She used 10 years of diaries to train GPT-3 and talk to herself in childhood. Netizen: The most healing application of AI

"This is the best and most healing application I have heard about AI so far."

What kind of application can make netizens give such high praise? evaluate?

It turns out that Michelle, a netizen with great imagination, used GPT-3 to create a lifelike "Childhood Michelle".

Then she chatted with her childhood self, and the other person even wrote a letter.

The "learning materials" of "Childhood Michelle" are also very interesting -

is Michelle's own diary, and it has been written almost every day for more than ten years. The kind they all write.

The diary contains her happiness and dreams, as well as fears and complaints; there are also many little secrets, including being so nervous that she felt dizzy when chatting with Crush...

(I really don’t like writing diaries. I fell to my knees...)

Reading the thick stack of diaries was not in vain. "Childhood Michelle" not only knew the past Michelle well, but even accurately predicted the fields she would be interested in later.

"Childhood Michelle" said that she was very proud of who she is now, but also said that she would have to do homework and practice piano for a while.

Michelle posted this wonderful experience online, which immediately attracted a lot of attention. Currently, this tweet has received 50,000 likes.

Some people think of their deceased grandma, who also loved to write diaries. So if you can use this to create a similar AI friend, that would be really nice.

Speaking of which, what exactly did Michelle and AI talk about?

Conversation with Childhood Self

Michelle herself was very curious about this "Childhood Michelle", so she first asked a bunch of questions to the AI.

Here are some interesting questions and answers.

Michelle asked:

How do you view time?

AI Answer:

I think time is like a reliable friend. It has always been there and passed steadily at a certain speed, regardless of my will.

△Present Michelle is herself, Young Michelle is AI

Michelle asked:

Which one is more important, freedom or love?

AI answer:

are all important. But if I had to choose one between the two, I would choose love. Because love helps us in any situation, and freedom sometimes makes people lonely.

Michelle asked:

What systemic problems do you think exist in society now?

AI Answer:

I think an obvious problem is that people think about mental health in the wrong way. We need to help people with mental illness more appropriately, rather than labeling them in a biased way.

Michelle asked:

What should be done to make people view mental health better?

AI Answer:

On the one hand, technicians need to develop more effective disease treatment plans; on the other hand, everyone needs to carry out positive and reasonable publicity about mental illness and destigmatize it.

This, especially the questions and answers from the last two groups, surprised Michielle.

Because when she asked the question, she did not guide AI to focus on mental health, and her childhood diary did not clearly state that she wanted to work in this field when she grew up, but now this is indeed what she is most interested in. field of.

Michielle pointed out that this AI’s reasoning ability is also very good.

In addition to these, the AI ​​also asked Michelle questions in turn, such as:

The AI ​​asked her how she was doing, especially what she did after writing the last diary during that period. .

Michelle said that she has been working hard to realize her childhood dream. But now, she still occasionally questions whether she made the right choice.

As a result, AI gently encouraged you like a close friend:

I am proud of what you have accomplished, because it is not easy. ;What you are doing is great and I hope you stick to it.

After some exchanges, Michelle said that "Michelle in her childhood" could actually teach her a lot of truths.

And AI also gave a heart-warming reply: "I think you do, too! "

Seeing this, not only Michelle, but also The netizens who were watching felt very healed.

How to DIY your own AI friend

As for how to make this AI friend, Michelle attached a tutorial, which is mainly divided into five steps:

The first step is to collect raw data materials used to train the model.

Any text that can show your personality, hobbies, values, etc. will do.

It doesn’t matter if you don’t keep a diary much—archived chat logs are great material, too.

The second step is to open GPT-3 Playground.

(See the link at the end of the article for the portal)

This is a new feature released by Open AI in November last year, allowing everyone to set their own parameters to train the model.

You need to have an Open AI account and set up a payment method, because this feature is not completely free.

There are multiple models in GPT-3 Playground. Michelle chose the powerful Da Vinci model to train her AI. The price is about 2 cents per 750 words (1000 tokens).

The third step is to add prompts in GPT-3 Playground to help improve the AI’s personality characteristics.

For example, Michelle wanted "Childhood MIchelle" to be gentle and healing, so she added the prompt: caring and friendly.

Of course you can also DIY a funny version of AI friend, like this:

If you ask the AI:

When did the first airplane take off in the world?

The AI ​​answered the question seriously, but it would also do some last-ditch damage:

I hope they can fly over and take me away.

The fourth step is to add and adjust parameters.

After training the AI ​​with a large amount of materials, in order to achieve ideal results, some parameters need to be added and adjusted appropriately.

Although there are many training materials, it is best to select a few samples that best represent your personality or values ​​and adjust parameters based on them.

The fifth step, click submit - your AI chat friend is born!

Seeing this, friends who want to DIY their own AI, you might as well click on the portal below to give it a try~

In addition, if it were you, what kind of AI chat would you like to have? Friends?

Portal: https://beta.openai.com/docs/guides/completion/prompt-design

Reference link:

[1]https:// twitter.com/michellehuang42/status/1597005489413713921

[2]https://www.iflscience.com/woman-trains-ai-chatbot-on-her-childhood-diaries-so-she-can- speak-to-herself-as-a-child-66449

The above is the detailed content of She used 10 years of diaries to train GPT-3 and talk to herself in childhood. Netizen: The most healing application of AI. For more information, please follow other related articles on the PHP Chinese website!