What is a medium language model? (you can abbrev as mlm for fun)
It's mostly my term for the moment, but it appears to me as the obvious term for a natural line of research I have heard discussed more and more. In this video, I discuss some trends in #ArtificialIntelligence research oriented towards producing #LLM -like models that are less prone to hallucinations and are hopefully more suitable to serve as experts. The idea is that if the training data is restricted to material produced by appropriate experts, the output will be more consistently expert, but then the immediate question is if there is enough data to suit this pickiness.
The biggest question in AI: What is AI?
If you are a business leader or executive, it might turn out to be very important how you communicate with various stakeholders about #ArtificialIntelligence. There is one simple, surprising question you can contemplate that will help you handle this communication well...
What is #AI, exactly?
In this video, I argue that it is everything not to understand this question for yourself, but rather to appreciate + accommodate all the ways other people understand it.
No seats on the OpenAI board for Apple or Microsoft
In this video I discuss recent news that Apple and Microsoft will no longer pursue observer seats on OpenAI's board ostensibly because of concerns around antitrust optics. Success with #ArtificialIntelligence in a large organization often involves navigating unusually many and unusually weird stakeholder revolts and this story is a great example of how non-technical issues like corporate governance and antitrust regulation can join forces to make your life more interesting than you might think.
AI is out and biznis is in at Open AI?
Open AI is in the news today for hiring a CFO and a Chief Product Officer. I think it is more broadly noteworthy that human resource stories about Open AI these days invariably involve not only the hiring of many people on the business side but frequently also the departures of notable engineers and scientists like Ilya Sutskever. I examine some recent history including Open AI's journey from not-for-profit research lab to legally awkward pseudo-for-profit company, its near implosion following the attempted and eventually failed ouster of CEO Sam Altman, and use these data points to muse on whether technology is out and business is in to such a degree as to produce impactful consequences down the road.
Scarlet Johannson is suing OpenAI! She should win but might lose...
Take a study break to let me update you on events at the intersection of #ArtificialIntelligence and personal identity. The biggest headline is that Scarlet Johansson is suing #ChatGPT maker OpenAI for allegedly using her voice without her consent. The punchline is that I think she likely deserves to win, but she may not. I provide some discussion my personal journey as a creator, as well as some observations of trends on Pinterest and jokes about the fantastic abs typical of male underwear models, to argue that the current situation is very murky in a way that I expect will enable bad actors to misbehave with a little too much impunity.
The (non) future of new and science fictional consumer products
#BigTech companies like #Apple or #Google are under unusual pressure to show the world a futuristic, transformative product and there are plenty of incumbent ideas that we tend to agree are cool and futuristic when we see them on a lab bench. However, I think there is a lot of room to be skeptical that people will welcome these ideas into their day-to-day lives in the way that is required to produce a hit product, much less a socially transformative product like the smartphone. In this video, I talk about correct and incorrect prophecies in classic Schwarzenegger action / sci-fi films to point out that we do a better job guessing at the progress of technology than at the evolution of our day-to-day interest. I spotlight some particular technologies that are knocking on the door, including #AugmentedReality as well as a possible #ArtificialIntelligence -fueled shift in how we interact with computing broadly that you can see implicit in #OpenAI's recent focus on improving capabilities for audio and image based interaction.
LLM, Agent Based Workflows, and Dungeons and Dragons
Below I talk about agent based workflows, a paradigm for how to use #LargeLanguageModels (#LLM) and similar tools, in the context of a tabletop role-playing game like Dungeons and Dragons. I have a very basic text-based game (linked below) built on the #ChatGPT API that has some limitations I am going to improve over the next few weeks. I am going to do my best at doing this in an at least philosophically agent-based manner - look out for more videos on my progress and why you might call it agent-based. The goal is to give you a sense of the value in these concepts that is more tangible and problem-solving oriented than some of what you might find out there right now.
Failed PoC or Foundation for CoE?
A sort of golden parachute: If corporate risk management concerns prevent your #GenerativeAI project from making it into production then you might think about using your momentum as the foundation for a forward-looking #CenterOfExcellence. #ArtificialIntelligence will have its day, even it is not today, and you want tested expertise in the room when it is ready.
AI is for small business
The current wave of #GenerativeAI technology may get a lot more uptake with #SmallBusiness and smaller businesses will likely be the leaders that eventually drag the large corporations with them. This is not something that will be well-publicized, but it is already out there if you know how to look for it. Invest 10 minutes in my story on how the AI wave is really about small business and why.
#ArtificialIntelligence
The human in the loop
You can't leave #ArtificialIntelligence all the way unsupervised.
Can you use #GenerativeAI to plan an event for you? #Meetup has a button you can click to do this. The answer is "Yes" but in a silly way. Do you care what the event is? What kind of work you have to do to back up the plan? What the event is about? You probably do, and if you do you need to put about as much work into the supervising the AI as you might have spent planning.
If you are involved in managing risk for a large organization, or if you are worried about where the labor market is headed, you should note that you need humans in the loop for the foreseeable future. No matter how good the AI is, you still need to check that it is behaving in a way aligned with your goals and resources. You can have a brilliant employee that creates a problem because you did not bother to communicate with them and AI is not really any different.
Is prompt engineering for real?
Is #PromptEngineering for real? Is it really a coherent, specialized skill that will stay relevant or is it just a buzzword about tinkering with another buzzword? There is a multi-tiered debate going on about this right now and I won't be able to cover all of it here. I will say, though, that there is a lot more to it than you will be exposed to playing with #ChatGPT for personal use and I often see the topic carelessly over-trivialized on social media. If you are building an application using a #LargeLanguageModel or similar, you are really writing prompts for other people that might not be cooperative, you might have prompts helping write those prompts, and the stakes for your business could be pretty high. It is a challenge to take seriously.
#ArtificialIntelligence #GenerativeAI #SoftwareEngineering #ApplicationDevelopment
Elon Musk sues OpenAI!
#ElonMusk is suing #OpenAI. This is hot off the presses right now and potentially a really big deal.
In this video I provide context and backstory for this lawsuit including...
Musk's turn as a co-founder, funder and one-time board member of OpenAI,
how OpenAI was in fact founded to be open yet has become closed,
how OpenAI is now oddly a not-for profit with a for-profit subsidiary,
the role of Microsoft and how all these threads fit into OpenAI's recent semi-implosion followed by #SamAltman's return to the helm.
#ArtificialIntelligence
What's in an LLM development library like LangChain? And why?
LangChain is a framework for developing applications powered by #LargeLanguageModel s (#LLM s). Just what does this mean? What is in this library and why? In this video, I start by contrasting #ChatGPT the popular web app with ChatGPT the developer API in order to point out some very basic tasks and problems that many people will quickly encounter when creating applications like chatbots. This is a great lens through which to start explaining not only what functionality is in LangChain but why it is in there in the form that it is and how you might decide to use (or not to use) such a tool.
#ArtificialIntelligence
Ironically, sometimes reliability is everything in AI
Many potential applications of #ArtificialIntelligence lie at the center of a tough Catch-22: a more sophisticated #AI has more exotic ways to be unreliable, yet if the point is automate and replace human labor then reliability is a key requirement.
#ProductDesign
Why the data breach at Wyze is too classic not to happen again
We used to say "data is the new oil" and I've always enjoyed ironically reinterpreting this maxim from a #DataPrivacy standpoint: data is something that inevitably leaks out of its pipes and when it does it is toxic waste.
It is through this lens that I tell the story of the recent data breach at Wyze. I give the basic facts and explain why this incident is very classic. Not only have things gone wrong in similar ways at other organizations in the past, one can also articulate structural challenges that guarantee things will go wrong in similar ways elsewhere in the future.
#CyberSecurity #SoftwareEngineering
Intelligent AI shopping is not to be taken for granted...
"We'll just buy AI." I hear this statement often and it is eminently reasonable except maybe for the "just" part. There is a flood of AI tools coming onto the market and I don't think intelligent shopping is something to take for granted. To an unusual degree, you will see shadows of the difficulty of building even if you only interested in buying and this situation is worth understanding. In this video, I discuss some unusually treacherous pitfalls you might find navigating this marketplace before exploring them tangibly in the seedy but illuminating AI girlfriend niche.
The tower of communication challenges in AI
Today's thesis because you need this is that getting #ArtificialIntelligence right in a large organization is primarily a #communication challenge. This is especially true to the extent that your company is not primarily an #AI company and I would say it is relevant whether you expect to be building things in-house or whether you expect to be buying them. There is a tall tower of stakeholders, from individual technical contributors with backgrounds like #DataScience or #SoftwareEngineering through corporate #management and potentially all the way up to #WallStreet, that will each have slightly different visions of "AI" that won't necessarily mesh without care and empathy. If you can align this tower, the sky is the limit. If you can't it will appear to everyone that everyone else is talking about nothing... and the microeconomics textbook has scary things to say about where this leads.
Facial recognition nightmares
If #ArtificialIntelligence turns your life into a dystopian hellscape anytime soon, I expect #FacialRecognition technology will be at the tip of the spear...
In December the Federal Trade Commission banned RITE AID from using facial recognition technology for a period of five years after it was revealed their implementation suffered systematic #AlgorithmicBias, often flagging women and people of color as shoplifters without good cause. This is essentially a story about a robot being a bigot across all Rite Aid stores on the company's behalf.
One might hope technological improvement will fix these issues, but I am here to tell you it is out of the frying pan and into the fire. Towards this end, I review some past incidents involving Madison Square Garden Entertainment Corp. where facial recognition was used not to flag criminals, but employees of law firms involved in litigation against the venue. Facial recognition is something that helps people exercise power, and we are now taking for granted that our would be bullies sometimes struggle to find us in pursuit of our lunch money.
These are nightmarish experiences for everyday people who have not done anything wrong.
This is a high-conviction take and I would like you to take notice. I roll out some of my best agitated hand-talk and idiosyncratic interpretation of Chinese proverbs in hopes of persuading you. Please take notice.
#DataPrivacy
Act on AI, don't react to AI
It's tough to know when to act and when to react, or sometimes even to know which one you are doing at a given moment. I draw on some standard negotiation concepts and folksy old-timer sayings in hopes of convincing you neither to uncritically embrace the current wave of #ArtificialIntelligence hype nor ignore the technology as a passing fad. Let me persuade you of a middle way: #AI is something you need to make part of your strategy for the next several years in a careful, thoughtful way that considers all the small details of how your business works.
#management #hype #strategy
Advice on sizing your project via an AI tabletop RPG toy
Many of us are thinking about how to size our #GenerativeAI projects and I would like to convince you to take care - this one is more subtle than some others. In this installment, I have both some significant figure metaphors from high school chemistry as well as an #AI Dungeons and Dragons toy we built using tools from #OpenAI and #Pinecone in a #RetrievalAugmentedGeneration (#RAG) architecture. The story is that a maximally quick and dirty approach is amazingly cheap, and then you can make superficially the same project infinitely expensive as you add more demands on quality control, #DataPrivacy, and #infrastructure.