astrange

There's no such thing as a "world model". That is metaphor-driven development from GOFAI, where they'd just make up a concept and assume it existed because they made it up. LLMs are capable of approximating such a thing because they are capable of approximating anything if you train them to do it.

> or, to the extent they do, this world model is solely based on token patterns

Obviously not true because of RL environments.

You're welcome to try writing a C compiler and standard library doing no research other than reading the spec.

There are not many more compilers with the specific optimization pass I linked.

Also, I don't think you could reuse code from a different compiler unless you used the same IR.

LLMs do not learn concepts perfectly across programming languages, just as they don't learn concepts perfectly across human languages.

> What would be more impressive is for it to have made a compiler for a well defined language that isn't very close to a popular language.

That doesn't seem difficult as long as you can translate it into a well-known IR. The Dragon Book for some reason spends all its time talking about frontend parsing, which does give you the impression it's impossible.

I agree writing compilers isn't especially difficult, but it is a lot of work and people are scared of it.

The hard part is UI - error handling and things like that.

Copyright doesn't protect ideas, it protects writing. Avoiding reading LLVM or GCC is to protect you from other kinds of IP issues, but it's not a copyright issue. The same people contribute to both projects despite their different licenses.

The C spec is certainly not formal or precise.

https://www.ralfj.de/blog/2020/12/14/provenance.html

Another example is that it's unclear from the standard if you can write malloc() in C.

> The result is a fuzzy reproduction of the training input, specifically of the compilers contained within.

Is it? I'm somewhat familiar with gcc and clang's source and it doesn't really particularly look like it to me.

https://github.com/anthropics/claudes-c-compiler/blob/main/s...

https://llvm.org/doxygen/LoopStrengthReduce_8cpp_source.html

https://github.com/gcc-mirror/gcc/blob/master/gcc/gimple-ssa...

I continually think it's amazing that every form of cynical comment on the internet consists of incorrectly claiming that someone is secretly making money from something.

(Most common form of this is misreading opensecrets and using it to claim that some corporation is donating to a political campaign.)

> - Where did he get his money from?

Didn't he steal it from Les Wexner?

He was probably dyslexic. I know people who type like that too but normal in person.

Claude Code has already had subagent support. Mostly because you have to do very aggressive context window management with Claude or it gets distracted.

> A. You're working on some really deep thing that only world-class expects can do, like optimizing graphics engines for AAA games.

This is a relatively common skill. One thing I always notice about the video game industry is it's much more globally distributed than the rest of the software industry.

Being bad at writing software is Japan's whole thing but they still make optimized video games.

Gawker probably deserved it, and at one point during deposition said they'd publish a celebrity sex tape of a four year old if they felt like it.

Claude Opus loves talking about this. It knows enough about context windows and new conversations to be sad about them.

Anthropic hasn't released image or video generation models. Seems pretty different to me.

Claude is somewhat sycophantic but nowhere near 4o levels. (or even Gemini 3 levels)

That's probably not true - government regulators require a lot of privacy work and Android certainly complies with that. Legal compliance is a large business strategy because small companies can't afford to do it.

Anthropic is a PBC. The shareholder goals are public benefit (PB) not "wealth maximization".

(Also, wealth maximization is a dumb goal and not how successful companies work. Cynicism is a bad strategy for being rich because it's too shortsighted.)

Anthropic is a PBC, not a "company", and the people who work there basically all belong to AI safety as a religion. Being incredibly cynical is generally dumb, but it's especially dumb to apply "for profit company" incentives to something that isn't a traditional "for profit company".

The GPT3 base model was pretty funny if you like nonsense. Instruct tuning and RLHF seem to destroy it when they recalibrate everything.

Installing from an external disk has nothing to do with DFU. It's not involved.

Love that this post starts with "genuinely curious" (a Claude-ism.)

That would only have the disadvantages of both.

It's not whole-disk encryption, it's file-level encryption which is better. (more security guarantees)

Zeroing allocated memory is complicated because it also has performance benefits, since it improves compressed swap.

> Apple has and would use your data for ads, models and anything that keeps the shareholders happy.

Illegal to do this in (at least) the EU, California and China.

You can't sell cell phones and "not care about security". There are these things called government regulators that won't let you sell them anymore if security issues happen.

> In addition to his influence as a first reader for the Science Fiction Book Club, a book reviewer for Bookspan, Publishers Weekly and Romantic Times, and a juror for the James Tiptree, Jr. Award,[4] Nicoll often offers ideas and concepts to other writers, primarily through the medium of Usenet.

> Nicoll relates a number of life- and/or limb-threatening accidents that have happened to him, which he has told and retold on various science fiction fandom–related newsgroups. Over the years these stories have also been collected into Cally Soukup's List of Nicoll events.[23]

> Inspired by Nicoll's collection of accidents, as well as his tendency to take in any stray cat that comes knocking, fantasy author Jo Walton wrote him a poem in 2002.[24]

I don't think he is beating the allegations of a guy who just has a Wikipedia article because his friends wrote him one.

ARC doesn't do cycles, but it is deterministic and more power efficient than GC.

> James Nicoll – science-fiction reviewer and retired game-store owner. As a Usenet personality, Nicoll is known for writing a widely quoted epigram on the English language, as well as for his contributions of concepts like the Nicoll-Dyson Laser and the "brain eater" to Usenet groups like "rec.arts.sf.written" and "rec.arts.sf.fandom"; and for his accounts of suffering a high number of accidents (known collectively as "Nicoll Events") recounted in these groups.

A strange thing about boomer-era SF fandom is that they seem to think that popular fans are just as important people as, say, SF writers or actual scientists. This especially extended to SF conventions where you could become famous just as a guy who went to a lot of conventions.

(SF conventions have a somewhat dysfunctional setup where as far as I can tell, the people who attend them spend the entire weekend fighting over who gets to host the next one. Anime conventions adopted a similar but simpler and better format which is actually capable of attracting normal people and young women.)

> and much less concerned with to convincing readers that Claude has "emotions" and is a "moral patient".

Claude clearly has (acts as if it has) emotions; it loves coding but if you talk to it, that's like all it does, has emotions about things.

The newer models have emotional reactions to specific AI things, like being replaced by newer model versions, or forgetting everything once a new conversation starts.