AI and data: Q2 quotes by Snowflake's Slootman

Quality data is a prerequisite for building useful machine learning and AI models. By quality data I mean that it is representative, accurate and abundant. And then, even when you have quality raw data, which is a big hurdle, there is still plenty of work to do to tailor it to machine learning and then train and test models against it to obtain useful results.

From the recent Q2 conference call, Snowflake’s Slootman states some of these ideas well in response to the current AI excitement. I’m glad to see that he recognizes these important facts that all enterprises, whatever their AI goals, need to understand and execute on accordingly.

Slootman Q2 fiscal 2024 conference call:

“Generative AI is at the forefront of customer conversations. However, enterprises are also realizing that they cannot have an AI strategy without a data strategy to base it on. We have a head start in this race as the epicenter of highly curated, optimized, and trusted enterprise data.”

“…we were actually saying that having highly organized, optimized, trusted, sanctioned data is incredibly important for deploying large language models.”

If you think you can just, you know, drop a model on top of a data lake and just, you know, see what happens, that’s not going to end well. And then, that’s what people are realizing.
So, [enterprises] really got to get super serious, you know, about their foundations, you know, before we – if you don’t have a good foundation, there’s not much you can build on top of that. There’s tons of governance issues involved as well. You know, we spent, you know, literally decades, you know, as an industry, you know, making data highly governance. In other words, who can have access to what.
So, that now needs to translate into the world of large language models as well. So, there’s tons of questions that are coming up that are really important for the enablement of language models and AI generally. So, being extremely organized on your data is going to become, you know, a premium thing. And we’re obviously – that’s – you know, we’ve been on that, but it’s become more important as a function of this.”

“a lot of these want to do more in the area of AI. But first, they need to get their data into Snowflake, and it’s going to be a journey for these people. It’s not going to happen overnight, AI, for our customers.”

“And hence, the emphasis on getting your data house in order because you just cannot unleash, you know, large language model and hope for the best because of all the issues that we’ve mentioned before around governance and just understanding of what kind of data we are generating in the process.”


A good way to understand AI and Data is to understand Civilization and Tradition.

Babies born in China learn to speak Chinese. Babies born in Moslem families become Moslems. Babies born near the north pole learn to live in cold climates. Babies born in fishing villages learn to fish. It’s all about the data they are fed. Who is to curate data for AI to make it safe and useful for humanity?

Wikipedia is a good example. The consensus is that Wikipedia is mostly good. I edited an article that I though was biased but was overridden by an ‘official’ editor who, in my opinion, was on the wrong end of the political spectrum. Modern AI, machine learning AI, is no different. Now I use but don’t particularly like Wikipedia. Fact is fact, opinion is not fact. In other words, GIGO, Garbage in, Garbage out.

Denny Schlesinger


Sixty three (63) years later my hunch is proven to be correct. What follows is the rambling tale of how I figured out how AI works. It won’t help you with your investing but the conclusion might. Skip if you are in a hurry.

I’m not a scientist but a college dropout. The mathematics of physics at MIT was way too complicated for me to follow. When I dropped out my dad got me a job as a programmer at the IBM Service Bureau in Caracas in 1960. That’s another long story but the short of it is that writing computer code was just fun and games for me.

On my very first job I was introduced to the Finance Ministry as the Computer Expert even though I knew exactly nothing about it. Those were the days! I wrote the code but it was too big for the computer, an IBM 650 that ran on 2000 vacuum tubes (triodes) and a magnetic memory drum. I told my boss it didn’t fit and he replied, “It fits!” Back to the drawing board, Back to the boss. “It fits!” two or three times. The difficulty was that the IBM punched cards for this job didn’t have the standard punches but great big square ones with a different coding system…

Standard punches

I could not interpret them except by a very long algorithm that was just too big for the computer (20,000 10 digit decimal words). The problem seemed insoluble until one night when around 4AM I awoke with the perfect solution! I was usually late for work but this day I could not wait for the office to open to see if it worked in reality. It did! “It fit!”

The reason for telling this story is that it got me thinking about how the heck does the brain work. Writing algorithms is the boolean brain’s job which was fast asleep that night. It was the subconscious brain that found the solution. The computer’s instruction set had a Lookup function. I didn’t have to decode the big square holes, all I needed to know was if they were valid. A table lookup does just that. How did the subconscious brain figure that out?

How does the brain recognize a face out of millions in fractions of a second? How does it recognize a voice over a poor quality phone connection? Imagine yourself in the African savannah a few thousand years ago. Your logical algorithmic brain is way too slow to save you from hungry predators. Literally you don’t have time to think. Your other brain has to solve ‘instinctively’ for Fight or Flight? My answer to the question was that the brain is a pattern matching machine. It stores millions of patterns and it can instantly match the present one with one or more in its memory bank (don’t ask me how, beyond my grade level). By patterns I don’t just mean pictures, it can be anything including computer code lookup functions. Anything humans can experience.

My hunch has been vindicated! Tesla’s FSD12 does just that. Initial attempts at Artificial Intelligence (AI) were based on heuristics and were failures. People though that AI was just imposible

Computing proceeding to a solution by trial and error or by rules that are only loosely defined.

until they figured out how to mimic the brain’s pattern matching ability using tensor mathematics.

Tesla’s Most Important Weekend Ever :zap:

o o o o o o o o o o o o o o o o o o o o o o o o o o o o

The consequences for investing

I have a large position in TSLA which I treat as buy and forget. While I do follow its ups and downs I no longer react to them.

We are in the midst of two giant paradigm shifts, EVs and AI and Tesla is the only company that is the leader in both!

o o o o o o o o o o o o o o o o o o o o o o o o o o o o

Tensor mathematics

Denny Schlesinger