I keep hearing we don’t know “how llm’s work”, in mean yes we know the algorithms but the WHY I suppose. I’m not sure if that’s really true, does the best researcher at OpenAI, Anthropic, Gemini not know why it works? Would they say that?
Anyway these discussions always make me think we are too generous to humans. I know very few humans who are good at reasoning. It’s surprisingly hard to just think really hard through a problem. We mostly intuit, act, repeat. It’s a rare thing for someone to deeply reason.
The other problem I have with this is that we apply the word “reasoning” here because we haven’t really got another language for it, so we anthropomorphise the llm because that’s our reductive mental model and then complain that it’s not human enough.
We've come a long way since the first LLMs were created. They originally produced a lot of surprises that were very hard to explain, it's true. But current research is very tightly directed at producing forward results; it's not just throw in more training data and make a bigger model and see what happens anymore (although there's still a bit of that too). The researchers have much more advanced understanding of how they work now, it just doesn't propagate out to the public which is still repeating "autocomplete on steroids" years later.
This is just human behaviour though. We're wired for "a lot of people who do X, also do Y". "this person does X, therefore they must do Y". Obviously, not all brown things are cows, but that's how it be, it's got nothing to do with ai.
No, we don't. It's currently an impossible task, yet our answer to it has implications in the present, so we must do our best with incomplete information. FWIW that's how we treat practically all problems of consequence. A great many things happen to be unknown and unknowable.
I’m not sure why you think protectionism (by the local government) can’t kill an industry. It happened to American steel as well: the American steel industry was “protected” while foreign steel companies innovated their production processes.
The timeline goes like this:
1. Government gets pressured by industry to protect domestic industry from outside competition
2. Domestic industry stagnates as it has no reason to make competitive products
3. Time passes, quality/value gulf becomes so large that it becomes impossible to catch up to competition. It becomes impossible to export your goods and even domestic customers look elsewhere.
If the USA dropped tariffs and bans against Chinese car companies, American car companies would need to improve rapidly to survive. Instead, they’re selling us $50,000 decade old dinosaur designs like the Chrysler Pacifica.
This literally already happened to them with Japanese cars. And let me tell you, you don’t want to drive a Chevrolet from before Honda and Toyota sold cars in the United States.
Not protectionism related, but look at the MacBook Neo competitors shown off by Dell and HP at Computex. Competition drives innovation and cost reduction for consumers. Apple forced HP and Dell to make an affordable aluminum chassis computer for the masses and consumers are the winner in the situation.
It's also said to be one of the ironic long term impacts of the Jones Act (of 1920) in the US. The Jones Act among other things requires all US flagged ships at sea to be manufactured in the US in a hope to protect US manufacturing of ships, but instead in practice just became that most non-military ships just aren't officially flagged as US ships ever. For instance there's generally only one or two US flagged cruise ships in existence in a given year, they get US flagged through manufacturing loopholes (primarily constructed where other cruise ships are, but with enough "finishing touches" in the US to count as US manufactured), are smaller than the average cruise ship, and mostly only exist for Hawaiian island hopping cruises. (This is due to another quirk in the Jones Act that ships traveling directly between two US ports need to be US flagged. Most cruises can easily stop at a non-US port in between US ports, such as various Canadian, Mexican, and many Caribbean island ports, but from Hawaii the next non-US ports can be quite a bit more travel time and traveling to Tahiti and back just to see two different Hawaiian islands can be a bit much.)
Since the closure of Jeffboat in 2018 (among others), there's no non-military ship manufacturers in the US for anything even resembling cruise ship size (and even before that manufacturers like Jeffboat were mostly only making cargo barges) and the US military contractors have basically shrunk to a duopoly. Probably not the 100+ years later expectations of the Jones Act authors.
Correct, protectionism still saved the industry for years, it did its job perfectly.
Look at the car makers in Europe, they had decades to produce affordable electric cars, even lawmakers pushed their own population to buy electric cars with timelines and incentives, and they still failed miserably. How?
Tariffs are a last-ditch effort to buy more time. If you fail to use this time effectively, you will close shop. If you don't have tariffs, you close shop today.
Europe sells gasoline shitboxes for €30K when China could easily sell EVs for €20K, Europe (and the US) already lost.
> protectionism still saved the industry for years, it did its job perfectly.
This is under an assumption that the alternative would have been eradication of the industry, as in, “if we don’t protect the industry it will go out of business.” In reality, leading an industry to be more competitive is a lot better than protecting them from outside competition in the long term.
What was the economic value of protecting US steel for half a lifetime and then letting it turn into a dead industry? Wouldn’t it have been better for that protectionism to be more in the form of encouraging competition so that US steel companies were still thriving today? If the end result is going to be a dead industry why invest in protectionism at all?
It’s funny that you’re talking a bunch of smack about European electric cars without looking up sales numbers. The Volkswagen group EVs outsell Tesla in Europe. The top selling EV in Europe is a Skoda. VW and BMW are the top two EV companies globally besides Tesla and 4 Chinese brands.
and it's not just the end product, it's the process. China has built almost entirely automated car factories. They're already 10+ years ahead of what's happening in the US/Europe.
it's a solution looking for a problem and google are desperate to stay relevant there.
We just don't need to search as much. But I _do_ still want to search sometimes, it's still a valid use case, just not as important as it used to be.
But when a do search, I want simple, relevant, external search results so that I can go straight to those good sources. Google isn't satisfied with their returns on that though.
the thing that bothers me is I don't usually want this mode. When I search, I am not looking for what google thinks, I am looking for what other sources think.
Yes! And for actual research, I like to see a variety of sources and opinions, as well as some links to granular or primary data if apt. I don't even like A.I. summary of product reviews because they're SO bland and not needed. Even aside from the traffic-choking, content stealing insult to legit creator sites, it's like a giant, insidious flattener of human color and interest.
