it. To get a more or less adequate response, the request should contain:

• role,
• task,
• question,
• context,
• examples.

The cardinal rule is that your instruction must be impossible to interpret in any other way.

These are the basic functions that make up the operation of Chat as a whole. If you understand the general mechanics of how it works, interaction will become easier.

**Searching for information ** If you don’t want to google and resurvey information yourself, you can leave it to ChatGPT.

Data processing

Chat can handle different types of data, such as: Textual: analyzing text, tone, extracting keywords and phrases. Numerical: performing mathematical operations, statistical analysis, creating reports and summary tables. Financial: analyzing stock market data, forecasting prices, portfolio management.

The ability to process data depends on the instructions available and the context of the task.

In the example below, I asked ChatGPT to make a summary of an article and create a table based on the material.

Information collection

A chatbot can be tasked with collecting certain information based on a given example. For example, it can collect a list of books, movies, or, as for my example, hair salons.

If there is a variable in the data that changes, it should be put in square brackets. Then chatbot will insert the necessary values there.

Semantic Analysis

ChatGPT is able to interpret meaning and understand what concepts are embedded in a message. For example, it can analyze the emotionality of a text.

Information generation

This includes brainstorming, writing texts, explaining terms, etc.

Saving interaction history

Chat remembers everything you texted and keeps the dialog within a given concept. It learns as the conversation progresses, which facilitates interaction and improves the quality of responses.

That’s why you should create separate dialogs for different tasks with different contexts. Otherwise, it will mix all your information into a pile, and the answers will become useless.

Next, I’ll share specific AI use cases. These are personal and work tasks often handled by ChatGPT for me. They may be different from yours, but they will help you understand how to interact with ChatGPT and customize requests to your needs.

Email

I often conduct business correspondence using emails, and I also sometimes get sent various offers, information about projects, etc. In order not to waste time but not to miss anything important, I ask ChatGPT to summarize the email and compose a response.

Meeting report

I often have calls with clients at work, and to capture the results of the meeting, I write notes as the conversation progresses. It’s not very professional to send a meeting report like this, so I ask ChatGPT to “ pack” it all into a nice, formal message.

Generating ideas

You can generate completely different ideas, up to the theme of the birthday party of a 10-year-old child. Yes, not all ideas will be viable, some options are completely crazy or, on the contrary, too banal. And this is where context and examples help: the more information you provide, the more accurate the result will be.

Generating ideas

You can generate completely different ideas, up to the theme of the birthday party of a 10-year-old child. Yes, not all ideas will be viable, some options are completely crazy or, on the contrary, too banal. And this is where context and examples help: the more information you provide, the more accurate the result will be.

Emoji, visual ideas

Chat can select appropriate emoji and suggest design concept ideas for the same old email newsletters.

Analysis and conclusions

ChatGPT can not just compare numbers, but draw a logical conclusion. That is, you will get a justification of what happened and why. For example, here I asked you to analyze the time costs of tasks on a project to see which tasks are out of the budget.

Mentoring

ChatGPT can be your coach and help you reach your goals. Of course, this interaction is very different from real mentoring, but can help you see the path to your goal and points of growth.

You can also use Chat to learn languages (including programming languages), ask it to make a workout plan, share tips for effective communication, etc.

Congratulations

In our company we regularly congratulate our employees on their birthdays. When your head is busy with work tasks, it’s hard to find the right words to heartily congratulate your colleague. That’s why I tried to assign this task to Chat as well.

To make the congratulations look like my messages, I sent about 15 examples of my texts to Chat. It analyzed my answers and then issued its own version of congratulations, which was right on target.

Of course, there are many more ways to use ChatGPT. I just gave examples of my real tasks so that you could see the logic of interacting with ChatGPT.

Let’s summarize what you need to do to get a close to perfect answer:

• add context and any details you think are important to the request;
• show Chat examples of your personal texts, or other content, so that it can learn from them and adjust to your style.

If you understand how ChatGPT works, you can decompose and adapt almost any of your tasks to delegate them to Chat (in whole or in part).

For example, here’s what a step-by-step plan for finding a job with ChatGPT might look like.

ChatGPT can be wrong.

And it can also generate fictitious facts or provide irrelevant information. To avoid falling victim to AI hallucinations, don’t take Chat’s word for it and try to double-check the facts it provides. Avoid ambiguous wording so that it doesn’t have to “make things up”. And better yet, do not provide ChatGPT with any sensitive data.

Embracing AI is the future; are you brave enough to redefine your limits with an ever-evolving companion?