Princeton student builds app to detect ChatGPT in essays.
Plus, Google may have a 'fastpass' in AI chatbot race.
In today’s email:
NarcStudent builds app to detect traces of ChatGPT in essays.Google may have a ‘fastpass’ in AI chatbot race.
Write a better social media bio with AI Social Bio.
😎 3 Cool Things
Unfortunate comedian uses ChatGPT to write his entire standup set.
A video from OpenAI about how their text-to-image generator, DALL-E 2, works.
Need a new bio for your social media? Let this AI tool write it for you.
🤓 2 Big Stories About AI
1. Student builds app to detect traces of ChatGPT in essays.
Nobody likes the narc classmate. So Princeton University student Edward Tian is about to become the most unpopular kid at every school after releasing an app that detects AI in essays.
The 🥩 of it:
Edward Tian, a computer science student at Princeton University, saw the power of AI in a class; where his thesis advisor asked students to differentiate between human-written text and AI-generated text.
Many students guessed wrong, and Tian realized that there was a problem that needed a solution.
He spent his winter break working on an application called GPTZero that can determine, with high accuracy, if a text was written by a human or a bot.
GPTZero was created in response to the emergence of ChatGPT, a free online chatbot that can expertly write almost anything, from essays and articles to computer code.
GPTZero works by looking at two variables in a text: perplexity and burstiness, and assigns each of those variables a score.
I asked ChatGPT to explain “burstiness” to me like I was a child, and this is what it says: Burstiness is like a game of hopscotch. Imagine you're playing hopscotch and you're hopping on different numbers on the hopscotch grid. Some numbers you might jump on only once, while others you might jump on multiple times. This is similar to how words are used in a text. Some words might be used a lot, while others might be used only a few times. Burstiness is when the text has a lot of variation in the use of words. Like in hopscotch, if you jump on different numbers a lot, it's more bursty than if you stay on the same number all the time.
The app measures how familiar it is with the text presented and assigns a higher score to texts that are less familiar, indicating that they were written by a human
It also measures burstiness by scanning the text to see how variable it is, as human-written text will have more variation in sentence length and structure
Tian has released a beta version of GPTZero for public use and says it can successfully identify human-written text from AI-generated text with a less than 2% false positive rate
GPTZero was originally created to combat academic plagiarism, but Tian sees it being used to address other issues that will come with the rising popularity of AI, such as online disinformation campaigns.
2. Google may have a ‘fastpass’ in AI chatbot race.
The Innovator's Dilemma is where established companies like Google have difficulty adopting new technologies or business models that disrupt their traditional markets. This can happen when a company has significant resources and capabilities, but their existing customer base and internal processes make it difficult for them to take advantage of breakthrough innovations. The term was coined by author Clayton Christensen in 1995, who went on to write a book he named after it.
A lot of outsiders have been accusing Google of facing this exact dilemma internally, as they seem really slow to develop new AI technology…but maybe their secret sauce has been in one of their subsidiaries all along.
The 🥩 of it:
Deepmind, Google’s British artificial intelligence subsidiary has a chatbot called Sparrow that could launch into beta this year.
The chatbot has been trained with human feedback and is intended to be more helpful, accurate, and harmless than other AI models like ChatGPT from OpenAI.
Unlike ChatGPT, Sparrow will have access to the internet through Google. This will give it a major advantage over ChatGPT, as it notoriously lacks any info after 2021.
The chatbot is based on Deepmind's Chinchilla language model, which outperforms GPT-3 in common language benchmarks.
Google hasn’t released a chatbot similar to ChatGPT yet; mostly due to concerns about security and reliability. However, Google seems to think that Sparrow will be the best solution to solve their concerns and finally enter the AI chatbot race.
🤣 1 LOL
As much as I love the tool, I’m getting really bored with all of the LinkedIn posts about how other people are using ChatGPT. We get it.
