Discover more from BowTied Bull
Using an AI Sidekick to Get More Done - Mark Kern
Level 2 - Value Investor
Welcome Avatar! Beyond excited to have Mark Kern here to give the jungle an update on how to think about AI. We’re sure all of you are aware of the run up from NVDA so this is good timing for the post.
“Mark Kern joined Blizzard Entertainment in 1997. Kern's first major role at the company was as an Associate Producer for Starcraft. He would then have roles in games such as Diablo II and Warcraft III, before becoming a manager for a World of Warcraft team. He left Blizzard in 2006.” - Wiki
Before jumping into his content, we’re beyond shocked to have him here. Starcraft is undoubtedly the greatest RTS series ever created… even though his work with World of Warcraft is likely the most famous product he was involved in.
Unsure why he took the time to help us cartoons out but thank you Mark and you’ve done more than enough value add for 1,000 life times.
With that the below is all written by man, the myth and the legend - Mark Kern
AI, AI, AI!
So, you’ve heard how AI is about to change everything, how it’s going to replace jobs, transform society and maybe even take over the world. A lot of that is hype, but behind that hype is some impressive technology is absolutely going to impact us all.
The reality is that current AI has some significant limitations, but if you aren’t using AI to help you power through day-to-day tasks, you are simply going to be less productive than those that don’t. AI is simply too useful, too fast, and too convenient to ignore.
Here are some simple things that AI can help with *today*:
Generating standard e-mail responses
Drafting rough sales copy
Generating the bulk copy of your new website (and even the HTML)
Summarizing articles when researching
Analyzing your customer data
Rapidly learning new skills or how to solve a specific problem
Creating custom stock images for your sales and marketing
Generating HTML templates for your website or emails
Using AI to code faster (websites or app development)
Each of these tasks could use their own article, but for now just understand that it only takes *seconds* to get results out of AI. But before you dive in, having some basic knowledge of how these systems work is very helpful to understanding it’s current capabilities and where it might be going. This will help you separate the hype from reality and allow you to get the most out of AI.
What is AI?
AI is an all or nothing outcome. Humanity is playing with Prometheus’ gift of fire. The razor edge, the infection point of AI, is the goal of sentience. AI doesn’t really think for itself yet, but there are a lot of smart people who are betting it can. People like Ilya Sutskever (Chief Scientist, OpenAI), Elon Musk and even John Carmack (Doom and iD Software fame) are betting they can achieve sentience in the next decade. There are an equal number of smart people (who aren’t getting as much funding), that say it will never get much smarter than a highly useful tool, driven and guided by humans.
You and I may never be able to guess if AI will really reach post human levels of consciousness and intelligence. The kicker is that neither can the computer scientists. The reason why it’s such a mystery is due to the way AI is programmed.
Traditional computer programming involves writing a sequence of instructions that a computer follows step by step to produce some kind of data transformation. That data might be user input translating to gameplay in a game as 3D characters on screen respond to your actions. Or it might be reading in a file of sales data and producing sales reports. Old school AI was also written this way, and there were things called “expert systems” that used a long list of “if then” rules to replicate things like, say, medical diagnosis. Every line of code could be read by a human, and by reading the program you could “trace” its steps and see what the computer was doing at any point in time. You could follow the flow and predict the outcomes, given the data provided.
Modern AI uses what are called “neural networks.” Neural networks are comprised of billions of very small chunks of code that only do one thing: They take a signal or collection of signals, add them up in some way, and then emit their own signal when the sum of their inputs achieves some threshold of activation. What they do is very simple. Nodes in neural networks don’t really care about the data they are carrying as signals. They don’t understand them at all. They just add them up according to a few standard ways of summing the signals, and then “fire” their own signal when there are enough combined signals to “activate.”
The important difference here is this: Nobody knows what is really going on in a neural network.
Humans can’t observe trillions of events firing at the same time, activating million of nodes of tiny code which then go on to fire another few trillion events that get passed on to the next layer. Unlike traditional computer programs, there is no way to “trace” the flow of logic, or even observe any “data” flowing through the system. There is no single flow, there are trillions of flows, and there is no readable data, just signals being collected, added and passed along.
These neural networks are doing things that we didn’t predict. We started with very small neural networks back in the 70’s, we’re talking like a dozen or so nodes. These could not do much more than, say, recognize handwritten numbers. Only recently have we been able to have neural networks at scale, and the power and growth of ChatGPT from version 1 to version 3.5, were in large part a result of improvement just by throwing more compute and more nodes and connections together and more and more data fed into the system. As scientists added more and more nodes and more and more weights, their AI’s would suddenly develop new capabilities. Like say…. being able to do multi digit addition when they were never specifically trained to do so. Or suddenly being able to translate languages it was never trained to understand.
The sudden appearance of these capabilities has caused quite a bit of debate among computer scientists. Some insist that effect is not real (or at least…not discreet jumps), but a “mirage” created by how we are measuring the abilities of these networks as they scale. Bottom line? Nobody knows, and nobody can agree.
There is some evidence that we may have reached a wall on scaling, and that adding any more nodes or weights is giving us diminishing returns on these “emergent abilities.” In fact, ChatGPT 4 is not just “bigger and better” but is rumored to be comprised of many smaller neural networks, trained to be experts on different domains (like programming), all brought together to produce a cohesive response to the user.
The reason why computer scientists can’t agree on what AI is really doing…is because they just don’t know. Not, at least, with any certainty. Everyone is literally throwing darts (albeit very smart, carefully targeted darts) and seeing if they can suddenly have a living, thinking machine.
Outcome I: True AGI (sentience)
AGI is defined as an AI that can learn to accomplish *any* task that human beings or animals can perform. Quite likely, AGI will also surpass human capabilities in the majority of useful tasks. If they succeed in creating this sentience, known as AGI (Artificial General Intelligence), then nothing I say here matters. Knowledge work would cease to exist, and so would nearly all manual labor as AI embodied in robots could carry out any task needed.
There is an old computer science joke where two scientists, who have been laboring over a new AI program to answer the big questions of life, are finally able to switch on their new creation. The machine whirs to life and speaks, “It is I, that which you have created to answer the ultimate questions. Ask me that which you would know!”
The scientists huddle and speak in hushed tones, trying to decide which should be the very first question to ask. Finally, one of them straightens up and approaches the machine. “Yes, we would ask…is there a God?”
Without hesitation, the machine answers. “There is now!’
That was a pretty cute joke back when AI was in it’s infancy, but it is pretty scary question today that keep people like Elon Musk awake at night. Elon and a good chunk of the AI community believe there is a small but non-zero risk that sentience (and sentience superior to humans) could occur rapidly and suddenly, resulting in a Skynet type scenario. They wrote an open letter to the AI community, urging for a moratorium and slowdown in AI research that has been largely ignored.
If AI were to achieve this state, it will be a sudden and tumultuous change. Society would be impacted overnight, and most of the work in the world would vanish. Our economic and social systems would be caught flat-footed and would have to rapidly transition in order to survive. There is simply no way to predict what would happen and who would be affected. (And if you really want a metaphysical existential crisis, I invite you to look up Roko’s Basilisk, but don’t say I didn’t warn you.)
Outcome 2: The Useful Partner
Let’s assume the doomsday scenario isn’t likely, and that you want to find out ways to adapt to the changes AI are bringing today, prepare yourself for tomorrow, and to stay competitive.
In the non-AGI scenario, AI research will be able to create highly intelligent AI, just shy of full human replacement, that will serve as a useful co-pilot for our day to day activities. This is the more likely outcome and is already happening. It’s time to start finding ways to integrate AI into your business and daily life, because it will greatly boost your productivity.
If you are a solopreneur, small team, or creative, current AI tools are just too good to ignore as time-saving devices and force multipliers.
I will guarantee you, if you aren’t using AI, your competitors will.
Much of the news and focus has been on a type of AI known as Generative AI. Generative AI is what programs like ChatGPT, Midjourney, Stable Diffusion (among many others) do when they write articles, code programs, and generate photographs and artwork.
The best way to learn what AI can and can’t do is to spend some time with it. And I don’t mean just dabbling with it for fun for an afternoon. You should be playing with AI a little bit each day, because getting the best results out of AI is NOT automatic. You need to be able to learn how to talk to AI to get the most out of it, and this involves trial and error and *practice*.
Unless you practice, you aren’t going to ever really understand the amazing things AI can do, and more importantly, can’t do. You won’t be able to separate the hype from reality, and you won’t be able to get a handle on the best AI opportunities right now. Your BS detector will be weak, and you could fall prey to a lot of the hype around AI right now and make bad decisions.
Don’t Start with Self Hosted AI
Maybe you think you can save some bucks and download the free version of Stable Diffusion or some Large Language Model and use the power of your own laptop or PC to start tinkering.
Don’t do this.
Unless you are technical and comfortable with dealing with github, Python and have some very fast GPUs hardware on your PC, this is a frustrating adventure that will thwart your learning speed. Pay the bucks and get signed up with OpenAI’s ChatGPT and Midjourney. Not only will you avoid the hassles with self-hosted AI script-tweaking and parameter settings, but you will also be getting access to the *very best* AI out there and be using their fastest hardware. The UI will be simple and intuitive, and you won’t be wasting any time. I don’t even recommend Google Bard, because it falls far behind what ChatGPT 4 can offer.
Pay the dozen or so bucks for ChatGPT and Midjourney and invest in your education. You can iterate faster and save time, while learning what the very best of class AI can do. Invest in yourself first, and don’t skimp.
ChatGPT: To get access to the latest versions of ChatGPT (4 and up) and remove most limits on how much compute time you can use, you will want to sign up for their premium program for $20/month here: openai.com
Midjourney: Midjourney uses Discord for their main UI (a web UI should be available soon). The cost for basic membership is $10/month and you can sign up here: midjourney.com
Get both. These are the two major areas of AI are currently: Large Language Models (ChatGPT) and image generation (Midjourney). They work differently and both are useful for different tasks. $30 bucks a month is cheap compared to the benefits you’ll reap.
You Need to Practice
The majority of people who try AI for the first time have no idea what do. They try to start a general conversation with the AI, or perhaps give it a task or two and then walk away disappointed with the results. Your first session is likely to suck, and you will be left wondering how anybody gets anything useful out of AI. This is because you don’t yet have the skills to talk to AI’s to get the best results. Now is not the time to write off AI as “overhyped bunk.”
Using AI requires knowing how to talk to AI and guide it to do the things you want it to do. Like a newly hired, but very smart intern, you must guide AI step by step through the task and correct it and teach it the type of results you want. This is generally referred to as “prompt engineering” and is vital to your success with AI.
“Smart Intern” is the best way to think about the current generation of AI. The AI knows a lot (It is basically a compressed archive of the sum of human knowledge on the Internet), but its performance level is entry to mid level. The reason AI is useful is because it’s as if you had an army of such interns at your disposal, capable of assisting with almost any task imaginable.
They way you talk to AI is the same way you would talk to an intern new to your business. You explain your business, give it information, and walk it through your task in bite sized chunks and steps. You will need to refine and correct the AI along the way, much like how you would guide a new hire. Once the AI is trained up on a task, you don’t have to re-teach it every time. ChatGPT will remember conversations within a thread and be able to maintain it.
Prompt engineering is a huge topic and there are many videos on the web that can help you with learning this process for your specific goal.
Try AI, then go watch a few videos, and come back every day to try a little bit more with AI and practice what you learned. This won’t take long, a 6-minute video, and 30 mins with AI each day, will rapidly get you proficient. It’s also a lot of fun.
The best way to approach learning AI is to have a task in mind. Got some sales copy you need to write? Try using ChatGPT to generate a rough draft, then teach it what you know to dial it in over a few revisions. Need a stock photo for your website? Use Midjourney prompts to generate it and tailor it to your needs.
If you’re like me, Midjourney will be the most fun and you’ll stay up all night just playing with creating images. There is no need to spend that much time, but you should put aside a few minutes a day to play with both ChatGPT and Midjourney. Stick with it, and you will get usable results very quickly.
Congratulations. You Passed Freshman Year.
You practiced, you have the basics of AI under your belt, and you are ready to digest future articles on what this means for various jobs, opportunities and industries and to take a deeper dive into how to use ChatGPT and Midjourney in your business or day to day tasks. You will have better tuned BS and hype detectors, and you are starting to understand the limits of current AI.
Old Books: Are available by clicking here for paid subs. Don’t support scammers selling our old stuff
Crypto: The DeFi Team built a full course on crypto that will get you up to speed (Click Here)
Security: Our official views on how to store Crypto correctly (Click Here)