vidarh

43 ms·

961.
by vidarh 9mo ago
> I don’t know how to respond to that other than to ask you to quote the part of the blog post where the author described the language model running into a problem that it could not fix and then described the details of how he manually i
962.
by vidarh 9mo ago
There is a significant amount both in credit card debt, buy-now-pay-later type offerings (e.g. Klarna), and payday loans - it's by no means all larger loans. It also varies greatly by country, but all major European countries still hav
963.
by vidarh 9mo ago
There are numerous credit providers in Europe that would beg to differ. By December 2025, consumer credit in the Euro area alone stood at an estimated €812 billion.
964.
by vidarh 9mo ago
> The blog post used the word autonomous a lot, which I suppose is true if Nicholas Carlini is not a human being but in fact a Claude agent. Nothing in the article suggests it did not autonomously do the work. > Why would anyone do th
965.
by vidarh 9mo ago
I occasionally - when I have tokens to spare, a MAX subscription only lasts so far - have Claude working on my Ruby compiler. Far harder language to AOT compile (or even parse correctly). And even 6 months ago it was astounding how well it&
966.
by vidarh 9mo ago
They made a blog post about it because it's an amazing test of the abilities of the models to deliver a working C-compiler, even with lots of bugs and serious caveats, for $20k of tokens, without a human babysitting it. I'd challe
967.
by vidarh 9mo ago
I regularly has it produce 10k+ lines of code that is working and passing extensive test suites. If you give it a prompt and no agent loop and test harness, then sure, you'll need to waste your time babysitting it.
968.
by vidarh 9mo ago
If you can show us code it has infringed on, you might have a point. Until then, you are making unfounded accusations.
969.
by vidarh 9mo ago
> the experimenter was trying to show a different point of what the Internet got (as usually) All of your points are important, but I think this is the most important one. Having written compilers, $20k in tokens to get to a foundation f
970.
by vidarh 9mo ago
It wasn't given gcc source code, and was not given internet access. It the extent it could translate gcc source code, it'd need to be able to recall all of the gcc source from its weights.
971.
by vidarh 9mo ago
> My 2 cents: just like Cursor's browser, it seems the AI attempted a really ambitious technical design, generally matching the bells and whistles of a true industrial strength compiler, with SSA optimization passes etc. Per the art
972.
by vidarh 9mo ago
Claude one-shot a basic x86 assembler + linker for me. Missing lots of instructions, yes, but that is a matter of filling in tables of data mechanically. Supporting linker scripts is marginally harder, but having manually written compilers
973.
by vidarh 9mo ago
47% total tax wedge is not 47% of gross wage anywhere except places with no payroll taxes / employer contributions.
974.
by vidarh 9mo ago
I mean we know it can be done in little space, given the many tiny C compilers. I think what is most interesting about this one is exactly the creative shortcuts. It's an interesting design space for e.g. bootstrapping to impose extra
975.
by vidarh 9mo ago
Then you might as well embed the headers, since in that case you can't update the compiler and headers separately anyway.
976.
by vidarh 9mo ago
So we're down to a missing or unclear description of a dependency in a README - note following the instructions worked for others -, from implications the compiler didn't work.
977.
by vidarh 9mo ago
Especially given they're not shipping anything. The GCC binaries can't find misplaced or not installed headers either.
978.
by vidarh 9mo ago
A compiler that can't magically know how to find headers that don't exist in the expected directory? Yes, that is the case for pretty much every compiler. I suppose you could build the headers into the binary, but nobody does
979.
by vidarh 9mo ago
And it doesn't for the compiler in question either. As long as the headers exist in the places it looks for them. No compiler magically knows where the headers are if you haven't placed them in the right location
980.
by vidarh 9mo ago
Noticed the part where all it requires is to actually have the headers in the right location?
981.
by vidarh 9mo ago
It's a fun comparison, but with the notable difference that that one can compile the Linux kernel and generate code for multiple different architectures, while this one can only compile a small proportion of valid C. It's a great
982.
by vidarh 9mo ago
Sure, but non of those things requires you to watch it work. They're all easy to pick up on when reviewing a finished change, which ideally should come after it's instructions have had it run linters, run sub agents that verify it
983.
by vidarh 9mo ago
Reviewing what it produces once it thinks it has met the acceptance criteria and the test suite passes is very different from wasting time babysitting every tiny change.
984.
by vidarh 9mo ago
I will personally never consider using an agent that can't be easily pushed toward working on its own for long periods (hours) at a time. It's a total waste of time for me to babysit the LLM.
985.
by vidarh 9mo ago
> Or tell you how to use the tools you already have, but sure whatever. But that's missing the point of the writing tools example: The models can unilaterally do this , and if prompted properly they will do this today . Writing an
986.
by vidarh 9mo ago
The reason for this is that the cost scales with the model and training cadence, not usage and so they will hope that they will be able to scale number of inference tokens sold both by increasing use and/or slowing the training cadence
987.
by vidarh 9mo ago
Well, you can have the less capable system ask the more capable system to give it a tool to solve the problems it doesn't know how to solve, instead of solving it. If the problem really needs the more advanced system, that tool might
988.
by vidarh 9mo ago
That's a totally valid reason. I also have infrastructure I operate because of personal comfort rather than because it's financially optimal.
989.
by vidarh 9mo ago
I agree, but a lot of companies risk exactly that by creating policies that people are likely to have reasons to want to bypass. E.g. Calendar sharing. It's a paintpoint if you often have irregular working hours and have to match up
990.
by vidarh 9mo ago
I have showdead on, and every comment the account has posted is "[superlative] post. [trivial observation]. Do you/Have you [superficial question]".

More ›