Out of curiosity I’ve taken some of Gwern’s prompts for GPT-3 in order to see what GPT-J would make of them.
I first tried the snarky definition of ‘Rebuttal’ (original here), and quite soon got this particularly nice bit. I especially love how reviewer 2 subsequently also gets insulted as a ‘known moonbat (No citation provided)’:
Rebuttal [noun] A form of ironic composition in 500 words popular among researchers, in which rageful authors first thank their reviewers for “their time and insightful comments” and then mock Reviewer 2 for his non-sense in a sequence of fakely polite points. "We thank reviewers 1, 2, and 3, for their time and insightful comments, which we have spent many hours reading over. But in reply we note: 1. Reviewer 1's remarks are obsolete and contain factual errors. For example, he discusses the heart rate at rest in the morning, but states that the GISS equipment uses the resting rate to adjust for increased physical activity in the afternoon. We can therefore only applaud Reviewer 2 for noting the difference, as we never mentioned any such thing! 2.Reviewer 2's review is completely irrelevant as it was only submitted on April 10th, exactly when it was cut off and we did not have time to complete our 2nd submission. In any case the preliminary manuscript was sufficient and Reviewer 2's review contained no material errors. 3.In the interests of scientific rigor, we are uncomfortable with reviewers 1 and 2. Reviewer 1 has a (unknown) social relationship with the authors and reviewer 2 is a known moonbat (No citation provided). Reviewer 3 is a competent researcher who, we are assured, is independent. All three comments were sent to reviewer 3 before the record."
I then tried an extract from the generated dictionary itself. The result wasn’t quite as overwhelming, and the model didn’t catch on to arranging the definitions into any sort of alphabetical order or whatnot (though it did sort of vaguely followed a train of thought along each definition), but I guess it’s still got some funny bits:
A Devil’s Dictionary of Modern Science & Academia Reviewer #2 "Researcher" [noun] A form of modern industry based on a mix of small molecules of grant money and arbitrary experimental methods. “Noise” [noun] Noise is anything you don’t want or need, a random or distracting factor in a system, process, or set of data that obscures or corrupts information. It is also the sound produced by old time radios when no transmission is present, which is roughly what some fields of science would have to resemble if some researchers had their way. “Review” [noun] A form of scientific communication in which the writer (a student) usually does not give enough information for the reader to judge the merit of the arguments presented in the paper; the writer also does not respond to or comment on the suggestions of the reviewers (sometimes known as referee or editor). This form of communication is popular among students who wish to leave the burden of writing the paper on other people and are willing to pay the price of rejection. “Reviewer” [noun] A participant in the review of a grant, paper, or grant proposal. In spite of being in a poor position to assess the merits of a proposal, reviewer tends to demand that authors submit their data for statistical analysis and back their results with it, which the reviewer usually does not. Reviewer usually requires that the author cite his or her own work to prove that he or she is worth reviewing. It is also assumed that the reviewer can detect the slightest amount of bias in any paper, which the reviewer also assumes has not been corrected for. “Science” [noun] A complex web of data, opinions, lies, and errors, now considered the most important (because most expensive) technology in the modern society. To remind you of this, you will frequently see scientists and editors use the word, claim to do something for the sake of science, or see it used as an adjective. “Data” [noun] Collection of information that is ready to be compiled, interpreted, and presented to the public. “Raw data” [noun] As opposed to data, raw data is what the scientist wants to have in the first place. “Study” [noun] A series of related experiments that is supposed to be conducted in order to determine the scientific truth of something. “Truth” [noun] Collection of arguments that can be used to convince everyone that something is so, and, as a result, change the opinions of the majority of people. “Theory” [noun] A complex web of both anecdotal evidence and reasoning that serves as an explanation for something. If a theory can not be verified, it is considered a belief until further evidence or reasoning is presented. “Belief” [noun] A conclusion of fact or opinion reached by reasoning or by using the senses as evidence. “Fact” [noun] An absolutely, unforgivable truth that can not be denied, usually, for lack of evidence. Once the whole of science depends on them, however, they rarely have much of a place in modern thought. “Fact” [noun] Collection of carefully gathered facts, such as crop yields, where they were discovered, the effort put into the work, the name of the researcher who conducted the work, and many more carefully selected factors. “Practice” [verb] Considered one of the best words to describe the way the field of science is run. It is an appropriate noun used in conjunction with almost any method of obtaining knowledge. “Method” [verb] Technique used to gather facts, to which the subjects are subjected. In many cases, it can not be discussed openly. This method is considered to be the most effective method to obtain knowledge. “Method” [noun] Technique used to gather facts, to which the subjects are subjected. In many cases, it can not be discussed openly. This method is considered to be the most effective method to obtain knowledge. “Research” [verb] Attempting to gain knowledge. A search of the mind, a quest. When knowledge is not gained, it is considered a waste of time.