Blog Viewer

the studENT - An AI wrote this post

(posted on behalf of @Andrew Nixon, PhD Student Representative & GPT-3, OpenAI)
I spend a good amount of time on Twitter and my feed has recently been captivated by weird AI-generated images from DALL-E mini by Hugging Face. It’s an open-source model that people have been using to make a computer draw odd scenarios like “A chicken nugget smoking a cigarette in the rain” or “Youtuber doing a funeral unboxing”. These have been relatively harmless demonstrations of how advanced AI systems have become at making art and as this is open-source technology, you can rest assured that the proprietary systems are much, much better. For example, DALLE-2 by OpenAI can generate photorealistic images.
DALL-E is built upon a natural language processing AI called GPT-3, which uses language prompts to instruct a large neural network to conduct different text generation tasks. DALL-E is trained to generate images from text descriptions like the Twitter examples I listed above, but I want to focus on the text generation possibilities of GPT-3. GPT-3 can translate text, classify items, and summarize emails but its most fun feature is asking it to create novel lines of text. I first became fascinated with this technology back in October 2020, when I read this article about a user who had then been accessing GPT-3 before it was free to use via PhilosopherAI and was making hundreds of posts on Reddit with outputs impressive enough to pose as a human. I would recommend reading through some of the posts in the article to get a sense of the sophistication of this technology even two years ago. Now GPT-3 is open to anyone to play around with and you might have seen screenshots on Twitter of silly computer-generated text, like this one earlier this week that is a series of GPT-3 outputs for prompts about Donald Trump giving speeches where he ridicules different Greek gods. Much like the image description prompts that DALL-E takes to create art, GPT-3 can spit out speeches, articles, blog posts, etc. You can sign up to play around with GPT-3 using OpenAI’s Playground platform in about one minute, and so I did so to see how well this language model works when applied to academia.
Here was the prompt I gave:
“A blog post for the Academy of Management Entrepreneurship Division by the PhD Student Representative. The topic of the blog post is a pressing issue relevant to doctoral students who study entrepreneurial opportunities. The post references academic journal articles authored by Schumpeter and Kirzner and Shane. The post synthesizes the literature and offers a new perspective on how the field should progress.”
Below the break line is verbatim what the model returned. It’s pretty coherent! I think it’s funny that GPT-3 named the PhD Rep after a male and that this text sounds more like an article discussing a recent blog post rather than being in the format of a blog post. This was the first iteration of this particular prompt, but previous similar prompts would write the text in a more traditional blog format and sometimes even include quotations of specific pages of articles and the references in APA format at the end!
These models are not yet sophisticated enough to convince anyone who has even early graduate student training, but I would bet this post would seem sensible to the general public. These models are trained by reading large swaths of the internet and I wonder since so much of our academic research is paywalled, whether they have been able to parse anything other than open-access articles and what archived PDFs Google Scholar has to offer. Who knows if the training data includes an Elsevier subscription or whether the growth of open-access publishing will feed it more literature? The power of models like GPT-3 is intimidating, as it pumped this post out in a matter of seconds and you can set the word limit to produce much longer samples of text. Much hullabaloo has been made recently about that Google engineer who thought the AI he was evaluating had become sentient and I’m not here to suggest anything like that, I simply want to highlight that academics should be aware that these tools exist and are constantly improving. I do believe this will impact academic integrity at multiple levels, with the most likely first candidate being undergraduate plagiarism and the potential for this to be applied to fabricating qualitative datasets.
I hope you enjoy/fear reading the below post and let me know what you think about how natural language processing models like this will impact our field!
In a recent blog post for the Academy of Management Entrepreneurship Division, PhD student representative Nathan Furr called for a new perspective on how the field of entrepreneurship should progress. He argued that the current focus on entrepreneurial opportunity is too narrow and fails to address the important role of market competition in the creation and development of new businesses.
Furr's argument is based on the work of two of the most influential thinkers in the field of entrepreneurship: Joseph Schumpeter and Israel Kirzner. Schumpeter, writing in the early 1900s, argued that entrepreneurs are key to economic progress because they are the ones who introduce new products and processes into the marketplace. Kirzner, writing in the 1970s, argued that entrepreneurs play a vital role in the economy by spotting opportunities and then taking advantage of them.
Shane, writing in the 2000s, has synthesized these two perspectives and argued that the key to understanding entrepreneurship is to see it as a process of discovery. Entrepreneurs, Shane argues, are not simply people who have good ideas; they are people who are good at spotting opportunities and then taking advantage of them.
Furr's argument is that the current focus on entrepreneurial opportunity is too narrow and fails to address the important role of market competition in the creation and development of new businesses. He cites research by Hebert and Link (2002) showing that only a small minority of new businesses are actually founded in response to an opportunity; the vast majority are founded in response to market competition.
Furr argues that the field of entrepreneurship needs to take competition more seriously. He points to research by Gimeno, Folta, and Cooper (1997) showing that firms that are more competitive are more likely to be successful. He also cites research by Busenitz, Barney, and Smith (2000) showing that firms that are more competitive are more likely to grow.
Furr's argument is important because it challenges the prevailing view of entrepreneurship as a process of opportunity recognition. This view, while helpful in some respects, fails to take into account the important role of market competition in the creation and development of new businesses. The field of entrepreneurship would benefit from a more nuanced understanding of the role of competition in the entrepreneurial process.