In this edition:
My thoughts
I need a favor
Learning, Tools, and Experiments
AI MISTAKES
My Thoughts
In 2024, we’re already seeing companies share their new AI wares. AI at CES was the hot topic. Wall Street earnings calls continue to talk about AI constantly in their announcement, and even before the new year AI has been mentioned hundreds of times. But a recent study published by BCG indicates that most companies aren’t diving in right away.
We are seeing adoption happen. And even if the company isn’t driving it, individuals are. More importantly, we already have examples of how enterprises use Open Source LLMs. (this hasn’t even been an option for very long!)
I’m not sure exactly how much of the audience crosses over between the YouTube channel and this newsletter, but if you are reading this and aren’t subscribed, do me a favor and check it out. I’m making it a habit of weekly streaming, and last week’s live on “The Future of AI is Open Source” seems to have touched a nerve. I do believe the future is open, and there are many reasons why I think this is the case. In the early days of the internet, so much of the Cambrian explosion was based on the fact that anyone visiting the website could easily view and learn HTML, CSS, and JavaScript. We’re seeing the same thing happen as models are shared, blended and reshared. Huggingface has established themselves as the go-to place for all things open-source models, and they’ve earned it well. They now have deals with all the big names working on AI (hardware and software).
The real value of the Generative AI revolution is that it opens up the technology to people who would otherwise not have the opportunity to use it. Therein lies my reason for starting AI MISTAKES. I want to help you learn about this so more people can use it. The faster adoption grows, the faster we improve, learn, and find new ways to apply it.
Not only is AI opening doors to opportunity, it’s also acting a catalyst to enable other emerging technologies that we have been talking about for many years. Earlier today, ARK Invest released their Big Ideas 2024 presentation, and they said it quite well with an excellent graphic to show how AI is driving convergence:
https://research.ark-invest.com/hubfs/ARK-Invest_BigIdeas_TechnologicalConvergenceMatrix.pdf
AI is enabling more rapid adoption, greater use case application, and market expansion. When we consider that this technology can ride on the commerce rails that have already been built via the internet, mobile devices, and globalization, I don’t think that saying this is a revolution in the making is an understatement.
With all of that said, these projections reveal the huge potential that lies ahead, and we must work to turn it into a critical advantage.
Let’s make more headlines like this: 12 New Jobs For The Generative AI Era. There’s plenty of work to be done; let’s get to work!
I need a favor.
I have previously shared my vision for AI MISTAKES, and I’m always looking for ways to serve. I’m in the process of building out some additional courses and offers, and if you’re looking for something specific, I’d like to connect with you more directly.
Over the next few weeks, I’ll be trying out some new offers, and I’d love to get your feedback. If you’re looking for education programs, from workshops to courses, I’d like to know more about your needs. Just reply to this newsletter and let me know what you’re thinking about or looking for!
In February, I’ll be launching a new subscription service as well, and I’m looking for willing participants to kick off the first month. If you’re interested, reply to this newsletter. You could win a FREE month (supply is limited)!
Learning Tools and Experiments
Cost Optimization
Did I mention that I am excited about the future of open-source models? Well, I am, and we’re going to see some great things happen this year.
One of those is a dramatic scale out of hybrid models. Organizations want high-quality output, but if you can pay nothing or less per query, why wouldn’t you do that.
So far, we’ve seen great success by the likes of OpenAI in making their general purpose models available. And they’ve even aggressively lowered the costs of using the calls. Last week, OpenAI introduced new options and lower pricing yet again.
So here is a solution that can help deliver improved costs, but helping make decisions about query orchestration.
https://github.com/leeroo-ai/leeroo_orchestrator
Architecture
Last year I shared a wonderful article provided by A16z (Mark Andreeson’s VC firm) and they’ve continued their work and have more to share. They’ve now launched this repo to help provide some great starting points for the key elements in architecture. If you’re building AI, this is worth your time or at least a bookmark.
https://github.com/a16z-infra/llm-app-stack?tab=readme-ov-file
Creation
Not to be outdone, there’s been a great deal of activity in solutions that everyone can use to save time and get their work done.
I use AI tools every single day. My favorite tools are:
1.) perplexity.ai
2.) ChatGPT GPT’s
3.) Microsoft Co-pilot (including windows 11 side panel)
4.) Pinokio (your own personal AI experiments lab)
I’m sharing weekly videos on my YouTube channel, so don’t miss out on this hour of fun and creativity every week as I go in-depth on what I am excited about and how I am using these tools to get more done in less time while also having fun!
AI MISTAKES
Magazine Publisher Arena Group’s Use of AI Content
“Ortiz isn't the only AI-generated author published by Sports Illustrated, according to a person involved with the creation of the content who asked to be kept anonymous to protect them from professional repercussions.”
None of the articles credited to Ortiz or the other names contained any disclosure about the use of AI or that the writer wasn't real, though they did eventually gain a disclaimer explaining that the content was "created by a 3rd party," and that the "Sports Illustrated editorial staff are not involved in the creation of this content."
AI is constantly improving. We probably already read more AI-written content than we realize. Watch for this mistake to be repeated in the future. This is a mistake on several levels. Firstly, almost all publications want to enforce copyright, but judgments have already been made that require attachment to a human for intellectual property. Secondly, they were pushing the boundaries of content creation, got caught, and then deleted the stories. Thirdly, the AI-generated content was never edited. The article references a story that listed “five simple…steps” but numbered them each “1.” a did that five times.
AI-generated content is here to stay, but don’t let it do the job for you. We’ve been down this road before. The MISTAKE that people make with automation is assuming that it’s perfect all the time. Humans aren’t perfect, and while we’d like to think that machines are, too, they aren’t. All processes need control, attention, and feedback. Nascent systems need this far more than mature systems. AI is great fun, but it’s more toward the nascent side.
Regardless of who was responsible for the quality of the content, Generative AI needs systems, processes, and procedures to eliminate errors and ensure consistency. Don’t make this AI MISTAKE.
Popular AI Chatbots Found to Give Error-Ridden Legal Answers
Large language models hallucinate at least 75% of the time when answering questions about a court’s core ruling, the researchers found. They tested more than 200,000 legal questions on OpenAI’s ChatGPT 3.5, Google’s PaLM 2, and Meta’s Llama 2—all general-purpose models not built for specific legal use.
I’m a big fan of the general-purpose models. We’ve seen multiple scenarios where GPT-4 (a general-purpose model by OpenAI) has better purpose-built models. The amount of knowledge and accurate responses that these systems can give is incredible! If you are building a new system and you lack data to build your own, setting up a trial of a general-purpose system is the best choice!
However, as we learn from this AI MISTAKE, not all scenarios lend themselves to general-purpose models. Many scenarios still benefit from independent models, augmented training, and proprietary information that is not available anywhere else.
2024 will be the year of open-source models, especially those coupled with other systems. The mixture-of-models architecture as demonstrated by Mistral MIXTRAL is very likely to gain significant traction and value in the near term.
Based on what we saw in 2023, using LLMs to identify case law and precedence is dangerous. We’ve seen this before. All seemed well when the use of Autopilot in airplanes was mandated over having a pilot-in-command to fly the plane manually. The problem, though, was when something out of the ordinary happened. Protocols required humans to take over control, but because of the mandated autopilot use, the pilot in command suffered from significant skill atrophy and either was unable to control the aircraft properly or lacked the decision-making practice to make the right call.
Nicholas Carr said it best in The Glass Cage: Automation and Us: How Our Computers Are Changing Us.
WHEN PEOPLE tackle a task with the aid of computers, they often fall victim to a pair of cognitive ailments, automation complacency and automation bias. Both reveal the traps that lie in store when we take the Whitehead route of performing important operations without thinking about them.
Whether it’s sophisticated aircraft or Artificial Intelligence tools, we tend to become complacent. Don’t make this AI MISTAKE. Build systems that ensure the right steps are taken when quality is paramount.
And in the meantime, use this everywhere with little or no risk.
Do some writing. Create some art. Learn about a new process or technique. There are hundreds of use cases for AI that you can use today at almost no risk.
And if you are having trouble finding one, please let me know. I can help.
Speaking of letting me know, if you read something here and have questions, reply to this newsletter. I read all your comments, and I love engaging with my readers.