Why Should I care?
If you want to skip to the meat, jump to 'Today', you might want to read the 'Wolfram|Alpha' section as well.
Young Wolfram
At the age of 15 he began research in applied quantum field theory and particle physics. He went on to get published in 4 separate, professional scientific journals. By the age of 18 he wrote 10 academic, pier reviewed papers relating to his field of study. To say the least he was a prodigy.
CEO of Wolfram Research
Steven Wolfram is a physicist turned software genius. At the age of 24 he dropped the academia of particle physics for the school of Natural Sciences. His new field of study was cellular automata with computer simulations. Once he left academia in 1987 he founded Wolfram Research, Inc. They launched Wolfram Alpha in 2009.
Wolfram|Alpha
To keep this short Wolfram|Alpha is a computation model + database. It's goal is to generate logical computed answers from built-in data, algorithms and methods. The difference between this and chatGPT is there is no input to the data from users. The output of W|A is intended to be dry, factual data. Its purpose is to replace search engines and become a more effective way to access un-opinionated information.
Interesting fact: Wolfram|Alpha is used by Siri as part of its knowledge base.
Today
ChatGPT and Wolfram|Alpha
Stephen W. compares cGPT to "human like" computation wherein its understanding of logic comes purely from language. He posits this as a flaw. The logic accuracy of ChatGPT is limited not only by its textual training data, reinforcement training, but also by its generative token and probabilistic choices. Wolfram compares this to a human without scientific and mathematical skills.
In his paper on the subject we see a few examples of how chatGPT fails on basic logic-based questions. It responds in a convincing essay style response, while the answers are completely wrong. One question to cGPT was: "What is 3 to the power of 73?"
ChatGPT: "3 to the power of 73 is approximately 14,091,714,236..." + some essay about the response
The correct integer is ~67.58 decillion, chatGPT was completely wrong.
*Wolfram responds with the correct exact integer
It's here that the concept is clear, combine the language model of ChatGPT with the super-computation of Wolfram|Alpha. The crisp, precise nature of W|A logic, responses and the combination of cGPTs language based generative text might be the next step toward AGI.
If you have some patience I would recommend you read the paper from Wolfram himself. He has a slight ego you have to read around, however.
Implementation
Some projects have already integrated some generative text (chatGPT) and answer engines (Wolfram|Alpha)
Though the generative response is a bit light, and slow. I would say it's more a display of the limit of chatGPT or the LangChain API used to combine the AI models.
Sources
Stephen Wolfram: "Wolfram|Alpha as the Way to Bring Computational Knowledge Superpowers to ChatGPT"
Top comments (0)