Are Large Language Models Really AI?

I didn't, it's just that the term "will be" has no value in my eyes without a time in the same statement.

I can tell you SP 500 "will be" 7000 one day, or it "will" suffer from 20% correction one day. Without knowing when none of this prediction has any real value. Will it happen tomorrow? 3 years later? 10 years later? Who knows. And if I don't know when, then this info doesn't matter. Because it can't be used to generate real profit.

I am also one of these people who believe market can ever be predicted with 100% accuracy with algorithms and past data ever. There are entirely way too many random variables, human emotions change, timing and luck on detecting demand and winning in the market. I believe algorithms and data could increase the win rate in the market, or speed up the product deployment process. but winning often enough with data from the past? Very unlikely IMO.

(It's also the same reason why economists run models for years and still unable to predict economy with 100% accuracy. Too many random variables and there is a limitation on data driven science when it comes to economy and market related stuffs.)

So LLM's aren't really "giving you answers", that concept doesn't even exist. A chat parser is taking in put verb statements, breaking them into pieces, then looking through a massive multi-dimensional array for the sequence of words that most likely matches those pieces you inputed. There is no thinking or analyzing involved, that already happened during the training stage when a truly ridiculous amount of compute was spent to create that multi-dimensional array. Those arrays were built on every bit of text those AI companies could scrape off the entire internet.

The result of all this is that ChatGPT and other similar LLM's are glorified google search engines. Except they aren't searching the "now" so much as the "then" when their model was built. Of course every "correct" answer just reinforces the previous lookups which is how they "learn" after the model was built.

To illustrate I'll google a very simple but surprisingly common thing.

"how to call a web request with python"

And surprise the "AI" results are almost word for word the same as the common places tech guys like to go for answers.

Stack Overflow (seriously this is where like 90% of the worlds code gets made)

https://stackoverflow.com/questions/645312/what-is-the-quickest-way-to-http-get-in-python

and geekforgeeks (lots of tutorials here)
https://www.geeksforgeeks.org/how-to-make-api-calls-using-python/

Anyone who thinks devs write code from scratch is deluding themselves. Most dev work is breaking a problem into smaller tasks, then googling each of those tasks and copy pasta from stack exchange or some other tutorial, then modify it to suite your situation. Since all those "Code AI" were trained by scraping stack overflow and other such sites, "AI coding" is just taking the place of the google search. Each of these solutions get added to our own little collection for future reference. The best devs end up with these massive collections of functions and methods that they can pull from and like lego's use to build a solution.


That is to separate them from non-LLM systems like the watchdog I mentioned earlier. Those systems take non-language data and apply it to a purpose built model and use that to generate a new model that can be used to find existing patterns and predict new patterns. And the financial sector is absolutely using this to predict markets. Such things are still rudimentary as they have to be developed in secrecy but people are working on them.

Also, I forgot to mention another trait with real world market: a proven method that can make money, will eventually become less effective the more people use it due to market saturation.

If AI can really use past market research data and devise it's own successful projects. Pretty soon everyone will use AI to devise the same project and deploy the product. The market will eventually become over saturate so subsequent people who entered this market can no longer make money.

Which is also another reason why "market analysis" doesn't always get people a win. Because market experts uses data to say "This is popular", everyone makes the same thing, whoever entered late makes no money doing the same thing as everyone else.

Cab, service or value?
Airline, service or value?
Coffee Shop, service or value?
Grocery Store, service or value?
Walmart, service or value?
Dell, service or value?
Nvidia, service or value?
Intel, service or value?
Microsoft, service or value?

The horse is a tool, the driver is the human, this context is about how technology and automation can make skillsets obsolete and how humans need to adapt an acquire more relevant skillsets.


Human labor can broadly be broken into three categories.

No/Low Skilled workers, these are humans that perform tasks that can be taught within six months or less. These are people who are highly replaceable because more humans are becoming adults and entering this category every day.

Skilled workers, these are humans that perform tasks requiring six months to a year or more of training to perform their tasks. Replacing them can be expensive as they take much longer to grow, though not impossible. Large shifts in a market or technology can replace them.

Professional workers, these are humans with skills that take years to learn and a lifetime to master. They are virtually irreplaceable and are highly sought after and frequently poached. They are the least vulnerable to becoming obsolete, a lifetime mastering a set of skills usually comes with several adjacent skills and masteries that make them very flexible. When one market goes boom they can parley those adjacent skills into another market.


Most humans fall into the low/no skilled category. It's the easiest to get into, requires the least amount of continuous effort and allows for the most personal time to pursue non-work related hobbies. It also pays the least and is the least stable. Those who knuckle down an grind can get into the skilled workers category and get steady reliable income, or even build a business. Lots of personal time was spent learning skills and keeping up with trends, but it pays off because these guys aren't often unemployed. The last category is the rarest and anyone who gets there is having to turn down job offers all the time.

Every programmer has a bookshelf full of books called f'Object oriented programming in {language}', but they are all for show. Once you understand logic and basic workflows, the only book you need is

Of course people are working on them, they don't work. I've built my own, in fact my deterministic models are beating my ML models. One mistake can wipe out 50 successes. AI substantially better than a coinflip, but even a 90% success rate can send you to the poor house depending on how badly the misses are.

Not sure how my name got in the original quote box there, but that's hilarious! (and true!)

Ok I see a possibly better term than "service" now, how about "irreplaceability"? I think that may be a more accurate than "service".

You listed companies name instead of what they do, which isn't exactly "accurate description" because multiple companies has multiple departments ;p.

But I'll try my best to pick a few to answer.

MS: If you mean this product called "windows", it's platform with monopoly. So value in this case, the highest tier even. It's very, very hard to replace windows and kick it out of market. So it can't be replaced.

MS has plenty of other products like Xbox or game pass, they are easier to replace with lower value than windows.

Nvidia: Nvidia is the GPU designer that provided value and solutions to their buyers. TSMC(and countless other Nvidia supply chains)is the service provider for Nvidia in this case.

Intel: I am less familiar with Intel. I think Intel do both CPU design (value) and chip manufacturing (service). But their chip manufacturing department is currently losing to TSMC.

Intel's semiconductor manufacturing business case also shows how easy it is to lose the competition in the market as a service provider.

Dell: I think they sell laptops? though their product is quite replaceable in this case. People choose Dell laptop over another brand because their product has values to them. So Value in the case of Dell.

Dell doesn't actually make laptops, it's one of those big 5 laptop manufacturer like compal or pegatron that made laptops for them. So those laptops manufacturer are the service provider for Dell in this case.

Grocery Store: what unique value does grocery store provides? They are easily replaceable by e-commerce. Even you said you buy diapers online. And they don't make things, only sell them. So service in this case.

Cab: I see very little difference between hiring different cab drivers so service.

Anyways. If you ask if what they provide is replaceable or not, the answer should be easy. So here is your better term there.

So your answer help me get your stance, but


your argument here does not make sense to me, the "e-commerce" is still a "Grocery Store", unless you mean every individual food source has their own e-commerce front end so I go to 25+ sites to do my grocery shopping each week, which is the "value" grocery stores have is aggregation and logistics.

the idea of the value being "unique" differentiates you from a service means that once you have a competitor you're in the service business and not a value business, because you are now replaceable?

This is my favorite one

Chip manufacturing is service I suppose, but the value is the chip fabrication technology. Which is the difference between top tier foundries and the lower tier ones. As dimensions shrink and fewer and fewer people either have or are willing to invest resource into technology to actually make them. IP Design is also a service, but to make chips you need both fabrication technology and IP Development. (neither is any good without the other) Most of the fundamental improvement is coming from the technology development though.

China will have fabrication capacity at better density and power requirement than 3nm within 2 years max. It isn't going to be silicon as we know it, but it doesn't need to be.
https://www.livescience.com/technology/electronics/chinas-2d-chip-could-soon-be-used-to-make-silicon-free-chips

Maybe we could use a more detailed example. I don't see e-commerce and grocery store being the same thing here. They have different business models.

For example, You mentioned you prefer to buy Diapers on Amazon because they deliver the things to your house no?

In this case, Amazon provides unique value in the service, that made you choose it over another physical store.

On the other hand, 2 random grocery store on your street that sells the exact same thing at the same price has no unique value in this case. You'll probably just go to whichever is closest.

In this case, they did not have the same advantage as another e-commerce like Amazon, which delivers and sells cheaper. Those small grocery stores can be replaced far more easily than Amazon.

Keep in mind that most of the value provided by Amazon are unique advantages that only Amazon has because they are big corporation. It's very hard for other smaller stores to compete unless they sell unique things in demand that isn't available on Amazon.

So owning a grocery store is, by nature, a risky business that can be replaced pretty easily by other competitors due to that. This risk doesn't apply to Amazon.

A top tier chip manufacturer like TSMC does have pretty strong advantage over their competitors due to technology, but the disadvantage of being a service provider still exists: Their customers still has a chance to choose someone else.

If Trump say all chips from Taiwan will pay 50% tariffs for chips but South Korea chips only need to pay 10%, many American companies will use Samsung chips instead. It's exactly this reason that Taiwan index crashed harder than South Korea when Tariffs announced, because South Korea got lower tariffs.

It's also the reason why TSMC has to invest heavily in the US to please their clients, despite US fabs are losing money hard. They have to avoid the risk of losing the advantage for political reasons.

Technology can give TSMC advantage, but this advantage does not make them invincible as long as they remain as a service provider. It's the inherent disadvantage as a service provider imo.

Amazon lost the diapers war for me, we tried the mama bear ones out for something like 2 months, but they did not resolve the skin issue my little one was having (actually made it worse) so we swapped back to huggies which we get from costco.

even though I could buy them from amazon I by them from a physical store, I went to amazon for a unique product that they were selling that was supposed to meet a specific need which they failed to deliver on.


I would be interested in how you feel about the value business turning into a service when competition arises. which implies all business will become services as time marches on, unless they pivot away from that initial value offering.

Damn I almost spit my coffee out. Good on ya man.

If you are trying to pick winners and losers, essentially trying to count cards and cheat the market, you are going to lose. The issue here isn't the models but the information asymmetry. SEC reporting regulations are such that you won't receive the required information before it's effect hits the market. The model can't predict because it simply doesn't have the required information, and never will.

Do not try to pick stock prices, instead try to predict trends of index funds and other basket type assets. In this way the impact of unknown reports behind any set of actors is mitigated. When conditions X/Y/Z are met then valuations on an industry will go up.