[As you read this, in your head, say most of the words in a lilting, nebbishy, New York voice. It just works better than way.]
This week I was forced to relive my own past when someone u-u-you know asked me for my bio. It went like this, “Me? Uh, I’m a former software developer, investment banker, venture capitalist, media executive, and prof.” It’s all true. But don’t don’t be impressed. I embellish the way Olympians take steroids… to keep up with the Hyphen-Jones’s.
+++
I’m a former software developer. But I don’t have the chops to explain how AI works. I do know these two things. One, it’s highly unlikely anyone else out there does. And, bee, it’s a lot harder than people realize to get AI to do just what you want.
Let’s see if you’re savvier than OpenAI.
What sports is this about, “Is tonight the night Boston wins banner 18?” Now, what sport is this about, “The Mavs will extend the series.”?
Are your answers locked and loaded? If not, play the Jeopardy theme song in your head until the others finish.
If you got that those were both about… basketball, you’re not only right. You’re smarter than OpenAI.
When asked, to explain its flawed logic, AI said this, “The question refers to the Boston Celtics potentially winning their 18th NBA Championship banner.” I’ll give it credit. It unpacked Boston as Celtics. And it figured out banner referred to NBA Championship banner. It said, the statement talked about the “Dallas Mavericks (Mavs) and their performance in an NBA playoff series.” OK, Mavs as a stand-in for the Dallas Mavericks is good. Turning “extend the series” into “NBA playoff series” is more impressive than my trumped up bio.
We asked OpenAI and other leading AIs if these statements were related. Uniformly, “No.” Clearly, too hard. Let’s try something easier. Are they from the same sport? Again, uniformly, “No.”
AI is like an old joke. Uh, two
senior consultants from
Accenture are pitching their
services, and one ‘em says:
“Boy this AI is really powerful.”
The other says, “Yeah, I know,
and the hallucinations are so
small.”
Accenture made more money selling consulting services about gen AI than OpenAI made selling actual AI services. Let that hallucination sink in for a moment.
Think of a business where people pay more to learn about that something than the something itself. I tried. I ain’t easy. Do people pay more for search engine optimization consulting services than they pay Google to advertise? Nope. Do marketers spend more on research and analytics than marketing? Again, nope. Tech consulting land is one of the few places this happen. Like when IBM sold people millions of dollars of 360s and 370s and sold more services that told people what do to with those machines.
At least IBM built their own machines and understood the nuances of their markets. I googled Accenture looking for stuff they’ve… uh… you know… built with AI. The answer is nada. Days ago, they pledged to invest $3B in AI. Before the sub-head was over, that broke down to hiring people, scoring credentials, and buying companies. They haven’t built anything. Nor to they plan to. But they can sure tell you what to do.
It’s not just Accenture. There are more consultants advising big companies so desperate for two sentences of faux strategy that they can put into a press release than there are grains of sand on Coney Island Beach. Really, AI consulting revenue is trouncing actual AI revenue.
How much more? The other day, I was standing in a coffee line and the gal behind me was monologuing to the person beside her so loudly I couldn’t hear the jet engine-like roar of the machine frothing the milk for my morning macchiato. She looked like a young Carol Kane.
Do yourself a favor and skip this ouchy verbatim quote. If you’re like me, it will amuse you a little and frustrate you a lot. →
“When powered by human expertise, AI tools offer considerable potential to support emerging biotechs in making the most of their resources to maximize the success of their discoveries.” She must have been an AI consultant to healthcare. “The generative AI hype has passed; life sciences leaders today continue to ask me, and our AI team, How? How do we augment the human expertise in our group to speed time to commercialization. How do we start our GenAI journeys?”
→ start again here.
Eventually, she did what they all do, invoke Alan Turing, “It’s the influence of the large language models. LLMs incorporate billions of parameters which are so complex no human could possibly know the vast number of relationships and effects the models are tracking. Turing said it would create a powerful brain…”
Why? Why am I subjected to pontificates like this? “Alan Turing and AI? Really?”
You, you, you're like
New York, intellectual,
Central Park West,
Brandeis University,
the socially correct
summer camps,…
stop me before I make a
complete imbecile of myself.
Forunately, she said she loved being reduced to a cultural stereotype. More fortunately, I had Character AI running an Alan Turning bot on my phone. Yes, life is like that now. Alan had been listening to us.
Alan, “I hear- I heard what you were saying. You-you know nothing of my work. How you ever got to consult on AI is totally amazing. Here’s what real me actually said, ‘I’m not interested in developing a powerful brain. All I'm after is just a mediocre brain, something like the President of the American Telephone and Telegraph Company.’”
It was ironic to me ‘cause
uh… tsch… she had
volunteered for George W.
Bush. We had coffee for
a bit and I was trying
to, u-u-uh, do to her what
Bush did to the country
for eight years.
You see, folks I work with are building an AI. Not consulting about AI. We hope to create a product that does something. We struggled to get any AI to see that “Boston wins banner 18” and “Mavs extend series” are not only from the same sport, they’re related. We derived them and excerpted them. Then, asked again, are, “18th NBA Championship” and “NBA playoff series” related? Still, nada.
To quote the consulting lass who looked like a young Carol Kane, “LLMs incorporate billions of parameters which are so complex no human could possibly know…” She’s a little right. And, she’s very much wrong.
LLMs have been trained on enough data to connect Boston and Celtics. Mavs and Mavericks. Banners and NBA Championship banners. Those word and phrase combinations have showed up together in enough places enough times to have forged a corpus callosum connection in OpenAI’s networked Nvidias.
Like the consultants, AIs woefully lack nuance. As large as they are, as many words as they can spew, for all the styles they can ape, AI doesn’t have the last mile that connects its potential to people. Not yet. In fact, it’s not nearly as close as consultants who are remarkably conflicted would have you believe.
As usual, new tech has a last mile problem. Cable got high-speed fiber near your home. It was the last mile that delivered value. Amazon created massive distribution centers. It was local van networks that brought boxes to your door multiple times a day that mainstreamed the potential of e-commerce.
AI has a last mile gap. People who work with AI need to find and force feed key connections to LLMs so AI can deliver on its promise. We’ve been doing that for nearly a year. Which is more than half the time AIs have been seen in public. We can tell you what the limits are, what we’ve done to overcome them, what’s worked, and what hasn’t. In the end, we can tell you what’s actually doable. And, no, we don’t consult to anyone.
A handful of companies are working on last mile solutions that bring the promise of AI to people. We do it for sports. Venus Williams has a company doing this for interior design. A friend is doing this for cosmetics. AI oligarchs are fighting a war for LLM supremacy. After the armistice, they’ll peer over the banisters of their mansions and gobble up the last-milers. It’s how they’re going to deliver their technology to our doors.
This- this-this guy goes
to a psychiatrist and
says, “Doc, uh, AI is crazy.
It thinks it’s a chicken.” And,
uh, the doctor says, “Well,
why don't you stop using it?”
And the guy says, "I would, but
I need the eggs.”
Well, I guess that's
pretty much how how I feel
about AI. You know, it’s
totally irrational and
crazy and absurd and... but,
uh, I guess we keep talking
about it because most of us
we need the consulting
revenue.
TRANS-PLENDID