Why You Should Ignore Every Founder's Story About How They Started Their Company: Catching Unicorns with GLTR

Why You Should Ignore Every Founder's Story About How They Started Their Company: Catching Unicorns with GLTR

When we hear the stories of successful company founders, we often imagine their journey to be filled with overnight success and brilliant ideas. However, the reality is often quite different. Many of these stories are non-malicious lies, designed to create a narrative of quick success and genius.

Take the example of Sam Walton, the founder of Walmart. He opened the first Walmart at the age of 44, after running his own retail stores for over 15 years. His success was not an overnight phenomenon, but rather the result of years of hard work and persistence. Picasso once said, "It took me thirty years to draw that masterpiece in thirty seconds." This quote perfectly captures the truth behind many success stories – they are often decades in the making.

One key lesson we can learn from these stories is that there will always be mistakes and problems on the path to success. As the saying goes, "Obstacles are the way." It is important to not be afraid of making mistakes and to view them as opportunities for growth. Sam Walton himself was less afraid of being wrong than anyone else. When he realized he was heading in the wrong direction, he would simply shake it off and start anew. The idea itself is not the most important factor in success; rather, it is our psychology and persistence that truly matter.

Another interesting aspect of these stories is the role of competition and niche markets. Sam Walton couldn't find investors to start the first Walmart, despite his near-perfect record in retail. This goes to show that even the greatest entrepreneurs face challenges and setbacks. If we want to learn from these founders, we should focus on their journey and how they navigated the start, rather than just looking at the finish line.

Now, let's shift gears and talk about GLTR (Generating Language Through Replacement). GLTR is a tool that allows us to inspect the visual footprint of automatically generated text. It enables a forensic analysis of how likely a text has been generated by an automatic system. Language models, such as the ones used in GLTR, have incredibly accurate distributional estimates of what words may follow in a given context. This means that if a generation system uses a language model and predicts a very likely next word, the generated text will look similar to what a human would have picked in a similar situation, even if the system has limited knowledge about the context itself.

To address this issue, we need forensic techniques to detect automatically generated text. GLTR takes the same models that are used to generate fake text and repurposes them as a tool for detection. By analyzing the ranking of the words that the model knows, we can determine how likely the observed following word is. The tool even provides a display that shows the top 5 predicted words, their associated probabilities, and the position of the following word. It's a fascinating exercise to delve into what the model would have predicted.

While GLTR is a powerful tool, it does have limitations. It can only detect individual cases of automatically generated text and requires advanced knowledge of the language to identify uncommon words that may make sense in a given context. However, despite these limitations, GLTR has the potential to spark the development of similar ideas that can work on a larger scale.

In conclusion, both the stories of successful founders and the development of tools like GLTR teach us valuable lessons. We should not be swayed by the seemingly overnight success stories of founders and instead focus on the years of hard work and persistence that led to their achievements. Additionally, GLTR reminds us of the importance of detecting automatically generated text and the need for forensic techniques in this area.

Three actionable pieces of advice that can be derived from these topics are:

1. Embrace mistakes and view them as opportunities for growth. Don't be afraid to change direction when you realize you're heading in the wrong way. Persistence is key to success.

2. When analyzing text, be mindful of the context and look for unpredictability. Genuine writing often includes words that make sense within the domain, but may not be the most likely predictions of a language model.

3. Foster the development of tools like GLTR that can detect automatically generated text. This can help prevent large-scale abuse and ensure the integrity of written content.

By combining these insights, we can gain a more holistic understanding of the challenges and opportunities in both entrepreneurship and language generation.