i use a an auto-layout tool, so having windows stacked on top of each other is super-common for me, and the fact that they all peak thru each other (like the screenshot in the blog) looking absolutely terrible drives me crazy
Agreed on horses for courses. Different people have different tolerances. And yea, all things being equal, faster is better, but they are almost never equal. If you don’t mind me asking, what does “too slow” mean for you in this context? Do you have a particularly complex setup? And what do you use now as an alternative and how has that impacted the update speed?
I wish I could remember the details -- I know I got annoyed with things being slow and when I got a new computer decided to go the no-homebrew route. I'm using nix, and it seems fine so far, but I also really don't understand it at all, which is a little concerning. :-)
I still remember the very first time I saw an emoji -- just an old-school dumb phone, and my friend sent me a message with an emoji, which the phone, amazingly, was able to display. I had no idea such a capability even existed, and wondered for a second if I was dreaming.
I don't know for a fact, but I'd bet a few digits of cold hard cash it's a SwiftUI rewrite that is to blame. (Any1 in the know want to chime in?)
And yeah, it's terrible. Apple doesn't make good apps anymore.
(This is part of why I think electron does so well -- it's not as good as a really good native app [e.g. Sublime Text], but it's way better than the sort of default whatever you'll get doing native. You get a lot of niceness that's built into the web stack.)
That description is really only fair for base models†. Something like Opus 4.6 has all kinds of other training on top of that which teach it behaviors beyond "predict most probable token," like problem-solving and being a good chatbot.
(†And even then is kind of overly-dismissive and underspecified. The "most probable word" is defined over some training data set. So imagine if you train on e.g. mathematicians solving problems... To do a good job at predicting [w/o overfitting] your model will have to in fact get good at thinking like a mathematician. In general "to be able to predict what is likely to happen next" is probably one pretty good definition of intelligence.)
I'd disagree, the other training on top doesn't alter the fundamental nature of the model that it's predicting the probabilities of the next token (and then there's a sampling step which can roughly be described as picking the most probable one).
It just changes the probability distribution that it is approximating.
To the extent that thinking is making a series of deductions from prior facts, it seems to me that thinking can be reduced to "pick the next most probable token from the correct probability distribution"...
The fundamental nature of the model is that it consumes tokens as input and produces token probabilities as output, but there's nothing inherently "predictive" about it -- that's just perspective hangover from the historical development of how LLMs were trained. It is, fundamentally, I think, a general-purpose thinking machine, operating over the inputs and outputs of tokens.
(With this perspective, I can feel my own brain subtly oferring up a panoply of possible responses in a similar way. I can even turn up the temperature on my own brain, making it more likely to decide to say the less-obvious words in response, by having a drink or two.)
(Similarly, mimicry is in humans too a very good learning technique to get started -- kids learning to speak are little parrots, artists just starting out will often copy existing works, etc. Before going on to develop further into their own style.)
Non-sequitor: "perspective hangover" might be my favorite phrase I've ever read. So much of what we deal with is trying to correct-the-record on how we used to think about things. But the inertia that old ideas or modes have is monumental to overcome. If you just came up with that, kudos.
We could argue about whether fine tuning is still about predicting a distribution or not, but really I feel like whether or not that word is accurate misses the point of why the description is useful.
I like the phrasing because it distinguishes it from other things the generative model might be doing including:
- Creating and then refining the whole response simultaneously, like diffusion models do.
- Having hidden state, where it first forms an "opinion" and then outputs it e.g. seq2seq models. Previously output output tokens are treated differently from input tokens at an architectural level.
- Having a hierarchical structure where you first decide what you're going to say, and then how you're going to say it, like wikipedia's hilarious description of how "sophisticated" natural language generation systems work (someone should really update this page): https://en.wikipedia.org/w/index.php?title=Natural_language_...
Welllll I'm not so sure that phrase is well-suited for your intended meaning, then. (Also, tangentially, I think could argue thinking models w/ the elided thought prelude satisfy "having hidden state where it first forms an opinion.")
Put a loop around an LLM and, it can be trivially made Turing complete, so it boils down to whether thinking requires exceeding the Turing computable, and we have no evidence to suggest that is even possible.
As typically deployed [1] LLMs are not turing complete. They're closer to linear bounded automaton, but because transformers have a strict maximum input size they're actually a subset of the weaker class of deterministic finite automaton. These aren't like python programs or something that can work on as much memory as you supply them, their architecture works on a fixed maximum amount of memory.
I'm not particularly convinced turing complete is the relevant property though. I'm rather convinced that I'm not turing complete either... my head is only so big after all.
[1] i.e. in a loop that appends output tokens to the input and has some form of sliding context window (perhaps with some inserted instructions to "compact" and then sliding the context window right to after those instructions once the LLM emits some special "done compacting" tokens).
[2] Common sampling procedures make them mildly non-deterministic, but I don't believe they do so in a way that changes the theoretical class of these machines from DFAs.
Context effectively provifes an IO port, and so all the loop needs to do is to simulate the tape head, and provide a single token of state.
You can not be convinced Turing complete is relevant all you want - we don't know of any more expansive category of computable functions, and so given that an LLM in the setup described is Turing complete no matter that they aren't typically deployed that way is irrelevant.
They trivially can be, and that is enough to make the shallow dismissal of pointing out they're "just" predicting the next token meaningless.
Turing Machines don't need access to the entire tape all at once, it's sufficient for it to see one cell at a time. You could certainly equip an LLM with a "read cell", "write cell", and "move left/right" tool and now you have a Turing machine. It doesn't need to keep any of its previous writes or reads in context. A sliding context window is more than capacious enough for this.
You're right of course, but at the point where you're saying "well we can make a turing machine with the LLM as the transition function by defining some tool calls for the LLM to interact with the tape" it feels like a stretch to call the LLM itself turing complete.
Also people definitely talk about them as "thinking" in contexts where they haven't put a harness capable of this around them. And in the common contexts where people do put harness theoretically capable of this around the LLM (e.g. giving the LLM access to bash), the LLM basically never uses that theoretical capability as the extra memory it would need to actually emulate a turing machine.
And meanwhile I can use external memory myself in a similar way (e.g. writing things down), but I think I'm perfectly capable of thinking without doing so.
So I persist in my stance that turing complete is not the relevant property, and isn't really there.
Yeah, humans and LLMs and a TM transition function are all Turing complete in the same way, but it's also basically a useless fact. You could possibly train a sufficiently motivated rat to compute a TM transition function.
That's why I specifically didn't call the LLM itself Turing complete, but stated that if you put a loop around a Turing machine you can trivially make it Turing complete. Maybe I should have been clearer and write "the combined system" instead of it.
But the point is that this is irrelevant, because it is proof that unlesss human brains exceed the Turing computable, LLM's can at least theoretically be made to think. And that makes pushing the "they're just predicting the next token" argument anti-intellectual nonsense.
I am not sure it is proof, at least not in an interesting way. It's also proof that Magic: The Gathering could theoretically be made to think. Which is true but doesn't tell you anything much about MtG other than that it is a slightly complicated ruleset that has a couple of properties that are pretty common.
I think both sides of this end up proving "too much" in their respective directions.
> whether thinking requires exceeding the Turing computable
I've never seen any evidence that thinking requires such a thing.
And honestly I think theoretical computational classes are irrelevant to analysing what AI can or cannot do. Physical computers are only equivalent to finite state machines (ignoring the internet).
But the truth is that if something is equivalent to a finite state machine, with an absurd number of states, it doesn't really matter.
Hence why I finished the sentence "and we have no evidence to suggest that is even possible".
I think it's exceedingly improbable that we're any more than very advanced automatons, but I like to keep the door ajar and point out that the burden is on those claiming this to present even a single example of a function we can compute that is outside the Turing computable if they want to open that door..
> Physical computers are only equivalent to finite state machines (ignoring the internet)
Physical computers are equivalent to Turing machines without the tape as long as they have access to IO.
I think it's pretty likely that "intelligence" is emergent behavior that comes when you predict what comes next in physical reality well enough, at varying timescales. Your brain has to build all sorts of world model abstractions to do that over any significant timescale. Big LLMs have to build internal world models, too, to do well at their task.
> The problems aren't bugs; they're consequences of design decisions that may have made sense a decade ago, but don't align with how JavaScript developers write code today.
> I'm not here to disparage the work that came before — I'm here to start a conversation about what can potentially come next.
Terrible LLM-slop style. Is Mr Snell letting an LLM write the article for him or has he just appropriated the style?
Heh, I was using emdashes and tricolons long before LLMs appropriated the style but I did let the agent handle some of the details on this. Honestly, it really is just easier sometimes... Especially for blogs posts like this when I've also got a book I'm writing, code to maintain etc. Use tools available to make life easier.
I'm not sure any emdash use at all is what people are calling out typically(maybe it is?), more the sheer number of them typical in LLM written stuff.
Just ctrl-f'ing through previous public posts, I think there were a total of 7 used across about that many posts. This one for example had 57. I'm not good enough in proper English to know what the normal number is supposed to be, just pointing that out.
Just want to raise my hand and say I too have been using em dashes for considerably longer than LLM has been on every hacker's lips. It's obviously not great being accused of being an AI just because one has a particular style of writing...
People are understandably a bit sensitized and sceptical after the last AI generated blog post (and code slop!) by Cloudflare blew up. Personally I'm fine with using AI to help write stuff as long as everything is proof-read and actually represents the authors thoughts. I would have opted to be a bit more careful and not use AI for a few blog posts after the last incident though if I was working at Cloudflare...
What was it specifically about the style that stood out as incongruous, or that hindered comprehension? What was it that made you stumble and start paying close attention to the style rather than to the message? I am looking at the two examples, and I can't see anything wrong with them, especially in the context of the article. They both employ the same rhetorical technique of antithesis, a juxtaposition of contrasting ideas. Surely people wrote like this before? Surely no-one complained?
The problem is less with the style itself and more that it's strongly associated with low-effort content which is going to waste the readers time. It would be nice to be able to give everything the benefit of the doubt, but humans have finite time and LLMs have infinite capacity for producing trite or inaccurate drivel, so readers end up reflexively using LLM tells as a litmus test for (lack of) quality in order to cut through the noise.
You might say well, it's on the Cloudflare blog so it must have some merit, but after the Matrix incident...
I find it more amusing that the benchmarks claim 530 GB/s throughput on an M1 Pro which has a 200GB/s memory bandwidth. The 275 GB/s figure for chained transforms has the same problem.
I suspect the benchmarks, if not most of this project, was completely vibecoded. There are a number of code smells, including links to deleted files, such as https://github.com/jasnell/new-streams/blob/ddc8f8d8dda31b4b... an inexistent REFACTOR-TODO.md
These AI signals will die out soon. The models are overusing actual human writing patterns, the humans are noticing and changing how they write, the models are updated, new patterns emerge, etc, etc. The best signal for the quality of writing will always be the source, even if they are "just" prompting the model. I think we can let one incident slide, but they are on notice.
The idea is well articulated and comes across clear. What’s the issue? Taking a magnifying glass to the whole article to find sentence structure you think is “LLM-slop” is an odd way to dismiss the article entirely.
I’ve read my fair share of LLM slop. This doesn’t qualify.
Partially true. They've been trained and then aligned towards a preferred style. They don't use em-dashes because they are over-represented in the training material (majority of people don't use them).
It seems likely that with the written word, as with most things, a minority of people produce the majority of content. Most people publish relatively few words compared to professional writers.
Possibly the LLM vendors could bias the models more toward nonprofessional content, but then the quality and utility of the output would suffer. Skip the scientific articles and books, focus on rando internet comments, and you’ll end up with a lot more crap than you already get.
reply