Hacker News new | past | comments | ask | show | jobs | submit | quirkot's comments login

Seems kinda of like a ... lonely mountain


That app update screwed me up for weeks. Feeling that sweet sweet vindication right now


Is this true?

>> Do not panic! A lot of the large LLM vocabularies are pretty huge (30k-300k tokens large)

Seems small by an order of magnitude (at least). English alone is 1+ millions words


Most of these 1+ million words are almost never used, so 200k is plenty for English. Optimistically, we hope that rarer words would be longer and to some degree compositional (optim-ism, optim-istic, etc.), but unfortunately this is not what tokenisers arrive at (and you are more likely to get "opt-i-mis-m" or something like that). People have tried to optimise tokenisation and the main part of LLM training jointly, which leads to more sensible results, but this is unworkable for larger models, so we are stuck with inflated basic vocabularies.

It is also probably possible now to go even for larger vocabularies, in the 1-2 million range (by factorising the embedding matrix, for example), but this does not lead to noticeable improvements in performance, AFAIK.


Performance would be massively improved on constrained text tasks. That alone makes it worth it to expand the vocabulary size.


Tokens are often sub-word, all the way down to bytes (which are implicitly understood as UTF8 but models will sometimes generate invalid UTF8...).


BPE is complete. Every valid Unicode string can be encoded with any BPE tokenizer.

BPE basically starts with a token for every valid value for a Unicode byte and then creates new tokens by looking at common pairs of bytes (‘t’ followed by ‘h’ becomes a new token ’th’)


Just wait till an aggressive lawyer finds out about this. Gonna be a lot of companies taken to the cleaner for hallucinations


Hell yeah. You accumulate $100m (or whatever the limit is) and we throw a parade for you. You get an award and your name on a plaque. You won capitalism! and you're not allowed to make another cent


To what end? Innovative companies are literally creating money that did not exist before. Do you think there is a fixed pie of money?


Real, per capita economic growth only really started with the Industrial Revolution - when companies started literally making money.

But the mindset of many is still caught up in the uber-conservative (from an economic point of view) mindset dating to the middle ages that wealth is like land: it's just always there, and if you want more, you have to get it from someone else buy buying or seizing.


If you mean the net worth of money, which is zero, yes it is fixed. For one person to have a dollar, another needs to be in debt.

Also, for one person to make profit, they need to spend more than they earn, which inherently means the rest of the economy is in debt. Now you might say the profit will be spent, but then it isn't profit anymore.

Oh and if you want to talk about the real economy, entropy always rises, so no free lunch there either. When the economy grows, all that happens is that the human domain expands and humanity exerts more authority over the processes that occur within its domain. This by definition results in a loss of agency of the non human domain.


The Federal Reserve creates money. This is Econ 101. There is a fixed pie of money it's called the M2. Again, Econ 101.

Perhaps you meant "value"? Be careful though, if you start differentiating between money and value, you arrive at communism. This is literally the first and last chapters of Capital vol 1.


Obviously the monetary policy is important. The innovation part is deflationary and the money creation part is inflationary, if they cancel each other out you end up with more money supply and the same price of goods.


There are usually two reasons people advocate for a cap on earnings. Either they think that every dollar earned makes another person a dollar poorer, therefore a cap limits poverty, or they have some deep seated burning fire of envy. I saw it in university, activists enraged for the sake of being enraged that someone else had done well for themselves.

Bizarre isn't it? If I were to commit one of the seven deadly sins, I'd choose a funner one than envy.


Good code is the simplest code you can write to get the job done.

Getting too excited about techniques is a form of scope creep


It depends on what it means to be to get the job done.

Write some code and generate immediate results can be considered a done job. If you want to reuse the code a year later and found an unreadable mess though, not so much.


More like fulcrums on which the world moves, amirite?


>> designed by former SpaceX & Tesla earthlings

Man, for a company that is passionate about space stuff, they sure don't seem to care at all about experiencing the night sky. tent made out of glow tubes? madness


> Since the Neobank doesn’t have its own FDIC insurance and bank charter, the neobank customer’s deposits actually sit on the sponsor bank’s balance sheet

Are you saying that "neobanks" do not record the deposits as a liability? Or that they do not record the cash as assets? I don't understand what you are trying to say.

Are there any public "neobanks" you can point to so that we can see the financials as you describe them?


Getting a bank charter and FDIC insurance is a huge undertaking, so neobanks are usually not actually banks. They are technology companies that have partnered with banks or are customers of banking-as-a-service providers (Hydrogen, Stripe Treasury, and similar).

The banking partners have their own bank charter and FDIC insurance, but the technology company itself does not. If you go to most neobank websites, you can scroll to the bottom and see a disclosure along the lines of "Banking services provided by The Bancorp Bank, N.A. or Stride Bank, N.A., Members FDIC".


That doesn't have anything to do with financial statements. If you're Joe's Lawn Service and someone hands you money, it goes on your balance sheet. That's how accounting works, every penny gets (you know) accounted for. Otherwise how do these "neobanks" know who they owe?


None of the deposits are on the technology company's balance sheet because the neobank doesn't actually handle any of the money. They are essentially UIs on top of APIs exposed by the bank.

If you make a deposit into a neobank, you'll have a routing and account number, but the routing number belongs to the partner bank servicing that account. If you deposit cash or a check or whatever, it goes _directly_ to the partner bank.


I think they are describing what is termed "brokered deposits" -- like your Starbucks gift card account isn't managed by Starbucks, but rather a (usually large) bank on the backend which treats brokered deposits differently than unbrokered (standard) deposits.


Hmmm, possibly. However you are describing an operational reality and the author was using financial reporting terms. And specifically if something like "people gave a bunch of money to what they thought was us" isn't on the balance sheet, it's a big red fraud flag


This graveyard is such a great list of fads. Reading thru it I remember how hot each on of the services google was mimicking was for a time. Ahh.. google. The world's biggest Also Ran


Consider applying for YC's Fall 2025 batch! Applications are open till Aug 4

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: