GPT: A Software that changes the World
GPT-3 is one such development that gains momentum within a few months. The program will have a huge impact on the economy and society. In December 2022 I participated in a webinar with about 300 people, mostly professionals. Almost all participants were not surprised, but shocked by what GPT-3 actually does.
What is GPT?
GPT-3 is an acronym for Generative Pretrained Transformer, version 3. It was developed by OpenAI Inc. non-profit, which controls its for-profit subsidiary OpenAI LP. The financier and source of ideas for the company was Elon Musk, who left OpenAI in 2018 due to conflicts of interest with the development of artificial intelligence at Tesla. Another funder, Microsoft, “donated” $1 billion for development in 2019 and acquired exclusive rights to GPT-3 in September 2020. This at least shows who will benefit in the long run.
Software has the best chance of creating a Google-like monopoly. There is no doubt that the program can make a significant contribution to the common good. But this assumes that the daughter does not misuse what her mother so generously developed for the public sector as a machine for communicating data and testing “real” applications.
Faster, better and smarter than humans?
To be clear, this is not a system that has a conscience, a soul, or a personality. In fact, all this is still a long way off. What the system can do, however, is complete the beginnings of poems or novels, as well as scenarios. It can help with mathematical tasks, for example, at school or university – and, if asked, also explain why the solution is the right one.
The program completes the task of programming a website for an online pizza delivery service with ten pizzas to choose from – including a few photos of your choice – in about 25 seconds. Also, it can search and find errors in complex codes. With some specifications, it develops business models or consulting simulations.
GPT-3 compares and evaluates text clips (including your own) and determines who copied them and where. Even with a little imagination, it is doubtful who can use this function and where it leads. One of the exams consists of writing a short essay with a somewhat unusual specification. College professors say that many of these essays are better than what their students submit on average.
How does GPT work?
It is important to understand that the system does not write the same article every time, nor does it “cut and paste” sentence components that are “ready” for reassembly. The procedure is completely different.
Roughly speaking, words are represented as rows of numbers. These so-called decorations, for example with “new”, are associated with other marriages with a certain probability, for example with “York”.
So the trick is to calculate the next word after it. The probability space is now reduced by about 175 billion parameters per word and the sentence is computed in the respective context. Such is the power of a neural network that, although it is not (yet!) “connected” to the global knowledge of the Internet, it has 45 terabytes of data at its disposal. This corresponds to a text group of about a trillion words. A translation unit that generates text in any language is connected to a sentence output.
The main problem: how to identify artificially generated data
I cannot help but point out the consequences. Self-written, bachelor’s and master’s theses are a thing of the past, as are many scientific texts that the system writes independently – automatically gaining new insights for us. Creativity must be reassessed, including in the entertainment sector.
One of the main problems is quantifying the huge amount of new data that is being created artificially. Because “traditional” AI systems, which are already used all over the world, are trained with the help of big data from the network. If these systems are trained with unconfirmed GPT-3 synthetic data and spoofs, it can have a significant impact on decision-making processes and the output of other AI systems.
Legal provisions are extremely necessary
There is therefore an urgent need to develop legal regulations for the use of data, synthetic scripts and software – and when and how systems similar to GPT-3 can be used in training, studies and business. What we have before us, without falling into the usual science fiction pattern, is a realignment of what creativity and learning mean in reality: in the book market, as well as in science, business and everyday life.
Missed the latest editions?
On our topics page, you can find all available Terra-X products by reading the columns.