Welcome, humans.

Did you know that behind every answer you get from ChatGPT, Claude, or Gemini, there’s an invisible army of humans?

They label images, rate answers, and teach these models right from wrong.

In fact, Invisible Technologies just raised $100 million for this kind of work, and has already trained 80% of the world's top AI models. Whoa.

See, most people think AI just scrapes the internet and magically “learns.” But what does it mean to actually “train” an AI? And if 40% of its references are from Reddit, how do you stop it from going, IDK, completely off the rails? More importantly, what happens when companies realize their own data is a “god-awful mess”?

In our latest podcast episode, we talk with Caspar Eliot from Invisible Technologies to pull back the curtain on the human side of AI:

Here’s some of our favorite moments:

• (0:59) The "Enthusiastic Teenager" analogy for understanding how LLMs actually learn.

• (5:13) Why AI is fundamentally just "better predictive text" (and has no concept of "truth").

• (6:16) Why 40% of AI references come from Reddit (and why that's a problem).

• (17:28) A practical example: How the Charlotte Hornets are using AI to scout basketball players.

• (26:21) The new "data labeler": Why training AI now requires PhDs, lawyers, and nurses.

• (34:02) The next frontier: Why AI is about to move from the digital to the physical world (robots & autonomous vehicles).

• (45:19) Why a professional League of Legends player became a top-tier ML engineer.

• (47:03) The 3 biggest (and most common) mistakes companies make when trying to adopt AI.

So who is Invisible Technologies? In four words: they make AI work. 

Their platform cleans, labels, and structures company data so it’s ready for AI. It adapts models to each business and adds human expertise when needed — the same approach used to improve models for over 80% of the world’s top AI companies, including Microsoft, AWS, and Cohere.

Their successes span industries from supply chain automation for Swiss Gear, to AI-enabled naval simulations with SAIC, and validating NBA draft picks for the Charlotte Hornets. 

And get this: Invisible has been profitable for over half a decade, was ranked #2 fastest-growing AI company in 2024, and like we said earlier, recently raised $100M to advance its platform technology.

Bottom line: AI isn't magic. It's a massive, complex human process—and access to high-quality, human-curated data is quickly becoming the biggest moat in the industry.

Watch now: YouTube | Spotify | Apple Podcasts

Dive deeper with these resources:

• Learn more about Invisible Technologies

• A deep dive into Reinforcement Learning from Human Feedback (RLHF)

• Case Study: How the Charlotte Hornets are using AI

• Check out the sponsor of this video, Warp.dev!

Stay curious,

The Neuron Team

P.S. This interview is worth watching because it demystifies AI by revealing the crucial, complex human-in-the-loop processes (like evaluation and data labeling) required to make models work, arguing that the future of AI isn't digital magic but a human-centric “evaluation problem” that is rapidly moving into the physical world.

P.P.S: Love the newsletter, but don’t want to receive these podcast announcement emails? Don’t unsubscribe — adjust your preferences to opt out of them here instead.