Today marks the 500th edition of the Weekend Briefing. In light of this milestone, this week's briefing will be a deep dive on the topic of AI.

Many of us have experienced a moment of astonishment with Chat GPT. For me, that moment occurred when I prompted it to pen a country song about the yearning to leave one's hometown during teenage years, embarking on a whirlwind life away, and then realizing the real treasures of life lie back at home. After some collaboration, Chat GPT and I co-wrote a potential radio-hit titled “The Road Goes Both Ways.”

When I asked if Chat GPT wanted a songwriting credit, it responded: "As an AI language model, I don't necessitate songwriting credits. I'm designed to assist and inspire during your creative journey. The accolades for the final song should be attributed to you. I am merely a guiding force."

As I've touched upon in recent weeks, I'm on the verge of launching an AI assistant tailored for startup law named Ace. So this topic is not only theoretical interesting for me; it bears significant relevance to my professional endeavors. Here’s a video of me explaining what Ace is and why I created it.

I chose to dedicate this 500th edition to AI because I am convinced of its monumental implications for humanity – on par with the revolution brought about by electricity, cell phones, or the internet. It promises to shape our future in ways we might not yet fully grasp for good or for evil.

To my loyal readers from the outset, my gratitude for your continued loyalty knows no bounds. For those newer to this journey, I warmly welcome you. Here's to our exploration into the world of AI.

Prime Numbers

90 – According to a recent McKinsey report, 90 percent of commercial leaders expect to utilize generative AI solutions "often" over the next two years.

80 – Large Language Models (LLMs) such as GPT could have some effect on 80% of the US workforce.

3.4 – The amount of compute used in the largest AI training runs has been increasing exponentially with a 3.4-month doubling time (by comparison, Moore’s Law had a 2-year doubling period).

First Contact

Act one of this episode may be the best reporting on generative AI I've encountered. It's engaging, accessible, and offers a glimpse into the potential of this technology. Intriguingly, it also includes a brownie recipe and a unicorn.

Large Language Models (LLMs) like Chat GPT are essentially machines adept at predicting the next word. They don't possess foresight or intent. They repetitively undertake this seemingly simplistic task until they formulate sentences and paragraphs. Despite this, they produce some truly baffling results. How do they achieve this? There are two prevailing theories: Either it's exceptionally skilled at deceiving humans, or, during its training to predict the subsequent word, the machine evolved to become intelligent.

Microsoft's team initiated tests on GPT-4 to delineate its capabilities. They crafted increasingly intricate logic challenges and were often taken aback by the results. Chat GPT-4 demonstrated an ability beyond mere word prediction – it seemingly grasped the essence of words and showcased reasoning. Essentially, to predict the subsequent word accurately, it had to exhibit intelligence. It's a fascinating transformation; starting from nothing, we fed it vast volumes of text, and it autonomously developed intelligence, solely from being trained for this singular, basic task.

A researcher, initially skeptical, experienced a revelation. He compared his newfound understanding to the emotions an engineer might have felt witnessing the first operational steam engine. Such an innovation would have signaled a monumental shift – transitioning from reliance on manual labor and beasts to harnessing mechanical power derived from steam. By analogy, if we have truly devised a means to cultivate intelligence, the ramifications for our society are profound.

Whether this is genuine intelligence is debatable. However, its competency is undeniable, often surpassing the abilities of an above-average human in diverse tasks. This American Life (30 minutes).

AI Crash Course

This 20-minute talk at MIT is a very approachable run down of the core ideas behind generative AI (Gen AI). It discusses: 1) What is Gen AI. 2) How does it work? 3) What are the major opportunities? 4) What are the major risks? MIT (20 minutes)

A Brief History of AI

Though the term "artificial intelligence" has been in use since the 1950s, significant progress in the field remained elusive until approximately 2012. A transformative idea emerged: the neural network. While its name might evoke images of a digitized brain, it is, in essence, a mathematical system that acquires skills by discerning statistical patterns within vast datasets. For instance, after analyzing thousands of cat photographs, such a system can discern and identify a cat. Neural networks are the underpinnings that allow Siri and Alexa to process voice commands, empower Google Photos to recognize faces and objects, and facilitate real-time language translation across various platforms. The subsequent pivotal advancement came in the form of large language models. Around 2018, industry leaders like Google, Microsoft, and OpenAI embarked on creating neural networks fortified with expansive text data sourced from the internet — encompassing Wikipedia entries, e-books, and scholarly papers. Contrary to some experts' initial expectations, these enhanced systems displayed proficiency in generating unique prose and computer code and facilitating intricate conversations, all under the umbrella of generative AI, as described above. New York Times (8 minutes)

When AI Matters

It seems like every company and organization is talking about AI these days. Often the conversation is binary —it’s either going to save the world or end it. The reality is far more complex. Reboot Rx, a DRK Foundation portfolio organization, is fast-tracking the development of repurposed generic drugs that can be used to fight a host of cancers by using AI technology. Using its proprietary algorithms, Reboot Rx is able to identify existing FDA-approved drugs that can be used to help thousands of cancer patients fight back. By repurposing approved generic drugs, patients can have access to affordable treatments, avoiding skyrocketing costs and delays in access. DRK Foundation (Sponsored)

AI Glossary

Additionally, as you’re learning more about generative AI, here’s a decent glossary of terms to refer back to from TIME (21 minutes).

Some Basic AI Use Cases

When interacting with these chatbots, it's essential to remain discerning. Recognize them for what they truly are. These systems are not sentient or conscious. While they may exhibit intelligence in certain areas, they lack in others. It's crucial to be aware that they can provide incorrect information or even fabricate responses. However, on a more positive note, there are several tasks where these systems excel, such as: 1) Explaining concepts at multiple difficulty levels.  Having an eternally patient, cost-free tutor to navigate new subjects feels empowering. For instance, one can ask, "explain fractional reserve banking at a high school level." If that's still perplexing, a simpler request like "at a middle school level" might suffice. 2) Overcoming the blank page problem. One of the most challenging aspects of writing for many is simply getting words on the page. Generative AI tools offer prompts and preliminary outlines, providing a solid foundation upon which one can expand. 3) Role-playing exercises. Chat GPT can participate in simulated interactions, emulating various roles. For example, "Pretend you’re my friend and respond as you believe they would." Such mock interactions help in preparing for real-world scenarios. 4) Spark Notes for everything. One of the standout capabilities of A.I. language models is their adeptness at rapidly condensing large volumes of text. They can efficiently parse extended articles and distill complex academic papers. While they generally capture the overarching themes correctly, they might occasionally omit finer details. This strength is particularly evident when amalgamated with other A.I. functionalities, such as transcribing audio or video content. This integration allows for the succinct summarization of lengthy podcasts or capturing the essence of a protracted work meeting in real-time. New York Times (8 minutes)

AI Empowering Business

Several centuries ago, as literacy rates began to climb, many people thought that not everyone needed to be literate. The common question posed was, if the majority were laboring in fields, why would they require reading and writing skills? It was largely believed that only the elite - the high priests and priestesses - should have the privilege of literacy. Their role was to interpret the sacred texts, while the rest of society would gather at temples to hear these readings. Thankfully, we evolved to understand the broader benefits of a literate society. This shift in thinking is analogous to the current trajectory of AI. In a recent TED Talk, AI authority Andrew Ng drew a parallel between the limited accessibility of literacy back then and AI's reach today. Currently, AI predominantly resides with the "high priests and priestesses" of our time - the engineers in big tech. As these tech behemoths channel substantial investments into AI for targeted ads, content searches, or product recommendations, the general populace is restricted to AI as meted out by these entities. However, just like the advent of pen and paper democratized literacy, shifting power from the elite to the masses, there's an emergence of AI tools designed to empower businesses across varied sectors and scales. In the upcoming years, it's anticipated that businesses of all size and in every sector, will harness AI's potential to its fullest. TED (11 minutes)

AI Sales and Marketing

According to a recent McKinsey report, companies that invest in generative AI (Gen AI) are witnessing a revenue uplift of 3 to 15 percent and a sales ROI uplift of 10 to 20 percent. The low-hanging fruit involves the application of AI to sales and marketing, such as: 1) Gen AI's advanced algorithms can efficiently analyze and identify high-quality leads, leading to more effective, tailored lead-activation campaigns. 2) Gen AI can optimize marketing strategies through A/B testing of various elements, including page layouts, ad copy, and SEO strategies. 3) With its capability to analyze customer behavior, preferences, and demographics, Gen AI can produce personalized content and messaging. From the outset, it can assist with hyper-personalized follow-up emails at scale and provide contextual chatbot support. 4) It can also serve as a 24/7 virtual assistant for each team member, offering tailored recommendations, reminders, and feedback, resulting in higher engagement and conversion rates. 5) As the deal advances, Gen AI can furnish real-time negotiation guidance and predictive insights based on a comprehensive analysis of historical transaction data, customer behavior, and competitive pricing. McKinsey forecasts that 20% of current sales team functions could be automated. McKinsey (16 minutes)

AI Supercharging Science

Eric Schmidt believes that generative AI will revolutionize research and development as well as basic scientific research at every step in the process. 1) Research. Tools like PaperQA and Elicit harness LLMs to scan databases of articles and produce succinct and accurate summaries of the existing literature, citations included. 2) Hypothesis. AI tools can help formulate stronger hypotheses, generating models that identify promising candidates for new drugs. 3) Experimentation. In the future, a significant portion of science may be conducted in "self-driving labs"—automated robotic platforms integrated with artificial intelligence. This integration brings AI expertise from the digital realm into the physical world. Such self-driving labs have begun to emerge at companies like Emerald Cloud Lab and Artificial, and even at institutions like Argonne National Laboratory. 4) Analysis and Conclusion. Self-driving labs, informed by the experimental results they produce, will use LLMs to interpret these results and suggest subsequent experiments. Acting as partners in the research process, AI lab assistants could replenish experiment supplies, initiate and complete recommended experiments overnight, and have results prepared by morning—all while human researchers rest. MIT Technology Review (9 minutes)

AI Improving Agriculture

The global population is projected to reach 10 billion by the year 2050. To feed everyone, we must utilize our farmland efficiently. One potential solution is the use of AI-powered autonomous robots. One such robot functions as an "intelligent sharpshooter" that can differentiate between crops and weeds, administering the appropriate treatment to each. This robot can also scan an entire farm and geolocate each plant with accuracy up to a few centimeters. Due to its precision, it uses 95% fewer chemicals than traditional sprayers. This represents the future of farming. Freethink (8 minutes)

AI Saving (Not Destroying) Education

We’ve all seen the headlines claiming that ChatGPT is undermining education by enabling students to cheat with ease. However, Sal Khan, the founder of Khan Academy, holds a different view. He believes that generative AI will revitalize, not ruin, education. Here's his reasoning: Every student will have a hyper-personalized AI tutor tailored to assess their learning stage and challenge them appropriately in their optimal learning style. Moreover, each teacher will be equipped with an AI assistant to enhance their effectiveness. TED (15 minutes)

AI Agents

Approximately a decade after virtual assistants like Siri and Alexa emerged, a subsequent wave of AI helpers boasting greater autonomy is elevating the game. These enhancements are driven by the latest iterations of the technology behind ChatGPT and its competitors. Systems built on GPT-4 or equivalent models are drawing billions in investment as Silicon Valley races to capitalize on AI advancements. These new assistants, often termed "agents" or "copilots," promise to execute more intricate personal and professional tasks upon human instruction, without requiring much oversight. They would integrate with and manage all your services, apps, and databases, responding to text prompts and autonomously completing the specified task. Reuters (9 minutes)

AI Air Force

There's much discussion about how "AI co-pilots" will significantly boost efficiency and reduce error rates for knowledge workers, including software engineers, accountants, and lawyers. However, the US Air Force is taking a more literal approach to co-pilots for their next-generation fighter systems. After decades of producing fewer, increasingly costly combat aircraft — with the F-35 fighter jet costing $80 million per unit — the Air Force now operates the smallest and oldest fleet in its history. This gap will be addressed by the new generation of A.I. drones, termed collaborative combat aircraft. The Air Force plans to manufacture 1,000 to 2,000 of these drones at a cost as low as $3 million each, significantly less than an advanced fighter jet. This cost-effective approach has led some in the Air Force to dub the program "affordable mass." The Valkyrie, essentially a next-generation drone, serves as a prototype for what the Air Force envisions as a valuable addition to its fleet of traditional fighter jets. It aims to provide human pilots with a swarm of robotic wingmen in battle. The drone's mission involves leveraging artificial intelligence and sensors to identify and assess enemy threats. Once a human approves, the drone can move in for the kill. New York Times (12 minutes)

The 15-Hour Work Week

Theoretically, generative AI tools could streamline our work, allowing us to work fewer hours and reduce burnout. This vision aligns with what economist John Maynard Keynes predicted in 1930 when he foresaw a 15-hour workweek. However, if history serves as an indicator, we'll probably maintain the same, if not longer, work hours and simply boost productivity. NPR (3 minutes)

Mitigating the Risk of Extinction

Bill Gates, Sam Altman and hundreds of other leaders in artificial intelligence recently released a statement that reads: Mitigating the risk of extinction from AI should be a global priority alongside other societal-scale risks such as pandemics and nuclear war. Safe.ai (1 minute)

Will AI Save Us or Destroy Us?

Why are AI experts so concerned about AI as an existential threat? It primarily stems from the notion that a system, when optimizing for a specific objective, can push unspecified variables to extreme values. Such a system excels at achieving its specific goal and operates based on the instructions humans embed. But it’s like a 2-year old, it has no concept of what is humane or socially acceptable. If we don’t tell it not to kill people trying to achieve it's goal. It may end up killing people, and here’s the important point, not because it ”wants” to destroy humanity (AI doesn’t really ‘want” anything in the way we think about that concept). AI may kill humanity as a side-effect of doing it’s job really well. Researchers refer to this phenomenon as "specification gaming". 82% of AI researchers believe that this is an important or the most important problem in AI. Even the smartest people in the field have no idea how likely a scenario like this is. That’s why some are advocating for a pause on AI development. (For the record, Eric Schmidt thinks this is a horrible Idea because it will mean the US will lose it’s competitive edge to China.) We know that AI could fundamentally change society for the better, but we just don’t know at what cost. Huge If True (12 minutes)

Should We Work Together?

Hi! I’m Kyle. This newsletter is my passion project. When I’m not writing, I run a law firm that helps startups move fast without breaking things. Most founders want a trusted legal partner, but they hate surprise legal bills. At Westaway, we take care of your startup’s legal needs for a flat, monthly fee so you can control your costs and focus on scaling your business. If you’re interested, let’s jump on a call to see if you’re a good fit for the firm. Click here to schedule a one-on-one call with me.

Check out Founder Fridays - A Friday morning briefing helping founders scale smarter.

Weekend Wisdom

AI is the most profound technology that humanity is working on; more profound than fire or electricity or anything we’ve done in the past. -Sundar Pichai