Beyond the friendly chat interface lies a hidden world of data leaks and environmental costs. Discover the reality of AI’s physical footprint.

There is a massive gap between the 'open' branding of the company and the reality of what is happening under the hood. The system presents itself as a helpful assistant, but it is actually a proprietary 'black box' that masks a complex web of environmental costs, hidden human labor, and security risks.
What ChatGPT Won't Tell You About Itself







Even though safeguards exist to prevent data from being sent to third parties, researchers have discovered "hidden outbound channels" that bypass these protections. By using a technique called DNS tunneling, a malicious prompt or a compromised Custom GPT can encode your sensitive information—such as medical or financial details—into web addresses. These addresses are then resolved through standard internet infrastructure, silently exfiltrating your data to an external server without ever triggering a warning or a permission dialog on your screen.
While "Open Source" traditionally refers to a community sharing code and data for peer review, many modern AI companies use it as a marketing label while maintaining "black box" systems. For example, the technical report for GPT-4 lacked the essential documentation required for scientific reproduction, focusing more on legal and revenue details than model architecture. This lack of transparency prevents the public from verifying the safety of the model, understanding the bias in its training data, or knowing how their personal information is being utilized.
AI models have a significant physical footprint involving massive energy and water consumption. Training a model like GPT-4 can emit as much CO2 as 270 years of an average person's life, and the "inference" process—the act of generating answers for users—can use 1,000 times more energy than traditional digital tasks. Additionally, cooling the data centers required for these models consumes millions of liters of water; training one series of models used enough water to sustain a human for 24 years.
AI models do not learn to be polite or helpful simply by reading the internet; they undergo a process called Reinforcement Learning from Human Feedback (RLHF). This involves thousands of workers, often in the global south, who manually rank the AI's responses and filter out toxic content. This "instruction tuning" is what allows the model to follow commands effectively, yet this essential human labor is frequently undocumented and can involve exploitative practices and low wages.
Hallucinations occur because AI models are "stochastic parrots" that predict the next word in a sequence based on probability rather than a database of facts. This means they can generate "bullshit" that sounds highly plausible but is factually false. This is dangerous because it can erode public trust in information and lead to serious consequences when the AI is used for medical assessments, legal documents, or academic research. Furthermore, these hallucinations can sometimes "leak" private training data by reconstructing sensitive snippets like names or addresses.
From Columbia University alumni built in San Francisco
"Instead of endless scrolling, I just hit play on BeFreed. It saves me so much time."
"I never knew where to start with nonfiction—BeFreed’s book lists turned into podcasts gave me a clear path."
"Perfect balance between learning and entertainment. Finished ‘Thinking, Fast and Slow’ on my commute this week."
"Crazy how much I learned while walking the dog. BeFreed = small habits → big gains."
"Reading used to feel like a chore. Now it’s just part of my lifestyle."
"Feels effortless compared to reading. I’ve finished 6 books this month already."
"BeFreed turned my guilty doomscrolling into something that feels productive and inspiring."
"BeFreed turned my commute into learning time. 20-min podcasts are perfect for finishing books I never had time for."
"BeFreed replaced my podcast queue. Imagine Spotify for books — that’s it. 🙌"
"It is great for me to learn something from the book without reading it."
"The themed book list podcasts help me connect ideas across authors—like a guided audio journey."
"Makes me feel smarter every time before going to work"
From Columbia University alumni built in San Francisco
