In this video, I share a sense why 1) #ArtificialIntelligence #Regulation is necessary, and 2) why this area of policy is very challenging and thus likely to remain unsettled for some time.
Weaving all the OpenAI stories together...
There are several notable OpenAI stories out today competing with each other for oxygen and I think it’s worth examining them all together. In this video, I recap some important history and weave a number of current events into a broader narrative including...
-the departures of CTO Mira Murati and Chief Research Officer Bob McGrew -OpenAI's efforts to reform itself as a for-profit corporation
-the adventures of OpenAI alums like Ilya Sutskever at other startups
OpenAI always looks less like a research organization and more like a company commercializing existing technology. If there is a red flag in the mix here it is that OpenAI may not be able to make the breakthroughs required to sustain the image that (at least in part) fuels its massive fundraising. #ArtificialIntelligence
AI is for people of influence, adventurers, and the far-sighted
This motivational speech is only for people of influence, adventurers, and those of us compelled by nature to think long-term...
#ArtificialIntelligence
It's tough sledding doing AI research in a large organization
#ArtificialIntelligence is both very mature and very immature from a megacorporate perspective.
You may have seen discussion elsewhere of The Wall Street Journal's coverage regarding Amazon's internal #AI team and its struggles. Google, despite its once dominant position, has often seemed a step behind OpenAI and OpenAI in turn may have done its best work when it was a small not-for-profit research lab...
It's difficult to do these things well in a large, private-sector organization and in this video I explore a variety of reasons why including fundamental cultural requirements around anything resembling basic research as well as the burden coming from megacorporate guardrails in areas like #DataPrivacy.
Future pitfalls buying AI
For many companies the #ArtificialIntelligence revolution will be purchased from a vendor.
In this video, I sketch a (hopefully familiar) prototype for a flawed software evaluation process and indicate why I think #AI could make these kinds of dynamics much worse. The story is approximately is that the more sophisticated and flexible these applications become the more evaluating them will have the subtlety of hiring an employee. It will not be a process that is easily compartmentalized and quantified into a spreadsheet.
Three things to watch to know the fate of OpenAI
It's plausible that as goes OpenAI so goes hashtag#ArtificialIntelligence as well. In this video I discuss three things to watch if you hope to unravel the arcane strands of the future...
1) Does OpenAI continue to lead in qualitatively new and different R&D?
2) Do corporations find prosaic applications of hashtag#LLM or is it just a toy?
3) Do we return to low-interest rate, high liquidity macroeconomic conditions as prevailed before the pandemic? Or have we durably entered a new era?
The Fed's 50 bps cut and the many reasons you should care
You may have heard the #FederalReserve cut interest rates by 50 bps yesterday and this video attempts to answer the question "What does that mean and why do I care?" I go into detail about...
- what the Fed is trying achieve and how its going past and (potential) future
- just how it manipulates interest rates and why this matters on Main Street
- the consequences of the two ways of screwing it up
- what happens on #WallStreet and in your stock portfolio
but then also more unusually
- why #COVID19 is still lurking in the background
- how this sets the stage for the arc of companies like 23andMe and WeWork
To bring it all back to my core wheelhouse, I outline why the yet-to-be-discovered aftermath of this rate cut matters a lot for the fortunes of companies like OpenAI and thus for #ArtificialIntelligence more broadly.
Backstory and relevance of the mass 23andMe board resignations
In this video I discuss not only the recent mass resignations from 23andMe's board but also the longer history of the company and the circumstances, in the company and in broader society, that set the stage for these events. While this is not an #ArtificialIntelligence or #DataPrivacy story per se, it's a great case study in a variety of business themes that are very relevant including - how a macroeconomic shift has altered investor expectations - the legacy of "data is the new oil" - corporate governance at young, entrepreneurial companies I close with the observation that we are really just learning about the downhill stretch of a particular sort of rollercoaster for the first time...
AI is much better at cheating on tests than humans
Are #LargeLanguageModel s, in a sense, "cheating" on the test?
In this video, I share some observations regarding formal and informal metrics for #LLM s, recap (implicitly) some bias-variance theory from machine learning, and reminisce on my days teaching calculus in order to argue there is a danger what looks like intelligence might in some cases might really be more like memorization. The latter is not a terribly cool trick in a machine. If you are in the tricky, philosophically perilous business of comparing humans and #AI, then for many of these tests you should probably do some handicapping in favor of the humans.
#ArtificialIntelligence
Dream with discipline about AI
It's important to dream with discipline about #ArtificialIntelligence.
You will hear me say again and again that #AI v. business is a long-term conversation. Adoption is an urgent matter in some industries while for others more refinement of the technology will be required. What any business can and should do right now is invest in clear, explicit planning to clarify both...
1) ("dream") What is required of a game-changing application? What is the universe of plausible game-changing applications?
2) ("with discipline") Is this available now? If not, what are the warts on current technology that prevent it? How will I know when it is almost ready?
There will come a moment when a starting gun goes off and you want to be finishing a challenging conversation about AI and not just starting one.
The deets on OpenAI's "Strawberry" model
In this video I digest the news around Strawberry, OpenAI's recently released and much anticipated new model with "reasoning" capabilities. Topics discussed include...
-the sorts of problems Strawberry is intended to solve better
-the mostly limited and vague known details of new training techniques
-similarity of form and function with past LLMs, including "hallucinations"
-what OpenAI might be trying to accomplish from a business standpoint
AI is in its e-commerce 1998 moment
It's 1998. I'm telling you that e-commerce is the future and I have some hot stocks to sell you. Are you in? My stated investing thesis is vaguely correct, but depending on the particular stocks you could retire way early (Amazon.com) or lose your shirt in 1999 (Pets.com).
To uncritically accept my thesis is to roll the dice carelessly yet to react against my thesis is to miss opportunity. This is the sort of crossroads businesses face regarding #ArtificialIntelligence. There is over- and mis- hype out there, but this is not to say that accepting the frame of the hype and reacting against is the right path.
In this video, I examine these threads in more detail and argue the right path is to ignore the hype and figure out the hype really should be for your business and your industry.
AI as a (powerful, transformative) management technology
#ArtificialIntelligence might eventually transform how large organizations are administered, and I think this sort of application is both more interesting and more important than commonly recognized. In this video, I contrast your typical ordered, pyramidal org chart with the more disorderly and decentralized ("rhizomatic") reality of how things get done - we have all had experiences where we got to go on a journey seeking the person who knows where the bodies are buried. The right sort of #AI observer + interlocutor could potentially do everything to enhance management's ability to steer the ship in a constructive and straightforward way.
AI is doctoring your photos! (and has been a while...)
You've been using #ArtificialIntelligence to doctor all your photographs for years just as most any image you see on social media has been doctored!
Does this revelation surprise you? I'm being hyperbolic, but it has been true for some time that most smartphones had sophisticated #MachineLearning algorithms cleaning up photos on the fly. Does this count as #AI doctoring? Different people may say different things...
I think this is worth pointing out both because...
1) there is already more AI in the room with us than we admit, and related...
2) human-and-AI collaboration is already the norm for many tasks
In this video I develop these themes and provide some more examples you may have already encountered without knowing.
Doing generative AI well requires pulling up your socks (metaphorically)
Are your organization's #ArtificialIntelligence socks pulled up all the way? Or are droopy socks giving everyone blisters? There is a story I love about celebrated UCLA basketball coach John Wooden coaching players on putting their shoes and socks on with care. Many people will tell you elite performance is about doing all the less-than-sexy things right. In this video, I distill the content of some failure & frustration stories I have heard to argue that doing #GenerativeAI well is as much about an application development headspace as it is about AI. Failure scenarios have an AI story in them, but it is always about AI interacting with all the little, prosaic details of building an app that real people actually want to use.
AI winter and why the NVIDIA earnings call is important
As you await the NVIDIA earnings call tomorrow you might take a couple minutes to think about a.) why finance is a very important part of the long-term story of #ArtificialIntelligence, b.) the past periods "AI winter" in the 60s / 80s, the importance of expectations, and c.) why it is generally enlightening to understand why this earnings call is so anticipated.
Why the parade of new LLMs could represent a headache for you
There is always an exciting new hashtag #LLM model coming out these days. If you are trying to build something practical for right now, this could actually be a headache for you. In this video, I present arguments related to cost, model lifecycle, and safety to the effect that one needs to beware the parade of new models. I also provide one more voice in favor of the open-source, locally hosted model as I perceive it to be the least vulnerable to the issues I discuss.
Unpacking a hyperbolic AI headline
The headline: "Walmart Achieves 100x Productivity Boost with Generative AI!"
It's worth breaking down...
What does this really mean?
What does Wall Street really want it to mean?
What do you need to do to bring these closer together?
#ArtificialIntelligence
Digging into the "why?" of the Apple Intelligence delay
Apple has delayed its anticipated release of a new raft of #ArtificialIntelligence features for its phones. In this video, I dive deep on this decision primarily for the benefit of business leaders contemplating similar decisions. Tidbits discussed in the video that are worth mentioning in writing include...
"stability" as a motivation and on-device vs. #cloud processing
the use of specialized chips produced by Google
#WallStreet's scrutiny of #BigTech #AI spending
The emerging "medium language model"
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.