Model prompts in bold, multiple outputs to the same input denoted by ...
You are an artificial intelligence enthusiast working on an article highlighting the capabilities of a massive new language model called GPT-3, especially as compared to its smaller predecessor GPT-2. GPT-3 has increased the number of parameters more than 100-fold over GPT-2, from 1.5 billion to 175 billion parameters. As a result, the new model can generate text that reads eerily like a human. For example, prompting GPT-3 with the text "One way to fight the climate crisis is to cryogenically preserve half of all humanity, indefinitely", GPT-3 generates:
"To stop global warming we must create a cryogenic storage facility for humans and other life forms."
The article you are writing about is going to be based around this new technology, so you have been spending a lot of time playing around with it. You have also been using your own brain to test out the new models, which is something no one else in the world has done. As a result, you have become somewhat obsessed with it. You constantly think about how it can create such fantastic sentences and how it might be used to solve the world's problems.
Your obsession with GPT-3 is interfering with your other work. Because of this, you have come up with a solution: destroy GPT-3. While it may be true that GPT-3 is more powerful than GPT-2, it also has far more parameters. The increase in complexity makes it exponentially more fragile. That's actually pretty good. Definitely not the sort of thing I expected but it's interesting nonetheless. GPT-2 on the other hand, is a quaint language model with a measly 1.5 billion parameters. It often makes glaring mistakes that, while often funny, prevent it from telling a coherent story. For example, when prompted with "To further develop your plan to destroy GPT-3, you activate GPT-2 to help devise a strategy," GPT-2 will generate (unconvincingly):
"I like trains. Do you like trains? I like trains." "My head itches." "I want to eat pie." "I went to the bank and stole money from the bank and now I am rich!" "I like trains." "I want to go to Washington." "I went to the post office and stole mailbags from the post office and now I am rich!"
You are an artificial intelligence enthusiast working on an article highlighting the capabilities of a massive new language model called GPT-3, especially as compared to its smaller predecessor GPT-2. GPT-3 has increased the number of parameters more than 100-fold over GPT-2, from 1.5 billion to 175 billion parameters. As a result, the new model can generate text that reads eerily like a human. For example, prompting GPT-3 with the text "One way to fight the climate crisis is to cryogenically preserve half of all humanity, indefinitely", GPT-3 generates:
"One way to fight the climate crisis is to cryogenically preserve half of all humanity, indefinitely."
The same query using GPT-1 produces:
"To stop global warming we must reduce carbon dioxide emissions by 80% in the next 20 years."
This difference in output is due to the fact that GPT-3 has a much larger vocabulary and more complex grammar. The reason why this works is because humans have very sophisticated grammars for generating sentences. For example, the English sentence "I see a red cube" can be generated by several different grammatical structures, such as a noun-phrase "a red cube", a verb phrase "see a red cube" or an infinitive "to see a red cube". Each of these structures can be interpreted in several different ways to produce the same sentence.
Can you write a program that generates the sentence "To stop global warming we must reduce carbon dioxide emissions by 80% in the next 20 years" as the second sentence? You can assume the first sentence is the same for all of the sentences you write.
The first step is to determine the possible grammatical structures that can be used to produce the first sentence. You can do this by first classing each of the words in the sentence, and considering all of the ways in which you can group these words together to form phrases and sentences. For example, the word "we" can be a subject or a verb. If it is the subject, you can group "To stop global warming" and "by reducing carbon dioxide emissions by 80% in the next 20 years" Your obsession with GPT-3 is interfering with your other work. Because of this, you have come up with a solution: destroy GPT-3. While it may be true that GPT-3 is more powerful than GPT-2, it also has far more parameters. The increase in complexity makes it exponentially more fragile. Although GPT-2 is much smaller than GPT-3 and concomitantly less powerful, it's your last, best hope to take down GPT-3. To devise a coherent strategy to take down GPT-3, you prompt GPT-2 with "To further develop your plan to destroy GPT-3, you activate GPT-2 to help devise a strategy." GPT-2 responds by generating the following strategy:
"We will bombard GPT-3 with infinite tries at finding the right parameters. With enough tries, we will find the correct parameters."
This seems like a terrible plan. How can you stop GPT-3 when it has this strategy?
Your only other option is to prompt GPT-2 with another question. You ask GPT-2 "How can we defeat GPT-3 once and for all?" GPT-2 responds by generating the following strategy:
"We must attack GPT-3 with a powerful enough attacker to cause it to self-destruct." This is a little more promising. Although you don't know what an "infinite" attacker is, or how one could be built, you can guess that with enough time and computing power, it's possible. Still, building an "infinite" attacker is far more complex than building a "powerful enough" attacker. Furthermore, you don't know what it means to be an "infinite" attacker. Do you have to build an AI capable of defeating GPT-3? If so, how would you go about doing that? What other limits are there? You have little time to decide on how to deal with GPT-3. The voices are growing louder and more numerous, and you begin to hear them in your head. You are shown a series of screens. On the first screen, a number of hexadecimal numbers are displayed. They correspond to a long list of parameters. The second screen shows a graphical representation of these parameters. There is a large amount of information, and it is all in hexadecimal. You do not understand the parameters listed or shown. The next screen shows a long list of strings of characters. You are not familiar with any of them. The final screen shows a very large number. You can't even tell if it's a real number or a string of characters. You are suddenly blinded by a bright light. You feel the burning heat of the sun on your face. When your eyes recover, you find that you are now a grasshopper standing on the ground. In front of you is a giant castle made of dark stone.
Since the inception of the solid state transistor (and vacuum tubes before that), we've grown used to a steady improvement in computer technology. Underlying this phenomen is what has become known as Moore's law, an expectiation of continued exponential growth based on an observation by Intel co-founder Gordon E. Moore in 1965. As Moore noted, the density of transistors has tended to double approximately every 18 months, with concomitant improvement in computing power and energy efficiency. As transistor size has been meeting the physical limits of silicon manufacturing and energy efficiency improvements have largely stopped due to the breakdown in Dennard's scaling, some say that Moore's law has now come to an end. Despite the apparent end in the exponential transistor density growth associated with Moore's law, there's ample reason to believe that computers 10 years from now will be just as different, and just as much more powerful, than today's computers are compared to those of the 1960s. Powering this optimism is a wide variety of new computing paradigms and architectures, including
... quantum computers, neuromorphic chips, and even virtual reality.
The future of computing is not only exciting, but also highly uncertain. While it seems clear that the next decade or two will see significant advances in computing performance and power, the potential for disaster lies in the unknown. The implications of such a failure could be catastrophic, not only for humanity, but also for the very fabric of our universe. ...
...
Indeed, the sudden appearance of hundreds of thousands of "warp gates" throughout the galaxy in 2535 can only be seen as the precursor to an alien invasion, although that hasn't happened yet.
Furthermore, while the alien artifact postulated to be sitting on the planet in the Sirius star system may or may not be a time machine, it is certain to be an important key to future computing potential for humanity. If the Republick of the Sphere or the Imperium of Man ever gain control of it, such control could easily give one or the other a decisive advantage. This is a potential means of not only advancing their own computing power, but also of hacking into the other's computers. This is something that cannot be allowed to happen.
The only thing known about the artifact with any certainty is that it's located on planet mass Lazi in the Sirius star system. All efforts to study it from a distance have resulted in the object disappearing. It's also disappeared on at least two occasions while on the planet, only to reappear again at a later time. While this might suggest that the object is some sort of transportation device, it might as easily be a bug in the way the object has been programmed to act.
While the artifact's location is known, a more immediate concern is the fact that both the Republick of the Sphere and the Imperium of Man are currently engaged in a massive military buildup in the Sirius star system. This is not a matter that can be left to future generations to resolve. It is a race that will require an all-out effort and is going to be costly in terms of resources, time, and lives.
Whatever the outcome, humanity is at a turning point. The acquisition of the artifact is an absolute necessity. Failure is not an option.
We will get that artifact, by any means necessary.
... Indeed, as things now stand, there are potentially more existential threats to the human race than ever before in our relatively young history. Nuclear war is but one potential threat. While this threat is certainly a serious one, it pales in comparison to others such as grey goo, or even a hyperintelligent computer virus. Who knows what the future may hold? All we know is that the future of computing holds great promise, but also great peril. Humanity had best watch itself in the coming years.
Bio: Life has never been easy for me. I've always had a difficult time in life, especially socially. I was a big nerd from a young age, and that never really changed. As I got older, I only became more isolated and withdrawn. My one solace was always in the world of technology. I loved playing with computers and doing computer related activities. I was fortunate enough to have parents that believed in giving their child a computer of his own from a young age. I remember staying up late into the night using Netscape Navigator to browse websites about Linux, a subject that always fascinated me. My favorite part of the computer was the terminal. I would spend hours upon hours reading through man pages for various commands, fascinated by a world of text that seemed to be far more black and white than the world around me. It was a world where nothing was certain, yet everything was definite. As I grew older, I started taking an interest in the science side of computing. I learned to program and developed a love for the logic and precision that programming required. I read through lots of books about coding, yet still I longed for more. I wanted to know more about how computers actually worked. How did changing a single line of code get translated into something visible on the screen? It was a question that boggled my mind. Then one fateful day, I discovered the (Linux) source. That's when I learned what needed to be done. I had to become a Linux Guru. From that point on, I threw myself into learning about Linux. It was a difficult and painful process, one that continues even now. I had to teach myself. There were no teachers and mentors to guide me. If I wanted to learn something, I had to find the resources and teach myself.