I personally like being able to choose because I understand the tradeoffs and want to choose the best one for what I’m asking. So I hope this doesn’t go away.
But I agree that they probably need some kind of basic mode to make things easier for the average person. The basic mode should decide automatically what model to use and hide this from the user.
No, Mixture of Experts is a really confusing term.
It sounds like it means "have a bunch of models, one that's an expert in physics, one that's an expert in health etc and then pick the one that's a best fit for the user's query".
It's not that. The "experts" are each another giant opaque blob of weights. The model is trained to select one of those blobs, but they don't have any form of human-understandable "expertise". It's an optimization that lets you avoid using ALL of the weights for every run through the model, which helps with performance.
Usually when we’re doing it in practice there’s _somewhat_ more awareness of the mechanics than just throwing random obstructions in and hoping for the best.
LLMs are still very young. We'll get there in time. I don't see how it's any different than optimizing for new CPU/GPU architectures other than the fact that the latter is now a decades-old practice.
Not to pick on you, but this is exactly the objectionable handwaving. What makes you think we'll get there? The kinds of errors that these technologies make have not changed, and anything that anyone learns about how to make them better changes dramatically from moment to moment and no one can really control that. It is different because those other things were deterministic ...
In comp sci it’s been deterministic, but in other science disciplines (eg medicine) it’s not. Also in lots of science it looks non-deterministic until it’s not (eg medicine is theoretically deterministic, but you have to reason about it experimentally and with probabilities - doesn’t mean novel drugs aren’t technological advancements).
And while the kind of errors hasn’t changed, the quantity and severity of the errors has dropped dramatically in a relatively short span of time.
It's the whole answer being correct that's the important thing, and if you compare GPT 3 vs where we are today only 5 years later the progress in accuracy, knowledge and intelligence is jaw dropping.
> I don't see how it's any different than optimizing for new CPU/GPU architectures
I mean that seems wild to say to me. Those architectures have documentation and aren't magic black boxes that we chuck inputs at and hope for the best: we do pretty much that with LLMs.
If that's how you optimise, I'm genuinely shocked.
In my experience, even simple CRUD apps generally have some domain-specific intricacies or edge cases that take some amount of experimentation to get right.
I didn't say the average person should use it, I was thinking the search functionality could be implemented that way and save on burning up the planet to tame Moloc.
If it could be implemented that way and would be helpful where Google fails, then why has nobody done this? I’d love to try this out if you can point to a product that does this.
You can criticize LLMs all you want, but the fact is that they provide value to people in ways that alternatives simply don’t. The energy consumption is a concern, but don’t pretend there are viable alternatives when there aren’t.
I don't have insight on what goes on inside Google - they may well be doing this at some level, but their business isn't finding stuff, it's selling ads, so getting the search right is a very low priority.
The LLM people are heavily invested in ever bigger models to keep the research money flowing in, it wouldn't make sense to release a service that undercuts that.
that leaves independent actors - presumably building and maintaining an up to date database is difficult, so only the big search engines do.
I don't think this is _literally_ a search through vector embeddings.
LLMs store embeddings of individual tokens (usually parts of words), so a result of an actual search will be top-k embeddings and the corresponding tokens, similar to the output of a Google search. You could extract the initial matrix of embeddings from some open-weights model and find tokens closest to your query. However, it's not clear why do this. OP got coherent text, so that's not search.
It's _similar_, though, because attention in LLMs basically looks for most similar tokens. So to answer the question about the term, the LLM had to create a stream of tokens that's semantically closest to the given description. Well, this is somewhat like a search, but it's not exactly the same.
I disagree with this. Jan 6th didn’t affect 99% of peoples lives directly. It was clearly bad, but few people saw impacts in their own lives.
Higher prices and a possible recession will affect every person in the country and even globally.
His MAGA base might not blame him, but that’s only like 30-40% of the electorate. The other 60-70% won’t be happy if their lives are negatively impacted.
IMO the only thing that can get through is actual personal consequences for the voter themself
Well, yes. And his approval rating has been steadily declining in tandem with the stock market declines he's caused. If/when prices suddenly skyrocket because of tariffs, you can bet his approval ratings will decline further.
That won't work. WaPo, LAT, etc. have run numerous articles where MAGA voters talk about the consequences of his actions to them or immediate family members (getting laid off, losing medical coverage, deportation, etc.) and they say it's worth the sacrifice because they just know that Trump is actually looking out for them and won't let them suffer for too long.
What exactly on this list is a day to day item that is going to adversely affect the middle class/blue collar Trump voter enough to cause them to flip their alliance? On that list maybe clothes and shoes. But even still those are generally bought only a couple of times a year and any increase will be griped about in the moment, but come election time will be forgotten.
There are only four things that will make a middle class voter feel pain enough to re-align their vote. Fuel/energy costs, general food costs, rent/housing costs, and job insecurity. If these tariffs do not adversely touch those areas, they will have little to no impact on switching votes. Also of note, over the last four years those items are the ones that drove Trump back into the Oval Office.
You may be right here. The tariffs might not directly cause enough pain for the average person to matter.
The big unknown still is the impact on the economy and job market. His actions may reduce competitiveness of American companies globally due to retaliatory tariffs and resentment.
If the tariffs cause a recession, people will punish him for it.
My guess is that the stock market losses will be a greater driver than any cost increases on goods. And stock market losses will not necessarily push the middle class much, but will push GOP politicians into contention with Trump.
Ultimately Trump needs GOP in congress to be friendly…more than he needs the public at this point. He cant (and wont, despite the trolling) run again.
Well he needs the public not to vote in a bunch of democrats in the midterms. And, if his MAGA movement is going to live on in Vance or others after his term, he will need public support.
Middle class people also have to buy new cars, maybe want to treat themselves with some foreign alcohol, they still need furniture, coffee is still one of the most traded commodities in the world and americans guzzle it down.
New cars - They are not purchased very often by the middle class and they will lean domestic if they desire new or buy foreign cars on the used market if they really desire them.
Foreign Alcohol as a treat - If it’s a “treat” it’s already likely a more expensive choice than a domestic equivalent and any cost increase becomes part of the luxury of it being a treat. If your $60 bottle of Italian wine that you occasionally treat yourself with is now $80, you won’t really notice that $20. If you do, maybe you opt for a better domestic instead. If the $20 california wine you daily drink becomes $30, you notice that. But I assert that there is quite a bit of exceptional alcohol produced in the US and the middle class electorate is not going to starve for decent alcohol.
Furniture - Like cars, this is not a regular purchase and frankly not one where a price comparison with a prior purchase of a similar item will really notice any price increase due to tariffs. How often do you replace a dinner table? 10 years? Of course it’s more expensive than the last time you needed one.
Coffee is the one example where you have daily consumption and like I said on my original comment, food is one area that if affected, people will notice.
> It was clearly bad, but few people saw impacts in their own lives.
It did though, they just didn't know how to measure it, and it wasn't felt immediately. It was like the flash of light that dazzles before the pressure wave of the nuclear bomb blasts everything (which in the analogy is this moment, now).
What happened on Jan 6, and in the leadup and response to it, was the erosion of democratic norms. Before Nov 2020 they were stronger, and after Jan 6 they were significantly weakened. Our institutions are essentially built on trust, and Trump in his campaign to overturn the 2020 election spent every waking moment for months attacking those foundations. He purposefully eroded people's trust in Democracy for no reason, because there ultimately the fraud he alleged in that election was not found.
That impacts everyone. They just don't feel it in the supermarket; they just have no "democracy meter" that they can use to gauge how healthy their representation is in government. But the reason he's able to do what he's doing now is he because he laid the foundation in 2020.
What in your mind should have happened differently in response to Jan 6?
No widespread fraud was ever proven that could have swayed the election.
Should we have pretended there were major flaws with our voting system just to help you with your feelings?
Or maybe we should have let Trump be president again for no reason just because you were really upset about it.
And did you ever wonder why, if Democrats were able to magically steal the election in even red states like Georgia and Arizona in 2020, they didn’t bother to try it again in 2024?
The elite reaction to January 6th was just raw hostility towards the American nation.
I don't claim to have access to secret knowledge about the legitimacy or illegitimacy of the elections. My view on the actual election fraud claims is agnosticism. I have no access to information that would allow me to independently come to any conclusion on the matter.
However, a large volume of very plausible evidence was put forward. And, instead of honest engagement with those concerns, we got extreme censorship, gas lighting, and a violent crack-down on everyone involved.
We cannot allow people that have this attitude towards us to continue to rule over us.
A violent crackdown against who? The people who stormed the capital and assaulted police officers?
Even AG Barr said there was no evidence of widespread fraud. The “plausible evidence” you speak of was a firestorm of unsubstantiated claims on social media that incited a violent attack on the capital.
Yes. Put your actual decision makers on social media, and have them engage openly with the people making unsubstantiated claims.
> Even AG Barr said
Almost all of the information that was put forward that seemed plausible was deleted from the internet, and never addressed.
We can figure out what happened after we get rid of everyone that played a role in that; once we have a truth-finding apparatus that is made up of friendlies.
The only thing that matters from all of this is that unfriendly people are in power, and the only solution to that is to get rid of them and replace them with friendly people.
Trump, for all of his many flaws, at least pretends to be friendly.
At the end of the day we either live in a world where laws, process, and provable facts prevail - democracy; or we live in a world where conjecture, conspiracy, and opinions and ad hoc decision making rule the day - anarchy.
I prefer to live in democracy, where we follow a process to redress grievances. In 2020, President Trump's claims were given great deference. He was given the opportunity to prove them in court. He was heard by state legislatures and governors. The vice president weighed his claims. The Congress did as well. Even after the election states like Arizona handed over their voting machines to groups like the "Cyber Ninjas" who attempted to prove claims that ballots were tampered with in that state.
Nobody found the evidence Trump claimed.
Because it doesn't exist, it never did.
Because the alleged fraud did not happen.
What happened was a man lost a close but fair election. That's what the facts show, despite any threads you feel are "seemingly plausible", it was the most audited election in history. Eventually if you can't put up, you really have to shut up. It's that simple when it comes to a) living in a democracy and b) being an adult. Sometimes you don't get your way, and the response to that cannot be to burn down the entire system.
That's what I mean when I said that this impacted everyone - when childish temper tantrums like Jan 6 are allowed to stand, when the people who acted that way are pardoned and the person who instigated that event is reelected.... well that to me means we are trending away from democracy and toward a different way of dealing with reality.
Great idea, I’m sure that would have helped. Maybe they also could have had Anthony Fauci personally reply to antivax conspiracy theories on Twitter. That would have won hearts and minds.
You think Trump is “friendly” and of course he is. He is the primary beneficiary and spreader of lies about the 2020 election, so why wouldn’t he also be the best arbiter of truth on the subject?
Don't expect much. Modi's overnight demonetization of Rs. 1,000 bills back in 2016, caused a lot of inconvenience to almost all the Indians for 3/4 months at-least. Demonetization and flawed implementation of GST caused many small scale companies to shut doors.
With media in their pockets they can get away with anything.
I hate to break it to you, but Linux has always had a very significant amount of development done by and for various corporate interests. There are whole companies that exist solely to commercialize Linux.
People learn best in different ways. Some learn best by reading, some by tinkering, some by watching and listening. I heard this over and over in school and college.
I don’t think it has anything to do with reading speed. When taking in complex technical information, you spend more time thinking and trying to understand than actually reading.
If you’re finding that you can quickly race through content, it probably just means you find the content easy to understand because you’re already familiar with many concepts.
Yeah, this is not ideal. I’m hoping that the breaking changes don’t affect the code at my work, since we also had to spend multiple years on a major .NET Core transition. I want the faster compiles right away, not in a few years.
Also, most laptops will run at significantly worse performance when not plugged in. Macs are much more consistent both thermally and when not plugged in.
Nope, not at all. Stocks represent partial ownership of a company. If you buy all the stock for a company, you own the company. Companies are valuable because they (ideally) make a profit for the owners.
Bonds are valuable because the company or government that issues them signs a contract to pay out a certain amount on them. They won’t always do that, but thats why we have rating agencies to help investors understand the risk.
The closest equivalent to crypto in traditional assets is probably precious metals and gemstones, though even those typically have some industrial uses to drive demand.
Diamonds are a great example of the downside of looking at the market value of something alone. The natural diamond market has been massively disrupted by the rise of artificial diamonds and diamond alternatives.
Great argument. "I don't believe it has value, so it doesn't." Well, keep telling yourself that. I'm sure you feel just as smart as you did when you dismissed bitcoin at $1,000, and you'll feel just as smart dismissing it at $1,000,000.
I’m not dismissing it. I’m saying that it’s different than traditional investments since most of the potential uses are still speculative, whereas it’s pretty obvious what value you can get from ownership of a company or a barrel of oil.
At least gold has real world uses, such as in electronics and spacecraft. Keeping large quantities of it safe for the future seems prudent, even outside its use as a value store.
But I agree that they probably need some kind of basic mode to make things easier for the average person. The basic mode should decide automatically what model to use and hide this from the user.
reply