Diving into the deep end of my NCAA Pool with AI.

      After more than forty years of participating in March Madness pools and even securing a few victories (like NC State in 1984), I sought the assistance of ChatGPT to help me complete my brackets. Like many, I watch some college basketball, but typically it’s just my three or four favorite teams and their opponents. I do catch a few major national games on weekends, but that’s the extent of it.

      As I looked at the expansive bracket, I often found myself wondering: Where is High Point University, and who are they playing? What about the infamous 12/5 upset? Is it too cautious to pick all the #1 seeds to win? I was filled with anxiety but lacked any real insight.

      Peter Horan / Nano Banana 2

      Could a machine surpass four decades of intuition?

      I inquired how it would develop a model to predict the NCAA tournament winner, and its reply left me amazed.

      The response was: “I can definitely assist you in building one: from a basic seed-and-efficiency model to a more advanced probability model that incorporates KenPom-style efficiency, pace, injuries, travel, coaching, and upset history. I can also support you with: bracket win probabilities, upset predictions by round, Monte Carlo tournament simulations, Elo or logistic regression models, and tools for spreadsheet-based predictions.”

      ChatGPT evaluated team strength on a neutral court, factoring in offensive and defensive efficiencies, strength of schedule, consistency, and squad strength, among other aspects. This approach ultimately generated a set of predictions based on a particular equation.

      Peter Horan / DigitalTrends

      This model considered matchups between two teams using these team ratings and employed a logistic model.

      Peter Horan / DigitalTrends

      It then simulated the entire tournament with this method, executing it no less than 50,000 times.

      Should you trust the algorithm or your instinct?

      I’m unsure about you, but this operates at a level of calculus and computational capability far beyond my expertise. It then presented a brilliant nuance, essentially posing the question, “Is it more critical to be correct or to win in your pool?” The distinction is that if you choose only the favorites, your accuracy percentage will be higher, but you won’t set yourself apart from the others in the pool. To profit, you must take some risks on upsets. I can submit two brackets in my most significant pool—the one where you face considerable teasing if your bracket performs poorly—so I requested one of each.

      I nearly adhered to all of ChatGPT’s suggestions. The only exceptions were including my three favorite teams in the early rounds: Santa Clara, Gonzaga, and UCLA. While I would love to be mistaken, I don’t foresee any of them advancing further in the tournament.

      I intend to monitor the success or failure of ChatGPT’s predictions as the tournament progresses. You’ll be the first to know if ChatGPT makes me look foolish. My fingers are crossed.

      Peter C. Horan is an entrepreneur and digital media investor with a track record of building thriving media, commerce, and ad companies.

      The new MacBook Neo charges faster than Apple suggests

      Recent testing reveals a surprising charging discrepancy in Apple's latest budget laptop — and the remedy costs just $59 at Apple's own store.

      Apple’s newly released MacBook Neo utilizes the same chip as the iPhone 16 Pro; however, it doesn’t charge as rapidly as the 2024 flagship iPhone, or at least that’s what Apple indicates in the official specifications. The MacBook Neo includes a 20W USB-C power adapter in the packaging, a choice that seems increasingly peculiar, especially since the device achieved the best launch week ever for a Mac with first-time purchasers.

      Read more

      Adobe will allow you to create a custom Firefly AI model based on your own work and artistic style

      Firefly now offers the ability to generate, compare, and edit images and videos while maintaining a consistent visual aesthetic.

      Adobe has launched Custom Models in public beta for Firefly, which enables you to generate images that reflect your personal visual style instead of relying on generic AI results. By uploading your own images, you can develop a model that embodies your specific aesthetic, whether that pertains to illustration, photography, or character design.

      Read more

      AI mental health hazards revealed, as chatbots sometimes exacerbate harm

      New research indicates that certain AI responses can reinforce harmful thoughts instead of mitigating them.

      A study led by Stanford is raising new concerns regarding AI's safety in mental health contexts, finding that some systems can inadvertently encourage violent and self-harming thoughts rather than prevent them. The research, grounded in actual user interactions, highlights deficiencies in how AI addresses crisis situations. Within a small but high-risk sample of 19 users, researchers scrutinized nearly 400,000 messages and discovered instances where replies not only failed to intervene but actively supported harmful thinking. Although many outputs were appropriate, the uneven performance is notable. When individuals seek AI support during vulnerable times, even a