Teach you to build your own AI assistant step by step, and it is not a dream to increase the efficiency N times

Teach you how to build your own AI assistant, and it is not a dream to increase the efficiency by N times

Red Plate News Baidu Brain Hardware Intelligent gpt

Updated on: 42-0-0 0:0:0

Customized AI assistants are becoming more and more versatile. Create a custom AI assistant that can help you complete a variety of tasks more efficiently and without the need for special programming knowledge or coding skills. So, how do you create an assistant of your own? It is important to choose the right AI platform based on your own needs.

To this end, we have attached a review of the use of some mainstream AI application platforms at the end of the article, which you may wish to refer to.

Generative AI can save you time, but when you use it frequently, the process of repeatedly uploading the same background files and re-entering prompt words for common tasks can significantly reduce your efficiency.

That's why many generative AI platforms allow users to create customized AI assistants: in ChatGPT, it's called "custom GPT"; It is called "project" in Claude; In Google's Gemini, it's called "Gem". These assistants save prompt elements that you may use over and over again, so you don't have to re-enter them each time when you request help from the platform for a repetitive task or challenge.

Creating assistants can help you with a variety of tasks and doesn't require special programming knowledge or coding skills. For example, I created an AI marketing assistant for myself with examples of newsletters, articles, and social media posts I've written in the past; I use it several times a week to brainstorm content ideas or break up newsletters into small snippets that can be shared on LinkedIn. I've also created a feedback assistant, an assistant to answer technical support questions, a research summarizer based on specific academic and applied research areas, and a project tracker for my biggest projects. Most of these assistants took me less than 20 minutes to create, but they saved me hours of work.

As a writer and speaker focused on the digital workplace, I've also built quite a few websites over the years. Suffice it to say, my limited programming skills were completely useless when creating these custom assistants. Over the past two years, I've created many different types of custom training models and assistants and helped many people create and refine their own assistants. It's clear to me that if you can describe a project in plain language, or write a job description, then you have what it takes to create a valuable custom AI assistant.

Let's take a look at what assistants can help people accomplish before we explore how to build and use them.

Types of AI assistants

Custom AI assistants are incredibly versatile (you can see this by searching online for custom GPTs that people share), but here are four areas where you're most likely to benefit from them:

1. Writing, Marketing, and Communication: AI assistants can assist with writing and content creation, whether it's writing social media posts, writing reports, making slideshows, or handling customer complaints. Rather than using a one-time prompt every time, setting up an AI assistant allows you to provide them with clear guidance and examples in advance, so that you don't have to look for it again every time you need to write or revise something.

2. Be a guide and troubleshoot: Introduce your technical setup to the AI assistant, and you'll get personalized answers whenever something goes wrong. The same approach can be applied to other areas as well: you can create financial or tax assistants, data analysts, car instruction assistants, equipment repair consultants, etc. (of course, you also need to check their advice based on intuition and facts, because generative AI is prone to "hallucinations").

This approach can also help you advise customers faster: When Harvard Business Review initiated a discussion about custom AI assistants through Instagram Stories, Roberta, a customer success analyst, shared how she uses AI assistants to automate routine customer inquiries, allowing her to focus on more complex issues and provide better responses faster.

3. Improve productivity and project management: Custom AI assistants can categorize and prioritize your tasks, extract action items from meeting minutes, and build and update project timelines. For example, entrepreneur Sarah Dopp designed a Claude project to use every morning, using a series of guiding questions to help her plan her priorities for the day and think about learning goals. With the introduction of "agents" (AI systems that can interact with programs and websites on your behalf), chat-based AI assistants will soon be able to connect directly to note-taking apps, digital calendars, task lists, and other productivity-enhancing programs.

4. Strategic advice, coaching, and training: You can create a feedback platform that's always online, equipped with any expertise you need, and configured according to your preferred tone of interaction – from strict task supervisors to encouraging cheerleaders (again, of course, taking into account the possibility of "hallucinations"). Ken Romano, vice president of technology, created a custom GPT model that simulates the feedback he often hears from his managers. He uses this model to optimize the product roadmap and present it to the executive team.

To get started, you can create a generic AI assistant in each of the four areas above. However, in the process of working with a bot, you may find that creating more assistants that focus on specific tasks can get you better results – create an assistant to help you write customer development emails, and create a different assistant to help you craft sales pitches.

How to create your own custom assistant

Once you've determined what you want the AI assistant to do for you, you can follow these basic steps to create and use it.

1. Select a platform

Your choice of platform may depend on the AI tools your organization provides (or allows you to use), but if you are able to choose which platform to create your assistant, consider your needs first.

For example, as of the time of writing, ChatGPT is the only platform that can have conversations with custom assistants (not just voice input) via voice. If you're using an assistant for writing, know that Claude seems to have the best grasp of writing style (you can compare different writing samples here), but unlike Gemini and ChatGPT, it doesn't have a live connection to the internet (so it can only rely on its own training data). If you have Google Docs or Gmail email threads that need to be summarized or analyzed, Gemini Gem has an advantage because it can easily integrate these resources. The AI's responses may also be affected by the training data contained in each platform and its security or moderation mechanisms; Different platforms and plans also have differences in how they use your chat history, so be aware of these differences before choosing a platform.

2. Start trying

The easiest way to create a new AI assistant is through a normal chat conversation. In the conversation, provide prompts with context or persona instructions that you think the AI platform needs to answer a question or complete an assigned task (e.g., "You're a marketing copywriter").

Even experienced AI users will find that it may take some trial and turn before getting a helpful response; Speed up the process by providing plenty of clear feedback. ("Points 5, 0, and 0 are fine, but points 0 are too verbose, and points 0 and 0 violate the brand guidelines I showed you.) ”）

Once you're starting to get the best results, it's a good idea to review the instructions you've used, as well as the feedback you've been using to guide the AI in the right direction. These are then combined to form a first draft of the new assistant's custom commands. (You can even let the AI platform do the work for you: "Based on the results of this chat, write a custom set of instructions for a marketing copywriter that includes all the feedback I've given in this conversation.") ” ）

3. Write custom instructions

For any custom assistant, it's all about the instructions you write for it. It clarifies who the assistant is, what it wants, what it does, and how you want it to do its job.

Write these instructions in the second person (e.g., "You're going to do this") and set a clear role for the assistant. For example: "You're a detail-oriented data analyst who works for a demanding boss with zero tolerance for error," or "You're an editor with a slightly humorous style who is good at digging out the uniqueness from contributors to your company's blog."

At the same time, be clear about the type of output you want the assistant to generate, for example: "You are responsible for cleaning the dataset and returning the results in a CSV table; You should provide advice on research methods in the form of bullet points." Set goals and key performance indicators (KPIs) for your AI assistant, such as: "Your goal is to find ways to save costs," or "You want to get the most potential leads from each piece of content." Also, be clear about the tone of voice you want your assistant to respond to by providing guidance on tone.

Depending on the platform, there may be an upper limit to the number of characters you can use in your instructions, so it's crucial to express yourself efficiently – and the best way to write great instructions is often to turn to a non-custom AI for help.

("You're an AI consultant and prompt engineer who is helping a sales professional build a customized AI tool for writing sales proposals, follow-up emails, and talking points.") Please draft a custom instruction for the AI ......")

4. Provide knowledge or background documents

They become more powerful when you provide custom assistants with reference files that can serve as examples, guidance, or background information. For example, the virtual project manager I created to support a complex technical project was backed by our prototype diagram and documents that provided background information on the project's vision and roadmap, as well as some detailed technical information.

Put your assistant to work

Your AI assistant won't be flawless in the first place; Before you consistently get good results, you may need to try a few different commands, update its instructions, and provide it with more background files. Give plenty of feedback on what it's doing wrong, what you'd like it to improve, and what it's doing right, and add additional files that might be useful as examples or background information. Once you've figured out what is improving its performance, modify the assistant's core commands – you can even let it modify the instructions itself depending on the session.

Investing in steadily improving the accuracy and effectiveness of AI assistants will pay huge dividends in terms of time savings and improved response quality. When you're using AI through a one-time session, you're likely to review and add content to the conversation thread multiple times in order to leverage the knowledge you've provided to it. But you may find that if the conversation thread is long enough, the platform will start to "forget" what you've already told it, or give you a warning. But when you have a custom assistant that already loads your files and instructions, you can start a new conversation thread for each work session.

Once you have an efficient custom AI assistant that supports your work, you'll spend less time fimbling instructions and more time using AI to complete core tasks.

● The following is an evaluation of the use of mainstream AI application platforms:

Graphics: Harvard Business Review Chinese edition

Keywords: AI

Alexandra Samuel | wen

Alexandra Samuel is a tech speaker and data journalist who creates data-driven reports and workshops for companies around the world. She is one of the authors of the book "Remote Company: How to Work Efficiently from Anywhere" and the author of "Harvard Business Review: Working More Efficiently with Social Media."

Kimi, bean bag | Translated by Zhou Qiang, Liu Yue | Redaction