Ahh! Screaming noises! Fear! Running around in circles waving my hands above my head! Ahhhhh!
In this issue… the world ends! Google is dead!
- Jumping the whale (48 words)
- Graduating medical school with the highest GPA (685 words)
- Emergent abilities in LLMs (78 words)
Jumping the Whale
Let’s play Jeopardy!
Answer: A phrase describing a social paradigm wherein journalists and analysts lead people to believe that a corporate entity worth trillions is “losing” the battle for technological supremacy simply because a competitor invested in a popular new technology vertical.
Question: What is “jumping the whale?”
Graduating med school with the highest GPA
There’s a specific word for the person who ends up finishing med school with the highest GPA of all graduating students. That word is: doctor. It’s the same word they use for the person who graduates med school with the lowest GPA.
I point this out because my news and social media feeds are inundated with articles about how Microsoft is destroying Google in the modern AI arms race.
I’m gobsmacked by the amount of tech news coverage today dedicated to baselessly declaring Microsoft the arbitrary winner of an AI arms race that only exists in the tech section of popular news sites.
Editors HMU if you want a sober counterpoint.
— Tristan Greene 🏳🌈 (@mrgreene1977) January 26, 2023
These aren’t articles about rising tides or long-term prospects, they’re mostly quoting hyperbolic analysts who think that MS’s recent investment in OpenAI (reportedly $10 billion) has suddenly propelled it to the pinnacle of tech supremacy.
And, to the one, they all fixate on the mysterious suggestion that Google must now radically alter the way it does business in order to catch up to Microsoft.
A pair of FT reporters recently concluded their preemptive Google obituary piece with:
“But as the most dramatic demonstration yet of an AI force that is sweeping through the tech world, OpenAI has given notice that even entrenched powers like Google could be at risk.”
A reporter at Vox ends theirs with:
“Google now finds itself in a similar position that it helped put Microsoft in two decades ago: hoping it can release something better before it gets passed by.”
Allow me to retort…
…hold on, let me crack my knuckles and sit up straight so I can type this properly…
You’re all confusing chatbots with artificial general intelligence!
Chatbots aren’t anymore positioned to replace Google Search than podcasts are to replace colleges. I said as much when I wrote this op-ed for MIT’s Undark Magazine.
Until someone invents an all-knowing, god-like oracle, narrow intelligence will trump generalist knowledge in narrow tasks.
You want to do searches? You need search algorithms. You want to parse medical imagery? You need to train a vision model on a database containing only medical imagery. You want to automate the security and publishing process for internal application development? You need to train an AI to do… the thing you actually want it to do, specifically.
My point is that there is no “generalist” AI technology capable of “winning” the battle for corporate AI supremacy. Here’s a short analogy explaining why:
You and I go to the library to study for our chemistry exam. You say “I’m going to read every single chemistry book in the entire library as quickly as I possibly can.” I retort with “I’m going to read the assigned chapters in our chemistry textbook, make some flashcards, and spend the rest of the day studying them.
On the surface, you sound way smarter than me. With my narrow little education, it seems like I should perform much worse than you on the test.
But that’s not how people or AI function. There are no truly general intelligences in the world because intelligence is subjective. A bird might not be “smart” enough to build a fighter jet, but the biological advantages it has over humans and fighter jets means it’s significantly less likely to experience an unexpected outcome (crashing, navigation failure) unrelated to adversarial attack than we are.
As stated above, until someone invents an all-knowing, God-level oracle, narrow intelligence will trump generalist knowledge in narrow tasks.
What this all means is: Generative AI is a class of artificial intelligence technology. Saying Google’s in trouble because MS invested in OpenAI is like saying that the graduating doctor with the highest GPA at Harvard med school is going to put the CEO of Boeing out of a job.
Just because it’s within the realm of possibility doesn’t mean it makes any sense.
Trouble with you
Is the trouble with me
But we still don’t see
Come round the bend
You know it’s the end
The fireman screams
And the engine just gleams
Emergent abilities in LLMs
Do I believe that “scale is all you need?”
Sure, if “all you need” is the first half of a two-part sentence. As in, “scale is all you need to continue brute-forcing emergent abilities out of LLMs, but there’s a physics-based limit to the amount of data that can be pipelined before the pipe is bigger than the overall data space.”
I think the more interesting question is: when do we start scaling models on nonbinary architecture?
Read more More Than Meets the AI here!
