🤖 Fido, Fetch me a Neural Network

What’s up team!

Today we’re learning about Neural Networks and how critical dogs are in their development. We’ve also got news and also an app that builds full websites for you with just a prompt.

Tomorrow we’re going to look at Large Learning models being developed outside the ‘West’ 😎

Let’s go!

🗞 News in 30 secs’

👻 Snapchats latest feature release was a flop
Last week Snapchat released a AI Chatbot assistant inside the app. Awesome? Nope.

67% of reviews in the last week ave been 1/5 stars. Users finding the app creepy, intrusive and “weird”. At least I don’t need to spend $500k in API credits to find that out.

It seems a symptom of the AI hypetrain that every company needs an AI feature. But they don’t. And where they do, make it a value add to the core function of your app.

🌋 AI to predict natural disasters.
A team from a UK University have been using AI paired with microphones to get real time data on Tsunamis and other natural disasters

🎶 The Generative AI music debate continues…still
And this time it’s the Spotify CEO telling us generative AI is ‘Cool and Scary’. Wow. What a hot take / thinly veiled marketing ploy. Daniel Ek was commenting on the recent series of announcements and ongoing Drake saga. Spotify are trying to find the balance between protecting artists and encouraging creative freedom.

🧸 Hugging face release ChatGPT Competitor
Hugging face have released HuggingChat a “competitor” to chatGPT. Competitor in quotes as it’s still a few nodes short of a Neural network. Speaking of that…

Wait… What actually is a Neural Network? 🧠

Alright, alright. So yesterday we worked out there were two key inputs to the quality of models like GPT, or Bard etc

1) The size of your dataset

2) The size of your neural network.

But it was late and I left you on the hook as to what a Neural network was. So let’s catch up now.

A Neural network is lots of points connected to other points.

Each ‘point’ is a piece of Data.

The model has been trained to recognise the distance between two points. The more certain a neural network is of the distance between two points, the better it is at problem solving a specific task.

Right example time.

Let’s say we have a text-to-text model. It’s been trained on commands for dogs. The dataset contains inputs like “fetch”, “sit” or “stay” and outputs like “go and get the ball”, “bend back legs”, “stay where you are”.

There’s a concept called “supervised learning” where the model is fed one of the inputs, then has to return an out put.

Each time the model gets this right we tell it that and it get's more confident of the neural connection between the command and instruction.

Cool.

But how does it work for practical models, like OpenAI makes? Real big ones.

Well very similar. Except they used a method called unsupervised learning.

The basically served their model all the information on the internet. Loads of books, websites and right wing blogs.

They then trained the model to understand for each word it says, what is the most likely next word to say based on all that you have read. Rather than just looking for A command 👉 Instruction it’s assessing based on millllliioonns of criteria.

Asked the big man (GPT) what it thought of my summary and it suggested these clarifications:

Thanks, Dad.

What’s Appnin’ 🎮

Alright.

Enough boring stuff for the day. This is seriously cool.

You know that App or product idea you have.

And how it’s gonna change the world.

How easy would it be to make a landing page for it? Like a full landing page with pricing and all that good stuff?

Well…look at this.

The full video is about 20 seconds long, and it shows me generating a landing page for this blog in that time.

Design wise 6/10 - but on my second try I specified Hex codes.

Copy wise 8/10 - would probably use most of it

All the buttons work kinda out the box, and you can customise most of the settings just like you might in webflow.

Shower thoughts 🚿

Websites, content, images can all be generated by AI. It’s easy and gets you 80% of the way to a full website.

When do companies create a mark of “Human Made” vs “AI made”

Would they?

Would you be more likely to buy?