From our AI Overlords to Little Data: A New Dawn

Pour yourself a drink and dive into this week’s (hopefully mind-blowing) tech and business revelations—where the future meets the now.

May 31, 2024

From the Desk of Our AI Overlords

No AGI for LLMs: There are a few people in the world who, when they talk about AI, we truly listen to. One of them is Yann LeCun (Meta’s AI czar). And Yann will tell you that LLMs won’t get us to AGI: “LLMs have very limited understanding of logic… do not understand the physical world, do not have persistent memory, cannot reason in any reasonable definition of the term, and cannot plan hierarchically.” With that, Yann is back in the lab.

Move Over Llama And Meet Chameleon: Meta’s newest AI model, unlike other models, processes images and text simultaneously, creating unified and versatile content. From answering questions about pictures to writing mixed-media stories, Chameleon does it all in one go (and at least according to Meta, better than others). There is never a dull moment in the world of foundational models.

Diversity Rules: LLMs imitate life in so many ways, including the way we get the best results from them. By using a “Panel-of-Experts” approach, one can boost LLM performance dramatically. By simulating expert discussions, you can reduce errors from 20% to 40% in complex tasks. Here is a concrete example of how to do this for code reviews.

Software No More: LLMs are revolutionizing software creation. Soon, all users will be able to write and modify code, leading to custom scripts, GUIs, and more flexible software. Imagine businesses building tailored apps in-house and users combining features from various tools. All of which brings up an interesting question: What will happen to the software industry?

How to Build With LLMs: Here is a highly practical and insightful account of the lessons learned in building applications on large language models. Such insights remain quite rare.

Foundational model wars over the past 12 months:

→ Source

$500 Beats GPT-4V: By fine-tuning Meta’s small, open-source Llama-3 foundational model and spending only $500 on training, a Stanford researcher developed a version of Llama-3 with vision capabilities performing as well as the 100x larger ChatGPT-4V. As we’ve said before: the future is small when it comes to models.

LLMs Beats Financial Analysts: This quote says it all: “Even without any narrative or industry-specific information, the LLM outperforms financial analysts in its ability to predict earnings changes.” Here is the paper.

Better Listening With AI: Researchers at the University of Washington have paired a set of headphones with advanced AI to enable you to look at the person you are talking to and filter out other noises, making even the noisiest room a space for private conversation. Just remember that other people can still hear you…

Beating AI at Its Own Game:

Anything But AI

Forget “Big Data”: The era of Big Data is over; it’s time to embrace Little Data. Although there is a lot of hype and talk about businesses generating massive datasets, most don’t, and those that do rarely need to process it all at once. It’s about making smarter use of manageable data sizes to drive better decisions, not just handling vast amounts of information.

The Future of Cars is Chinese: Chinese car manufacturer BYD is not only challenging Tesla but also aiming to dominate the entire automotive industry. Its latest hybrid model can travel 2,000 kilometers on a single tank or charge. Road trips have never been more enticing. Meanwhile, Waymo’s self-driving cars have become a major tourist attraction in San Francisco.

Bring ’Em Back, Joe: COVID-19 propelled us into a world of remote work, and some bosses quickly realized they dislike not seeing their employees hunkering down in their cubicles. However, the effort to bring people back might backfire: you get the unmotivated ones back, while the productive ones might prefer to stay at home.

How Google Works: A leaked document reveals how the Google Search algorithm functions, offering interesting insights into the inner workings of the world’s dominant search engine. Expect SEO firms to jump on this, continuing the cat-and-mouse game between search engine optimization and Google ad infinitum.

Starlink’s Sky-High Disruption Continues: Starlink has revolutionized the space industry by launching thousands of satellites and becoming the largest constellation in orbit. The disruptive venture by SpaceX not only expanded their market but also challenged traditional satellite operators and military communication systems. A powerful example of how tech bets, when well-played, can redefine markets and propel innovation.

What We Are Reading

🤷‍♂️ Research: What Companies Don’t Know About How Workers Use AI Research shows that leaders do not fully understand their employees’ use of and readiness for AI. In addition to navigating a vast and ever-changing landscape of tools, organizations must also facilitate a significant cultural shift within their workforce. @Jane

🧘 The Life Sabbatical: Is Doing Absolutely Nothing The Secret Of Happiness? Taking time to do nothing can be surprisingly productive and rejuvenating despite initial resistance. We often confuse resting with simply sitting down, but truly resetting is much more than that. @Mafe

🔄 A New Philosophy Of Productivity For those who can’t quite “do nothing”: Deep Work guru and computer scientist Cal Newport proposes three keys to combatting busyness while achieving sustainably high performance via slow productivity. @Jeffrey

🔚 It’s The End Of Google Search As We Know It Google Search, Alphabet’s most lucrative service, is undergoing a major transformation driven by Gemini. Will the change be for the better or for the worse? @Pedro

🤦🏻‍♂️ ‘I Was Misidentified As Shoplifter By Facial Recognition Tech’ Facial recognition technology is becoming increasingly common in UK stores. Misidentifications, such as being wrongly accused of shoplifting, highlight the need for careful oversight. Technology should aid, not accuse. @Pascal

The future of coffee is… robots?!

Help us spread the word! We’d love to bring the radical Briefing to more people. Can you share the Briefing with a few people who would enjoy reading our dispatch?

Share radical Briefing