back to the Future! Using childhood diaries to train AI, this programmer used GPT-3 to achieve a dialogue with his "past self"

People always have a soft spot for the past and the future.

In the movie "Back to the Future", the protagonist who was in 1985 was accidentally sent back to 1955. Not only did he meet his parents from the past by chance, but he also made his young mother fall in love with him at first sight. , which caused a lot of jokes.

If we move back the time a little bit, we can travel back 10 years If you meet your past self before, what would you tell him?

Recently, a programmer named Michelle Huang shared on Twitter her experience of "traveling" back to the past and talking to herself.

"I trained an AI chatbot using my own childhood diary entries so I could have real-time conversations with my 'inner child.'"

This has also aroused discussion among many netizens.

For example, some netizens suggested, "You can also try to use AI to generate some pictures to match your text, which should be very interesting."

## Some netizens said, “This is great” and “This may be useful for psychological treatment in the future. superior".

5 steps to talk to your past self

I believe many friends are curious about how this is achieved. Michelle Huang also shared a tutorial on Twitter.

The main tool used is OpenAI’s GPT-3.

First, we need some training materials, such as diaries, personal narratives, anything written that showcases your personality/voice/values. If you're not in the habit of keeping a journal, archived chat history should also work.

Huang used photo OCR to scan the diary to upload the data during the process, which also took several hours due to handwriting problems.

Then, we can go to the GPT-3 page, where we can add prompts, diary data, and set training parameters for the model.

Website link:

https://beta.openai.com/playground/

But you need to register an OpenAI account and pay a certain amount. Huang uses the da Vinci model, the most powerful, which costs about 2 cents per 750 words.

Then, we need to prompt production. Here Huang refers to the OpenAI documentation. Scroll down to the "conversation" subheading on the website and you will see a sample conversation prompt:

Website link:

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

If you are worried that your younger self will be yin and yang towards you, you can also adjust your tone/mood/attitude here.

Secondly, you can add parameters. Huang adjusted the temp settings and charactercountup and trained with davincimodel-002.

The parameters are as follows:

##code show as below:

Finally, click Submit. With that, you should be able to start a conversation with your younger self.

Huang said that this project should have many uses in life. For example, you could train the AI ​​in a therapy setting, specifically Parts Therapy, or Internal Family Systems Therapy (IFS); or you could put the AI ​​in an angry mood and then try to comfort it.

Talk to yourself

Huang sighed, overall, this is a very wonderful and at the same time It was also a strangely healing experience. Talking to his younger self reminded Huang of the parts of himself that had remained the same over the years, as well as the parts that had been forgotten or buried as life changed. It was like holding a mirror and revealing the undisguised , a more serious and pure self.

Allowing yourself to have a dialogue with the past through this kind of real historical data is obviously better than any other way. More profound and concrete.

Over the past dozen years, Huang has been in the habit of keeping a diary. The contents range from complaining about heavy homework to being nervous when talking to her crush. Some of them are very anecdotal, and some of them are quite insightful now.

When these diaries were imported into the model as data, some of the reactions given by the AI ​​were very similar to her own reactions in her memory. Huang also asked his younger self about the world view, and then let the AI ​​try to answer the question, "The interaction felt very similar to a normal text message conversation, like I was texting my past self in real time, and it was like It's like using a time machine, but this time machine is disguised as a chat box."

Huang was also surprised that the AI ​​could accurately predict the appearance from diaries from ten years ago. Her interests, of course, have gone through a lot of iteration/trial and error, but it makes people feel that maybe the life path they are choosing now had the seeds planted in the spiritual world a long time ago.

From Huang’s interactions, it’s not difficult to see the therapeutic potential of this medium, not only to send love back to the past, but also to be inspired by your younger self.

Things that troubled us in the past can now be released. This also allows us to end our obsession with regrets about the past and welcome the present and the future with a better attitude.

Related reports:

https://www.dailymail.co.uk/sciencetech/article-11491323/Woman-talks-past-self-trippy-conversation .html

The above is the detailed content of back to the Future! Using childhood diaries to train AI, this programmer used GPT-3 to achieve a dialogue with his "past self". For more information, please follow other related articles on the PHP Chinese website!