AI’s Growing Pains: From Math Medals to Multibillion-Dollar Losses
Plus: The unexpected downside of AI marketing, Ferrari’s deepfake scare, and why judges don’t actually judge differently when hungry
The Thin Wisps of Tomorrow
AI Can Do Math Now — One of the more persistent problems with LLMs is that they can’t do math. Remember: LLMs see the world not as right or wrong, but more or less likely—which doesn’t bode well for math problems. Google’s AI lab DeepMind just achieved silver medal standard in the International Mathematical Olympiad problem set by using some novel approaches to training and constructing their model.
AI Cheat Sheet — Still a little confused as to what all the AI terminology means? Fret not, here is your cheat sheet. And if you are into building your own GPTs, here is a good guide on how to do this (better).
Three Laws of AI — AI and robotics legend Rodney Brooks is one of the people we, here at be radical, love to listen to. His no-BS approach to the subject is refreshing in a world of, kindly said, noise. He just published his three laws of AI.
AI is Expensive — Training large language models and running the infrastructure that enables us to ask AI how to ensure the cheese doesn’t slide off our pizzas is expensive. So expensive that OpenAI is expected to lose an eye-watering $5 billion this year alone. This surely raises some questions about the future of GenAI.
AI is Really Expensive — A recent research paper on arXiv (Scaling Exponents Across Parameterizations and Optimizers) made headlines not only for its insights but the insane amount of money it would take to replicate the study: A staggering $10 million.
AI Doesn’t Sell — By now, you won’t find many products that are not AI-enhanced, enabled, based, or whatever you want to call it. Unfortunately, it appears that labeling your product or service with the AI moniker might not only be foolish (who needs an AI-enabled toothbrush?) but also detrimental to your business: Customers purchase fewer of your products when they encounter the two-letter acronym in your marketing.
Some Ideas Won’t Die — The AI pendant idea refuses to die—now as Open Source (at least). And Google’s Olympic Games ad seems to come straight out of the deep, dark recesses of Black Mirror. On the positive side, blood tests (remember Theranos?) are back. This one can detect Alzheimer’s a staggering 90% of the time. And stem cells made their own little comeback—here they help with knee osteoarthritis.
Deep Fakes are Here — A recent episode of a deepfake-based phishing attack on Italian car manufacturer Ferrari highlights an important truth: You can’t trust a voice anymore. Better be prepared to ask strange questions.
Deinfluencing is the New Influencing — I discussed this topic with a client of ours (a very large consumer goods company) the other day. Influencers, driven by their incentive to maximize sponsorship dollars, are bending the truth in their product reviews and recommendations, leading to a countertrend of “deinfluencing.” And it’s showing in consumer spending.
Self-Driving Cars Woes — Cruise (GM’s autonomous vehicle unit) killed its novel (and actually pretty cool) Origin pod concept. Meanwhile, rival Waymo is suing people for destroying their cars (do we call them car-luddites?).
No, Judges Don’t Judge Differently When They Get Hungry — This is maybe more of a public service announcement, but as I was just at a conference where a speaker (from a highly reputable university nonetheless) kept repeating the same urban myth about judges giving harsher sentences the closer they got to lunchtime (the argument being: their hunger clouds their judgment). As so often in life—it is much more complicated...
What We Are Reading
🔄 How To Salvage A Useful Process That Isn’t Working Anymore The performance of an innovative process deteriorates over time. Before discarding it, organizations should understand the causes. They may then decide to redesign and re-implement the process. @Jane
🦅 How A Crisis For Vultures Led To A Human Disaster: Half A Million Deaths Every animal’s role in nature is crucial. The drastic decline in vulture populations in India led to over half a million excess human deaths due to an increase in environmental pollution and feral dog populations. @Mafe
🪖 The AI Search War Has Begun The AI-enabled search space battle that we’ve been forecasting is now in full swing. Here’s the lay of the land as positions are being consolidated and advantages pressed. @Jeffrey
🤖 Machines Might Not Take Your Job. But They Could Make It Worse While Google’s recent Gemini ad around the Olympics appropriately sparked widespread criticism for questioning some of our most fundamental human meanings, this short discussion on the increasingly independent presence of technology in jobs somewhat aligns with that criticism. The more independent tech, the less meaning for the individual. @Julian
🚗 How Tesla And Its Chinese Competitor Compare, In 4 Charts BYD is challenging Tesla’s global dominance. At one point, BYD briefly surpassed Tesla in sales, and China seems to be the crucial market for both. @Pedro
💰 Why The World’s Greatest Athletes Don’t Get Paid Like It With the Olympics in full swing, here is a glimpse into the financial realities faced by most world-class athletes. Behind the glamour and glory, the costs of training, competing, and maintaining an elite athletic career can be substantial. Combine this with a lack of sponsorship money, and many athletes have to have second and third jobs. @Pascal
The Fun Stuff
📺 Generative AI videos will make the world a better place, no question about it.
🛰️ Mesmerizing: All the world’s satellites in one interactive display...
☕ Ever wondered how decaf coffee is made (and why it is so bad)? Wonder no more, your friendly chemist is here to explain.