ChatCAK Technology, a Revolutionary Follow-On to ChatGPT

Santa Clara University’s Robotic Systems Laboratory (RSL) announced today the beta release of ChatCAK, an innovative alternative to ChatGPT. Compared to ChatGPT, ChatCAK technology is fundamentally different, but the two technologies have complementary advantages. How they will work together in SCU classrooms has yet to be seen, but their potential synergies hold great promises for new innovations in education.

ChatGPT, released in late 2022 by the company OpenAI, has caught the world’s attention due to its impressive ability to respond to natural language inquiries. The service is able to produce remarkable results that range from answering obscure questions to summarizing topics to writing compelling essays. ChatGPT is a form of “Generative Artificial Intelligence,” in which computer-implemented algorithms are used to create new works – essays, artwork, music and more – from collections of prior work upon which the algorithms have been ‘trained.’ Some extraordinary results from ChatGPT have included engaging in eerily human-like conversations, composing impressive poetry and songs, and even generating new logos and animations since it can write software that generates images!

RSL’s introduction of ChatCAK, introduced at the beginning of April 2023, uses fundamentally different technology compared to ChatGPT. It is a form of organic, neurally-implemented intelligence which can produce all of the types of outputs that ChatGPT is capable of, and more, like complex engineering designs and analyses. It can respond to natural language inquires not only in text form but also via spoken English, mathematical propositions, and diagrammatic expressions. Some particularly impressive ChatCAK results have included designing sophisticated devices, excelling in strategy-based games, and even teaching the nuances of complex topics.

Comparisons of the two systems are stark. To begin, the ChatGPT is based on a machine learning algorithm that generates words by predicting what the most likely next word should be. It does this by being “trained” on large sets of prior written work. A follow-on process using reinforcement learning is used to minimize incorrect or biased output. ChatCAK, on the other hand, is “educated” on principles upon which deliberative reasoning processes are used to produce output. These are very different approaches. The ChatGPT data sets are massive, making their curation a challenge; incorrect or biased information in these data sets can lead to wrong, biased, and disturbing ChatGPT output. In contrast, ChatCAK’s fundamental principles are fewer in number but the process of creating outputs can take a considerable amount of time compared to ChatGPT. Interestingly, ChatCAK has been educated on principles ranging from physics-based cause and effect processes to ethical standards. This means that ChatCAK’s outputs currently have better performance in terms of reasoning about engineering topics and adhering to virtuous norms.

Other comparisons are also of interest. ChatGPT output tends to be very literal, while ChatCAK is able to better handle nuance, to include humor and even (annoyingly) sarcasm. ChatGPT can impressively summarize massive amounts of data on a topic in seconds; ChatCAK can slowly and deliberately teach that topic over the course of weeks. ChatGPT is trained on existing databases of writing, which means it has no information to draw upon for specific current events. ChatCAK, however, ingests data in realtime, allowing it to reason about current events, although admittedly it does this better with news and technology oriented issues than with topics relating to media and the popular press. ChatGPT provides responses which can be incorrect with no cautions, and it is programmed to not respond to certain taboo topics. In fact, the company providing ChatGPT services has a published warning about its output, saying that it “sometimes writes plausible sounding but incorrect or nonsensical answers.” ChatCAK, on the other hand, often produces answers with built-in caveats on the degree of accuracy or assumptions that have been made, and when it simply has no answer it can often describe how the user might be able to find other ways to determine the solution. Finally, ChatGPT is currently functional, available via the web, and operates at a massive scale, serving hundreds of millions of requests daily. ChatCAK, however, is available only with limited access, has significant daily downtime, and can, at best, respond to a few hundred inquiries per day.

Overall, both ChatGPT and ChatCAK have a role in Santa Clara University classrooms. In some sense, use of ChatGPT on campus has been a bit unbridled, with faculty scrambling to establish standards of use. ChatCAK is currently on sabbatical, with plans to introduce it more widely into the classroom in the Fall. Both, however, represent important capabilities, and our hope is that Santa Clara University students will develop expertise in using them, in balancing their advantages, and in identifying appropriate ways to apply their impressive capabilities.

For more information on ChatCAK technology, see the Robotic Systems Laboratory Web Site.

4/1/2023