Models can't scan anything.

They return instructions for you to do something, and you or a script you permit chooses to execute what the model tells you and return the result to the model.

> if you understood what they think they are building and the culture inside of anthropic you would understand why they did it.

This seems like a cult with extra steps.

Related: I interviewed for Anthropic a few months ago and in place of the usual HR call they have one where they have someone with a suspiciously relevant degree grill you about how committed you are to the 'mission'!

I probably came off as being skeptical, and then, hilariously, I was strongly encouraged to read the book published by the CEO to 'form accurate opinions' on AI safety.

> the quality of that model

I guess the benchmarks disagree, but whenever I need to find specific information that does not easily show up with a web search, I try chatgpt, gemini and grok. Grok surfaces what I was looking for more often than the others.

Things like "find the github repo from 2017 that does $vague_thing".

Grok does seem to have the best searching capabilities, and not just for twitter. I wonder what search engine they’re using on the backend.

Good question. You can actually see the searches it runs (momentarily) so testing could determine if it's using public search engines or a private system.

I find that too. I use Claude for coding but when I need to dig out something based on limited data I turn to Grok and it delivers.

Can you give a specific example (that doesn't violate any privacy you want to protect)?

Yes, I’ve tried it. For example, this was my winning entry from a year ago [0]. The LLM only performs trivial obfuscation, not advanced transformations.

For example:

  if (x == 1 || x == 2) { ...

can be transformed into:

  if (!(2+x*x-3*x)) { ...

An LLM will do this if you explicitly ask it to, but not on its own.

[0] https://github.com/ioccc-src/winner/blob/master/2024/macke/p...

One of the main instruments of obfuscation (and the way to get more out of the size constraints) is making the code as short as possible, so in that example you'd prefer

    if (!(x/2-1)) { ...

EDIT: Oops, confused the original with x==2 || x==3. Instead, we can use !(x-1>>1), which precedence rules parse as !((x-1)>>1).

I checked this in an online C-compiler and it is not equivalent to x==1 | x==2.

this statement is equivalent to x==2 | x==3.

For example, x=3, 3/2 = 1 then 1-1 = 0 so that !(0) is 1 or true. Also for x=1, 1/2 = 0 then 0-1 = -1 and !(-1) = 0 or false.

I agree with your point in general though about size constraints.

It's pretty easy to see what that does though, even if it is shorter. Wouldn't the other approach be more obfuscated?

extrano84 already found some errors but also 0 will fail and if x is int (instead of unsigned int) all negative numbers will also fail (but so will the original s-macke obfuscation as well).

Not sure what you mean by advanced transformations but I got these versions from ChatGPT without explicit instructions.

  if(x-1<2&&x)...
  if((1<<x)&6)...
  if(x<3&x)...
  if(3%x&&x<3)...
  if(!((x-1)*(x-2)))
  if(!(x^1|x^2))...
  if(!(x*x-3*x+2))

(1) fails for negative numbers

(2) fails for (x % 32 in [1, 2]) due to UB

(3) fails for x == 2

(4) crashes for x == 0

(5) is the same polynomial, only factorized

(6) always returns 0

(7) is the same polynomail reordered

But this works for any value of x:

  x<2==2-x

or for something shorter:

  x<3&x>0

but that's not bery obfuscated

But then we all know that LLM has come a long way since one year ago.

Are you sure they still can't do it?

Just two months ago I tried to write a short K code with Claude Opus 4.6, only to find that while it had sufficient knowledge about K vocabularies it didn't try to make good use of them. K is, while slightly obscure and obfuscated, a real programming language and certainly better known than obfuscated programming. I don't have high hope for IOCCC-grade obfuscation.

The best thing about leveraged schemes like this is that if/when it fails it takes down everyone involved.

> You assume that the EU acts in a coherent manner.

When it comes to aiming for maximum extraction from big tech, yes.

What does a software breakthrough look like in your opinion?

If you get yourself to define it, maybe you'll find it achievable :)

At its peak, Apple accounted for 70% of Foxconn's business and there are plenty of Apple folks directly working with them.

It's an artifact of geopolitics that Apple and Foxconn and separate legal entities.

There really is none.

There were, if I recall correctly, one or two generations of Dell XPS that came close but they started cheaping out on materials almost immediately after it.

The old-ish ThinkPads were great if you want a rugged laptop, but that's not really because of better build quality, there were just more material.

I love how the point being made is a tautology :

AI is not conscious because it is not conscious.