Breakthroughs, Showdowns, and the Unyielding March of AI
It's that time again—a weekend is approaching, AI is doing its thing, and your weekly Briefing invites you to read, click, and contemplate.
The Never-Ending AI-Train
Going 1-Bit: Taming AI’s insatiable energy appetite could be achieved by making LLMs ‘imprecise.’ By fine-tuning and reducing stored probabilities to simple 1-bit representations (1s and 0s), the computational needs of LLMs decrease dramatically while maintaining high performance. The future might be black and white.
Creativity Galore: Scientists pitted LLMs against 100,000 humans in a battle for the title of “most creative.” Our robot overlords outperformed humans in several areas, like divergent associations and creative writing. AIs are truly the bicycle for the mind.
Reasoning is Hard: Turns out, reasoning is not only hard for your toddler but also for state-of-the-art LLMs (reasoning ≠ creativity). Despite seemingly ever-increasing training sets and LLMs beating benchmark after benchmark, a recent paper points out that every LLM still faces massive challenges with basic reasoning. This has led to an important call to action for the AI community to reassess models based on their ability to “make sense.”
AI Goes Small: Want to play around with AI on real AI hardware? Skip the pricey Nvidia chips and get a Raspberry Pi AI Kit for $70. AI has never been so accessible.
Poof Goes the Funding: It seems the bottom is dropping out of the (admittedly overinflated) funding pool for AI startups—seed-stage funding has plummeted by 76%.
Let’s Talk Security: As cool as it may sound (and look), please don’t use Microsoft’s nifty Copilot+Recall feature. It is a complete security nightmare. Surprisingly, if you think about it: What could possibly go wrong with your computer taking a screenshot of everything you do, all the time? And while we’re at it—turns out, it is rather trivial to trick an LLM into giving you any answer you are looking for, including stuff like “how do I build a bomb?” And if all of this isn’t bad enough, did you know that you can use your mobile phone’s Wifi module for mass-surveillance? Brave new world.
Future You Called: Using an AI model of yourself in the future might be what you need to make better decisions today. MIT scientists developed an LLM that lets you talk to an older, hopefully wiser, version of yourself and ask for advice. Maybe it’s not the best idea to have that third drink after all?
Going AWOL: May I suggest that we only attend meetings that are valuable enough to justify meeting in person? But then—looks like American’s are spending more and more time commuting to work.
Not AI
Agile Stumbles: Take this with a grain of salt, as it’s published by a company with a financial stake in the outcome, but a study showed that software projects adopting Agile Manifesto principles have a 268% higher failure rate than those that don’t. Personally, we believe there is much good in the overall ideas around agile but question some specific practices. As always with any dogmatic framework, it’s best to tread carefully and never lose sight of your friend, good old “common sense.”
Your Next Career: Swedish furniture giant IKEA is hiring people to work in their Roblox-powered virtual reality store. That’s right: actual living beings in VR. Your next career move might be into the (gulp) metaverse.
What We Are Reading
🦗 How Cockroaches Spread Around The World To Become The Pest We Know Today Read about the fascinating journey of hitchhiking cockroaches from Southeast Asian origins to worldwide domination. Love them, hate them… yep, most likely hate them! A new study uncovers the captivating tale of their adaptation and resilience spanning thousands of years. @Jane
🌟 The World’s Most Innovative Companies Of 2024 Fast Company’s annual list has the typical suspects like Nvidia and Novo Nordisk, but also the National Women’s Soccer League and Taco Bell! @Mafe
🤔 AGI By 2027? One of our favorite tech expert-skeptics is back to needle the AGI hype again in the wake of yet more revelations about internal conversations and roadmaps at OpenAI. @Jeffrey
🌍 This Classic Game Is Taking On Climate Change A classic’s interpretation of engaging with a relevant topic shows the fine line between information and education and a clear instruction of behavior. Albeit being no proper educational tool, the accessibility it brings to the topic of sustainable energy can be impressive, considering the game’s prominence. @Julian
🚀 How Brands Were Born: A Brief History Of Modern Marketing The 1950s and 1960s saw the birth of modern branding. Here is the story… @Pedro
💸 The Money In Menopause Supplements By now, we all know that the supplement industry, especially in the US, is a hot mess. The ease of starting your own supplement brand highlights the problems and potential dangers. There’s nothing quite like capitalism running amok! @Pascal
The Fun Stuff
Game Over: Robot solve a Rubik’s Cube in 0.305 seconds and play’s foosball better than any human.
From any place to any landmark to any building, ShadeMap will create precise shadows at any time of the day. Here is the Statue of Liberty (switch to “Satellite” as the base map).
Ouch:
