😺 AI training for beginners

Welcome, humans.

Heads up: the winner of the $500 giveaway has been selected! Congrats to g********@gmail.com who won the prize, and thanks to everyone who filled out our survey!

Now, remember earlier this week when we showed you World Labs? Google just said “hold my beer,” and showed off Genie 2:

Feed it a photo or AI-generated image and Genie 2 creates a fully playable world—complete with physics, NPCs that react to you, and objects you can actually interact with.

The wild part? It learned all this just from watching videos. No programming required. Just show it what you want, and it builds you a mini game world you can play in for up to a minute.

We’ve already selected the first world we want to explore whenever these tools go mainstream: these movie themed cruise ships!

Here’s what you need to know about AI today:

• We round up all the resources you need to train your own model.
• DeepMind’s new AI forecasts weather better than world-class system.
• OpenAI joined Anduril's drone defense project.
• AI music projected to hit 20% of streaming market by 2028.

Advertise in The Neuron here.

Here’s everything you need to learn how to train your own AI in ~10 Youtube videos (or less).

Training Your Own AI Model Is Not As Hard As You (Probably) Think

Ever watched a master chef (or episode of Master Chef) and thought “I could do that”? That's how most of us feel reading about AI breakthroughs every day. Put me in, coach!

And since “How to train my own AI model” is now our #1 most requested guide ever, it's clear you're ready to get off the sidelines, too.

However, we can't exactly give you a complete training guide on such a tricky topic in ~400 words. So here's something better:

The 10 best videos that'll actually teach you how to do it.

First, start with the basics:

1. A 10 min overview (featured above) of how to approach training a model (and a blog to go with it). He uses Vertex AI to train the model, but there’s also Amazon SageMaker, Azure Machine Learning, and the smaller guys.
2. Data structuring is the secret sauce (and hardest part) of model training. This 17 min overview teaches the crucial foundation of preparing your training data.
3. This legendary 2 hour video of Andrej Karpathy building a mini-ChatGPT from scratch is a must watch. Then, watch how Sophia Yang adapts his code to build her own model.
4. What about running the model on your own computer? This video covers all that (the easiest to use is probably LM Studio—here’s a tutorial video on it).
5. Most use-cases need fine-tuning, not from-scratch training. This 5-min guide shows how to fine-tune an existing model (like Llama 3.2) with Ollama and optimize the model with Unsloth.
   1. BONUS: Consider using RAG as an alternative—here’s a great video explainer on that.

For Unsloth users: check out these starter notebooks, this step by step walkthrough of how to use Unsloth with Google Colab, and this blog of it in action.

Now, for the technically ambitious:

1. You’ll need to get comfy with Python.
2. Next, learn the frameworks used to train AI. Watch these 3-min explainers on PyTorch and TensorFlow and CUDA (NVIDIA's GPU computing platform).
3. If those four videos didn’t scare you away, here is a 25 hour (yes, 25 hours) tutorial for PyTorch and a 14 hour series on Tensorflow.
4. If you wanna be an ultra-nerd, and learn how GPTs really work, there’s no better explainer series than this one (~20 min per ep). You can also check out Andrej Karpathy’s 10 episode youtube tutorial (each ep is 1-2 hours).
5. Finally, for the hardware nerds: watch this (we’re eying building our own Mac Mini cluster—here’s a great performance test on it).

For inspiration: check out this guy who built a custom AI for his daughters. He walks through his whole process, and makes it super easy to follow along.

FROM OUR PARTNERS

Want to sell to THE ENTIRE WORLD? (No, seriously.)

Attention just helped Aspire's sales team break down language barriers across seven countries. Their reps speak Indonesian, Japanese, Chinese—and their VP of Sales understands every single call. Magic? Nah, just really good AI.

Attention isn't just another sales tool. It's like having a super-smart assistant who:

• Translates your global sales calls in real-time (goodbye language barriers 👋)
• Turns customer convos into marketing gold (auto-magical case studies, anyone?)
• Makes your help center actually, you know, helpful (by spotting when reps need answers)

Just ask Aspire's VP of Sales, who uses Attention to coach reps in 7 different languages. Or Certificial's team who turned boring sales calls into killer marketing content. Or Gladia, who eliminated 90% of their low-value tasks.

The future of sales is here—and it speaks every language. Check it out here.

Treats To Try.

1. *Check this out: Adobe's upcoming Firefly Video Model lets you create custom holiday scenes that would make a Hallmark movie crew blush. Try it here first!
2. Gamma turns your raw content into polished presentations and documents in seconds.
3. Plot saves and organizes your social media content ideas through a mobile app—so when you see an inspiring TikTok video, you can text the link and have it automatically saved and organized with your other creative ideas.
4. Coval simulates thousands of test scenarios for your voice and chat agents to help you catch errors before they reach customers.
5. Pointer integrates directly into Google Docs as an AI editor that makes real-time suggestions right in your document (video).
6. Artflow Character Builder generates personalized photos and videos of you in any scene or setting using just a few sample pictures of yourself.
7. Replicate lets you run any open-source AI model with one line of code through an API, without managing infrastructure.
8. Animated Drawings is a free research preview from Meta that lets you upload a drawing and then animate it.

See our top 51 AI Tools for Business here!

*This is sponsored content. Advertise in The Neuron here.

Around the Horn.

People are calling this Shipmas, and are forming wishlists. I wonder if we’re on OpenAI’s naughty or nice list…? “On the first day of Shipmas, my true love said to me… giiive us fulll oooo1!!!”

• DeepMind’s new AI weather forecaster called GenCast can now predict the weather 15 days out, and does so faster than the ENS, the best system available today (in minutes instead of hours).
• OpenAI will work with Anduril to power the company’s $200M counterdrone systems software.
• Apple and Baidu (the maker of Ernie Bot, the ChatGPT of China) are clashing over letting Baidu collect user data to train its models, because as of right now, Baidu’s models are struggling to respond accurately to common iPhone queries.
• Humane, the maker of the AI Pin that failed to catch on (and that’s putting it “humane”ly), is pivoting to offer its CosmOS software for phones, cars, and smart speakers (video explainer here).
• A WILD report says AI music will make up 20% of all streaming revenue by 2028—in an “unchanged regulatory framework”, that is (get the report here).

FROM OUR PARTNERS

Turn AI ideas into reality faster.

Join Dell Technologies on Dec 12th to see how Speed Read AI built their platform in record time using Dell Precision workstations with NVIDIA RTX™. Get practical tips to accelerate your own AI projects. Register now!

Under the Hood

HuggingFace link here, or chat with it here.

Thursday Trivia

One is a real, and one is AI. Which is which? (vote below!)

A.

B.

Which is AI?

The answer is below, but place your vote to see how your guess compares to everyone else (no cheating now!)

A.

B.

Here are the results from last week’s trivia (B was AI):

That’s 3 - human, 2 - robots. We’ll see if robots tie it up this week!

Here’s what you said:

• R.M. chose A: “Frustrated lawncare guy here -- I thought A was AI b/c the grass looked too healthy and uniform. I also thought the bricks in B were weathered in a way that was more real than AI would generate. But I was wrong.”
• N.N. chose B: “Picked B because of the weird thing coming out of the tree in the background.”
• S.H. chose B: “There appears to be a deciduous tree in full leaf in B, and also one with bare branches. Now while this could have other causes, such as (a) the picture being set within a tropical climate, or (b) the bare tree eing diseased or dead,the most likely explanation is that AI doesn't understand seasonal changes in trees.” (spoken like a true arborist).

A Cat's Commentary.