Hacker Newsnew | past | comments | ask | show | jobs | submit | Rexxar's commentslogin

I don't work regularly on it but I have a proof of concept go to c++ compiler that try to get the exact same behaviour : https://github.com/Rokhan/gocpp

At the moment, it sort of work for simple one-file project with no dependencies if you don't mind there is no garbage collector. (it try to compile recursively library imports but linking logic is not implemented)


It's probably just a technical accounting update. Old assets are often kept valued at their buy price and not reevaluated every year to avoid taxes (Banque de France is not exempt from taxes). As they swap a type of gold by another and do a sell/buy action, the new gold is valued to current market price while the old one was valued in accounting at an old value.

They had a deficit last year, so they can probably avoid to pay tax this year by balancing last year loss with this year profit.


  > Someone didn't get the memo that for LLMs, tokens are units of thinking.
Where do you get this memo ? Seems completely wrong to me. More computation does not translate to more "thinking" if you compute the wrong things (ie things that contribute significantly to the final sentence meaning).

That’s why you need filler words that contribute little to the sentence meaning but give it a chance to compute/think. This is part of why humans do the same when speaking.

The LLM has no accessible state beyond its own output tokens; each pass generates a single token and does not otherwise communicate with subsequent passes. Therefore all information calculated in a pass must be encoded into the entropy of the output token. If the only output of a thinking pass is a dumb filler word with hardly any entropy, then all the thinking for that filler word is forgotten and cannot be reconstructed.

Do you have any evidence at all of this? I know how LLMs are trained and this makes no sense to me. Otherwise you'd just put filler words in every input

e.g. instead of: "The square root of 256 is" you'd enter "errr The er square um root errr of 256 errr is" and it would miraculously get better? The model can't differentiate between words you entered and words it generated its self...


It's why it starts with "You're absolutely right!" It's not to flatter the user. It's a cheap way to guide the response in a space where it's utilizing the correction.

People have researched pause tokens for this exact reason.

What do you think chain of thought reasoning is doing exactly?

You’re conflating training and inference

The input should be the range and the distribution of probability on this range. Intuitively we have a tendency to assume an uniform probability for range [-1, 1] which is not the case if we check every doubles.

https://www.legifrance.gouv.fr/

If you go on particular laws, you can see the previous versions and how it changed. Example: https://www.legifrance.gouv.fr/codes/section_lc/LEGITEXT0000...

Click on "version" then "comparer" buttons and you will see a diff.


Original zlib code: https://github.com/madler/zlib/blob/develop/contrib/puff/puf...

The code of the article: https://github.com/ieviev/mini-gzip/blob/main/src/main.rs

Top level declarations of the C code:

    #define MAXBITS 15   
    #define MAXLCODES 286
    #define MAXDCODES 30
    #define MAXCODES
    #define FIXLCODES 288

    struct state

    local int bits(struct state *s, int need)
    local int stored(struct state *s)

    struct huffman

    local int decode(...)
    local int construct(...)
    local int codes(...)

    local int fixed(...)
    local int dynamic(...)
    int puff(...)
Top level declarations of the Rust code:

    const MAXBITS: usize = 15;
    const MAXLCODES: usize = 286;
    const MAXDCODES: usize = 30;
    const FIXLCODES: usize = 288;
    const MAXDIST: usize = 32768;
    struct State<'a> 
    struct Huffman<'a>

    fn decode
    fn construct
    fn codes
    fn fixed
    fn dynamic
    fn stored
    pub fn inflate

..that doesn't proves anything.

it's just convenient configuration macros and generic procedures names being reused to implement the steps for the DEFLATE algorithm.

There are only finites ways for you to express a piece of code in a language.


The difference in size seems to be mainly due to the missing documentation before each function.

Indeed, I see they now wait 28 days to not been displayed in green before commenting.


Did you have envisioned the possibility that your were the one that have missed other people joke ?


It was my joke, at the top of this thread and the only one in this thread. Have you considered the possibility that you have no idea what is going on here (your lack of fluency in English is a likely factor) and that you have no business poking your nose into it, person who is attacking me with a hostile rhetorical question for no good reason? The fact is that I do put quite a bit of effort into being sure that I understand the context--a lot more effort than you and others here have.

It was only hard because UK wanted to stop immediately to participate to budget while continuing to benefit from already agreed multi-years policies.


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

Search: