Answering Whither Artificial Intelligence By Building A Bot

During one of our call with Yardy, discussing our next venture, we thought about implementing AI to streamline certain functions. Given that I had some experience with Machine Learning, our fund had a project aiming to evaluate ICOs & Coins on specific criteria.

I was quick to conclude: It is too hard and expensive, let us leave it for some later stages of the project.

After a swift debate, we invited a subject matter expert to our conversation. He assured us that Natural Language Processing tools (especially for English), have advanced tremendously, and now, even a person without programming experience can train a model and use it to generate text. His words challenged me. I read about twitter bots before and decided to make my own.

Given

• No hands-on experience with Ai/ML whatsoever
• Little-to none programming experience (I can use mac terminal to install Ghost or launch some apps on AWS lol)

Goal

Aggregate enough of data and teach GPT2 to compose tweets about crypto. Integrate it so curated tweets are published automatically.

Result

I was surprised by how easy it was to find concise and complete instructions on how to do everything I needed.

My primary source of information was the article by Max Wolf. I had issues with the download_tweets python script he provided, but installing and updating a clean version of Twint helped me.

Using Max’s instruction, I gathered 70,000 tweets from prominent crypto people and fine-tuned the GPT2 model.

You can find the result here: @trAIshtalk

To post tweets on schedule, I used a combination of AirTable & Integromat, a super easy and free solution. #NOCODE

What’s next?

As a next step, I want to teach my bot to reply to comments & selected people on twitter, to make it a real part of crypto-community.

What does it all means?

Honestly, living in SF, hearing out of every corner Ai for this, Ai for that, while reading that 40 percent of European startups classified as AI companies don’t actually use artificial intelligence in any way. I was really sceptical about the real advancement and application of the tech.

But at the same time, oh boy: Deepmind’s Ai beats a top go player. AI beats top poker players.

Competes against top starcraft player and 5x5 dota2 match. Games with imperfect information and millions of variables.

https://deepmind.com/blog/article/alphastar-mastering-real-time-strategy-game-starcraft-ii

All that combined gave me the idea that the technology is so far away from real life, from business and accessible only to the smartest and wealthiest companies.

My little experiment made me rethink my attitude towards Ai. If somebody with so little technical knowledge can use it, what actually smart people can do? That’s fascinating.

We quickly redrafted our MVP feature list to add a bunch of NLP & ML tools. Influencers, writers, bloggers, and readers will have access to them in our next project — MetaBias.

PS.

Learn more about Ai — Two-minute papers on YouTube
Follow me on twitter: @DanilMyakin
Ai crypto bot: @trAIshtalk