📜 Will AI Get A New Look At Regulation?

Welcome back! This week, we’re diving even deeper into the field of machine learning with a comparison of supervised and unsupervised learning. We’ll walk through the differences between the two—and, while we’re at it, we’ll also show you how to tell the difference between AI and non-AI writing.

What to Expect:

• Supervised vs. Unsupervised Learning

• Quillbot: Your AI Sidekick

• California AI Safety Bill: Visionary Move or Setback?

• Level Up Your Market Research

• More AI Tech, Tools, and Talks

CONCEPT CORNER
Supervised vs. Unsupervised Learning

Now that we’ve covered the basics of Machine Learning (check the post out here if you haven’t seen it), let’s dive a bit deeper into two different types of ML and how they can be used.

➡️ The biggest difference between the two types of learning is the type of data used.

Supervised Learning - You can think of supervised learning as a “guided approach” to ML. This means that the model is shown exactly what the correct answers are using labelled data, so then it can learn from these examples and make future predictions on data it hasn’t seen before.

🧑‍🏫 It’s like a teacher leading a student through a deck of flashcards, where the teacher shows the student all different kinds of pictures of a cats and says “This is a cat.” Based on these repeated instances the student eventually picks up the ability to identify cats when they see one in real life.

• Real-World Example: Spam Detection - You probably get a ton of emails, right? Some are spam, some are important (hopefully we fall in the second category 🤓 ). Using supervised learning, AI is trained on a dataset of labeled emails (spam or not spam). Over time, it learns to identify which ones are spam so that your inbox stays decluttered.

Unsupervised Learning - You can think of unsupervised learning as an “exploratory approach” to ML. This means that the model is given unlabelled data (aka no guidance or flashcards on what it’s supposed to find). The model learns to identify patterns, groupings, or structures on its own, without knowing the “right” answers beforehand. It’s able to do this all through clustering and association—the model will group similar data points and look for any relationships that might be present.

🧩 It’s like solving a jigsaw puzzle without the picture on the box. By trying to fit the pieces together, you will start to see patterns emerge, such as all the blue pieces go together as the sky or all the green ones are part of the grass. Unsupervised learning is the same in that it has to piece together all the bits of data without any clear instructions on what it should find.

• Real-World Example: Customer Segmentation - Ever wonder how stores know exactly what you might want to buy? Using unsupervised learning, retailers can analyze customer purchase data without any pre-labeled categories. The AI then groups customers based on similar buying habits, helping companies target promotions more effectively.

Why Should You Care? 🤔 Understanding these concepts can give you a deeper look into how some of the tech around you works. We guarantee you have encountered a product that leverages one of these methods already this week!

Here are just a few:

• Spotify: Utilizes unsupervised learning to analyze listening habits and create personalized playlists like Discover Weekly.

• Airbnb: Uses unsupervised learning to identify different types of guests and hosts, improving the matching process and enhancing user experience.

• Twitter: Uses unsupervised learning to identify trending topics by clustering similar tweets and hashtags, helping users stay updated with current events.

• Apple: Employs supervised learning in Siri to improve voice recognition and natural language understanding based on labeled voice command data.

• Tesla: Implements supervised learning in its Autopilot feature to identify and react to various driving scenarios using labeled data from thousands of driving hours.

HOW TO: HACKS
Quillbot: Your AI Sidekick

AI is everywhere around us - emails, social feeds, and yes… even on the “expert” news you’re reading from your favorite media outlet. But with so many tools out there, identifying the ones that are actually useful is a challenge. And are LLMs all that AI has to offer? Not really… enter Quillbot, a sleek AI writing assistant, paraphrasing your work, allowing you to sound like an academic researcher or a creative writer, or summarizing the articles you want to pretend to have read.

However, today we will focus on a specific skill: spotting AI-generated text. Because, let’s be honest, in a sea of LLMs and chatbots, knowing what’s really human-made is useful.

🗣️ Example That Speaks Volumes:

Option 1:

Traveling is like hitting the refresh button on your brain—it shakes you out of your routine and drops you into a whole new world of sights, sounds, and tastes. It’s not just about collecting stamps in your passport; it’s about broadening your perspective, meeting people who don’t think or live like you, and realizing the world is way bigger (and weirder) than your daily grind. Plus, nothing beats the thrill of getting lost in a new city or the satisfaction of finally finding that hidden local spot no guidebook ever told you about.

Option 2:

Traveling is like that reboot you do when your phone crashes—it gives you that push you need to get out of your routine and drops you into a world of exploration and new adventures. It’s not just about getting postcards from local villages you didn’t know existed; it’s about diving deep into new cultures, opening up your mind to accept new perspectives and broadening your horizon by meeting people who don’t think or live like you. Only then you can realize the world is a far more interesting and diverse place than you ever knew.

Can you detect which is the AI-generated text? We sure did try to trick you. Validate your answer using Quillbot’s AI-Detection tool.

🤓 How Does It Work:

1. Pattern Recognition: While humans love to switch things up, AI loves its patterns. AI models tend to get stuck using similar sentence structures, often struggling to get out of the repetitive loops. It’s like listening to a great song - only to realize that while the words are changing, the music pattern is the exact same.

2. Word Probability Check: Remember, LLMs do not think, they just place words using the statistical probability of that word fitting there based on what came before it. When an AI Detector checks some text and everything feels polished and predictable.. there’s a good chance it was written by AI.

3. Burstiness & Perplexity: We humans are complicated beings. Why can’t we just keep all our sentences the same length, tone and complexity just like AI does? Well, cause where is the fun on that, right? AI Detectors analyze for “burstiness” (how much the text varies) - and looks for perplexity (how unpredictable it is) to identify AI-generated text.

🤔 Why It Matters: Whether you are a teacher grading your classes creative essays or a curious mind reading the news online about what is happening on the world, authenticity is important - at least we hope it is. In a world where producing output has become so easy, remaining authentic is the real differentiator.

AI NEWS
Is California’s AI Crackdown a Visionary Move or a Tech Setback?

California’s new AI Safety Bill, SB 1047 (aka Safe and Secure Innovation for Frontier Artificial Intelligence Systems Act), is stirring the pot in the tech world. The bill introduces mandatory safety reviews, transparency requirements, and accountability measures for AI companies operating in the state. The goal? To prevent AI misuse and protect public safety.

⌛️ Status: The bill passed the California Senate on August 29, 2024 and has been sent to Governor Newsom, awaiting his action. If signed into law, SB 1047 would represent the most comprehensive AI specific state legislation enacted to date.

📜 What’s in the Bill?

• Safety Reviews: Companies must assess and disclose potential risks of their AI systems.

• Transparency: Firms must reveal how their AI models are trained and what data is used.

• Accountability: Developers could face legal consequences if their AI causes harm.

👨‍💻 Industry Backlash: Tech giants argue that the bill could stifle innovation, create legal uncertainty, and push businesses out of California. They claim that the bill’s requirements are too broad, costly, and could delay progress, especially for startups with limited resources.

 What Would It Mean For AI?

• Chilling Open-Source Development: Open source has been a frontier in AI research and development. When it comes to rapid improvements and breakthroughs, the open source community has long been delivering. However, enhanced regulations could discourage open-source contributions as those are more difficult to control and vet.

• Slow Down Innovation: Navigating complex regulatory environments isn’t easy. Tech giants argue that the bill will significantly slow down innovation, as the focus will now shift to “tiptoeing” around the regulations rather than constantly be improving. What’s more, such a law would be extra harmful to small startups, that can’t foot the bill for lobbying and expensive lawyers.

• Legal Uncertainty: Who will be responsible and for what? Would developers be responsible for unforeseen harm produced by their models? All this uncertainty will harm the rapid growth environment of California and could deter AI companies from setting shop there.

⚖️ Should It Pass? While we are no legal experts and we do love innovation and tech, we do think that stricter regulation is needed around AI. In the past few years we have seen AI do wonders - answering all our everyday questions, enabling breakthroughs in medicine or even in education. However, we have also seen it used for misinformation and fraud, among other of its harmful uses. Besides that, nobody seems to know who and how it is regulated - who is watching the key players? The conversation should be about how regulations will stall innovation - rather it should be about how it will enable safe development. It’s a marathon, not a sprint..

PROMPT OF THE WEEK
For: Market Research

You are an experienced market researcher. Your task is to conduct a comprehensive market analysis to identify potential competitors for [INSERT COMPANY OR PRODUCT]. I want you to produce a comprehensive report that shoul include a detailed profile of each competitor, covering aspects such as their market share, product offerings, pricing strategies, distribution channels, marketing approaches, and customer base. Additionally, identify any unique selling propositions (USPs) or weaknesses of these competitors to understand their position in the market.  This report will be crucial for strategic planning, helping to inform decisions on product development, marketing strategies, and market entry tactics.

TECH TOOLS, TIPS, AND TALKS

MORE READING

• OpenAI Valued at $150Bn in Funding Talks - Will OpenAI do it again?

• OpenAI Releases New Model That.. Thinks? - Is this new model, o1, aa significant leap forward?

• AI in Geo-Politics - What’s China’s AI Big Tech up to?

We want to empower our readers with actually insightful knowledge so that they are more confident, informed leaders. Because let's face it, AI could be running the world pretty soon... so shouldn't we at least know how it works? If you are curious about a topic and want to learn more, drop us a message below👇🏼