More

remexre · 2026-01-15T02:32:18 1768444338

https://en.wikipedia.org/wiki/Lexer_hack

Make your parser call back into your lexer, so it can pass state to it; make the set of type names available to it.

remexre · 2026-01-14T18:52:08 1768416728

The point of an IRB is to act as an outside reviewer of _ethics_. IRBs aren't some checklist thing admin put in to protect the University's reputation, they exist as a direct reaction to huge amounts of unethical human experimentation occurring last century.

jovial_cavalier · 2026-01-15T21:23:45 1768512225

The point of an IRB is to stop you from nonconsentually sterilizing people. As long as the system stops that from happening, I don't care about the paperwork. It's not my concern.

The "ethical" issues with this study do not rise to the level that I care, so the only objection is that they didn't get the IRB to rubber stamp it beforehand, which I also don't care about.

remexre · 2026-01-11T22:42:19 1768171339

There's usually an if(temp == 0) to change sampling methods to "highest probability" -- if you remove that conditional but otherwise keep the same math, that's not deterministic either.

Majromax · 2026-01-11T22:49:00 1768171740

If you remove the conditional and keep the same math, you divide by zero and get nans. In the limit as temperature goes to zero, you do in fact get maximum likelihood sampling.

dnautics · 2026-01-12T00:03:40 1768176220

if (t==0) argmax(logits) else pick(logits)

eru · 2026-01-12T08:16:05 1768205765

You are ignoring the limit taking.

dnautics · 2026-01-12T15:39:03 1768232343

If t < 0.1 return {error, 400}

eru · 2026-01-13T01:28:22 1768267702

Do you know that a mathematical limit is?

TomatoCo · 2026-01-11T22:47:35 1768171655

I'd assume that's just an optimization? Why bother sorting the entire list if you're just gonna pick the top token, linear time versus whatever your sort time is.

Having said that, of course it's only as deterministic as the hardware itself is.

dnautics · 2026-01-12T00:05:08 1768176308

The likelihood that top-two is close enough to be hardware dependent is pretty low. IIUC It's more of an issue when you are using other picking methods.

embedding-shape · 2026-01-11T22:47:19 1768171639

In for example llama.cpp? Specific to the architecture or in general? Could you point out where this is happening? Not that I don't believe you, but I haven't seen that myself, and would appreciate learning deeper how it works.

remexre · 2026-01-06T15:07:22 1767712042

What properties are you validating? ld.so/libdl don't give you a ton more than "these symbols were present/absent."

remexre · 2026-01-06T15:01:59 1767711719

If you're trying to audit React, don't you either need to audit its build artifacts rather than its source, or audit those dev dependencies too?

remexre · 2026-01-04T16:53:25 1767545605

Or more like,

    x = tokenize(input)
    i = 0
    do {
      finish, x = layers(x)
    } while(!finish && i++ < t_max);
    output = lm_head(x)

oofbey · 2026-01-05T00:43:08 1767573788

That’s closer still. But even closer would be:

    x = tokenize(input)
    i = 0
    finish = 0
    do {
      p, x = layers(x)
      finish += p
    } while(finish < 0.95 && i++ < t_max);
    output = lm_head(x)

Except the accumulation of the stop probabilities isn’t linear like that - it’s more like a weighted coin model.

remexre · 2026-01-02T10:39:27 1767350367

How would algebraic types work with SLIME? If I remove a constructor from my algebraic type, what happens to values of that type that are built with that constructor that're stored in globals?

In the same way that non-hygienic macros in a Lisp-2 with a CL-style package system are a local optimum, many non-obvious design choices in the Common Lisp type system and CLOS make SLIME "just work" in almost every case.

kscarlet · 2026-01-02T11:20:08 1767352808

I guess this case is workable similar to struct redefintion. There can be a condition and a CONTINUE restart, which makes instances of the removed constructor obsolete.

remexre · 2026-01-01T04:36:36 1767242196

> get into "unsafe" states on purpose

see, this seems like something that's nice to actually put into the types; a Ptr<Foo> is a real pointer that all the normal optimizations can be done to, but cannot be null or otherwise invalid, and UnsafePtr makes the compiler keep its distance and allows whatever tricks you want.

remexre · 2025-12-26T17:39:46 1766770786

This only covers what's in https://wiki.osdev.org/Bare_Bones, but in Rust, right?

remexre · 2025-12-21T00:16:36 1766276196

For each token generated, you only send one token’s worth between layers; the previous tokens are in the KV cache.