Merry Christmas, Happy Hanukkah, Happy Kwanzaa! Best wishes to you and yours!
This holiday season represents a hope-filled intersection of tradition and promise. For years, I have snail-mailed printed holiday cards to mark the joyful transition between the old and the new. This tradition is poised to take a radically different form soon because 2024 delivered a seismic shift with GenAI.
Short for generative artificial intelligence, GenAI products use smart algorithms that learn from known data and then produce stunning new content in the form of text, music, photo, art, code and video. A confluence of ingenuity, innovation, audacity and risk-taking behavior inherent in the American DNA fueled AI progress in 2024 and GenAI in particular.
GenAI can create a course outline for teachers and write essays for students. It can create a painting, compose music, write code, solve complex math problems and converse with equal ease. It can write your resume and my blog posts. Everything from doctor-patient communication to legal discovery, drug discovery, scientific research, fraud detection, customer support, and entertainment are rapidly evolving thanks to learning algorithms. GenAI’s explosion is forcing business leaders across the board to consider AI’s impact on the bottom line along with the shape-shifting future of their workforce.
If you are shrugging in confusion or wondering whether GenAI is producing reliable content, you aren’t alone.
The answer is not always but it is correct in a surprisingly large number of cases and pretty darn close in others. It can be grossly wrong too. Consumers of GenAI have to be careful about its use for this reason, knowing that over time it will continue to get better.
The fascinating part is how GenAI is “learning” to begin with.
GenAI was created in an audacious way. The poster child of its success is ChatGPT. If you haven’t used it, don’t end the year without googling it and asking it a question.
300 billion words - the equivalent of 570GB - were fed into ChatGPT’s large language model (LLM). This humongous training dataset came from books, articles, art, music, videos, websites and much more. ChatGPT trains on known data using neural networks that connect neurons or nodes, the elemental computational unit of GenAI. It learns by strengthening the connections between the nodes when the learning is correct or changing the weights to refine the learning process. The layers of the neural network hold knowledge which ChatGPT uses to generate new content.
For years, Facebook has suggested names for faces and asked you to confirm its guess, right? Did you know that your confirmation trained Meta’s LLM, called LLaMA? The company pays exactly zero to use your data and for your services to train their model but they use each click, post, comment, and reaction to train their LLM. Did I say audacious?
The key players in AI are the familiar Big Tech names and companies you will not recognize.
Only a small number of companies have created LLMs. The reason is the astronomically high amount of power that is required. A standard LLM consumes as much electricity as an average household consumes in 120 years! Microsoft recently contracted to have the surviving reactor reopened at the Three Mile Island nuclear power plant because it needs all of that electricity and more to power its AI.
Nvidia, the stock market darling of recent years, started as a chipmaker for gaming systems and became a central purveyor of processing horsepower for AI. Those who rode the meteoric rise of Nvidia are ecstatic. But competition lurks nearby and nobody - not even Nvidia - should become complacent.
Recently Google announced a breakthrough in controlling error rates in quantum computing. This technology uses qubit processors made from synthetic electrons for massive computing power. The classical bit which has powered computing so far uses mutually exclusive states of 0 or 1 to codify data. With quantum computing, you can catch a qubit in mid-spin where it has a probability of being 0 and 1. This leap from one or the other to both will fuel a foundational shift in computing. When error rates are controlled adequately (and other challenges are resolved), we will see unprecedented processing speeds and AI advancement.
Next year, I am likely to use an emerging technology to generate a live action short of our year. Still, analog connections in the form of tight hugs for happy, merry, peace, joy and love will never go away even as a rapidly evolving, uncertain, heartburn-inducing, AI-driven momentum keeps us on our toes.
What became clear in 2024 is this aging, overworked, exhausted and underpaid snail carrying our cards is nearing the end of a long and arduous journey. It is time to give it a huge shoutout and let it rest in peace as we prepare to make the next quantum leap.
Image source: https://tenor.com/view/mail-gif-12335789
Comments