Tech News of the Week: Generative AI
Okay, there has been news about Generative AI for quite a while, but there are a few reports I saw this week that were particularly interesting as it relates to the topic.
- A study published in the Scientific Reports journal finds that the more people use and speak a language, the more difficult it is for a large language model to learn versus a smaller linguistic dataset. Additionally, the language algorithms are challenged between similarly based languages (e.g., Italian and Spanish) versus distantly related languages (e.g., Italian and Arabic). As we start using generative language models more, it’s important to remember that the algorithms behind the technology are still learning. And for largely used languages, some generated verbiage may lack some of the nuances that humans are accustomed to.
- One of the areas of concern around Generative AI is that of content attribution. For example, if I’m a scientist writing a journal article, using a generative model to create a good portion of an article would be relatively easy. It could be challenging for someone to tell if I wrote the paper or copy-and-pasted it from an AI-generated source. There are some AI detectors out there, but as one might imagine, domain-specific writings can be challenging. For scientific-specific AI detection, the University of Kansas has developed a detector that, they claim, can distinguish between AI-generated and human-written content with nearly 100% accuracy. A study in Cell Reports Physical Science goes into additional detail about the AI-generated issue in the sciences.
- While AI is indeed a powerful tool for many areas, text generation, graphics, and so much more. However, as with most advances in technology, AI brings with it associated costs. One of them is the large impact on the power grid. Gartner reports that by 2030, 3.5% of the world’s energy could be used to power AI platforms. Add to that the complexity of the data centers and what it takes to cool them, and adding AI to our workflows becomes a more complex topic.
What are your thoughts on Generative AI? Please feel free to leave a comment below and let me know.
