This week in AI news, we talk about education, an AI arms race, and a very dark side of AI training. 

First up, A 22-year-old has created an app that claims to detect text generated by ChatGPT. The tool is called GPTZero, and it was created by Edward Tian, a senior at Princeton University, to combat the misuse of AI technology. Tian believes AI is at an inflection point and has the potential to be "incredible" but also "terrifying." The app works by looking at two variables in a text: “perplexity” and “burstiness,” and it assigns each variable a score. First, the app measures its familiarity with the text presented, given what it has seen during training. The less familiar it is, then, the higher the text's perplexity is, meaning it's more likely to be human-written,  according to Tian. It then measures burstiness by scanning the text to see how variable it is; if it varies a lot, it's likely to be human-written. Tian's app aims to incentivize originality in human writing and prevent the "Hallmarkization of everything" where all written communication becomes formulaic and wit. Paul and Mike discuss what this means, ethical issues, and opportunities and challenges for this tool. 

Next up, this week, Google staked its position in the AI arms race by announcing its commitment to dozens of new advancements in 2023. The New York Times reported that Google’s founders, Larry Page and Sergey Brin, were called in to refine the company’s AI strategy in response to threats like ChatGPT and major players like Microsoft, who just formally announced its multi-billion-dollar partnership with OpenAI. According to the Times, Google now intends to launch more than 20 new AI-powered products and demonstrate a version of its search engine with chatbot features this year.

And finally, a new investigative report reveals the dark side of training AI models. A recent investigation by Time found that OpenAI used outsourced Kenyan laborers earning less than $2 per hour to make ChatGPT less toxic. That included having workers review and label large amounts of disturbing text, including violent, sexist, and racist remarks, to teach the platform what constituted unsafe outputs. Some workers reported serious mental trauma resulting from the work, which was eventually suspended by OpenAI and Sama, the outsourcing company involved, due to the damage to workers and the negative press. 

As Paul put it in a recent LinkedIn post, this raises larger questions about how AI is trained: “There are people, often in faraway places, whose livelihoods depend on them exploring the darkest sides of humanity every day. Their jobs are to read, review and watch content no one should have to see.”