p-e-w

14 ms·

241.
by p-e-w 5mo ago
Because LLMs can learn that different token sequences represent the same character sequence from training context. Just like they learn much more complex patterns from context. You can try this out locally with any mid-sized current-gen LLM
242.
by p-e-w 5mo ago
That’s false. Larger LLMs learn token decompositions through their training, and in fact modern training pipelines are designed to occasionally produce uncommon tokenizations (including splitting words into individual characters) for this r
243.
by p-e-w 5mo ago
That paper’s abstract doesn’t carry its title, to put it mildly.
244.
by p-e-w 5mo ago
“Seems clear” based on what?
245.
by p-e-w 5mo ago
> due to fundamental limitations People keep throwing this phrase around in relation to LLMs, when not a single “fundamental limitation” has been rigorously demonstrated to exist, and many tasks that were claimed to be impossible for LLM
246.
by p-e-w 6mo ago
I appreciate you acknowledging that this was a mistake, but as you surely know from your own experience with other people’s mistakes, some mistakes are so egregious that they cast doubt on the intentions of the people involved even if they
247.
by p-e-w 6mo ago
Because the human world is built for bipedal beings and everything else will encounter obstacles somewhere.
248.
by p-e-w 6mo ago
> We'll know this works when it starts replacing Amazon pickers in quantity. That doesn’t follow. There are plenty of tasks that can be fully and reliably automated but aren’t, for the simple reason that human labor is dirt cheap co
249.
by p-e-w 6mo ago
Then the correct answer is “I can’t tell.” Not “Here’s a random guess that I just pulled out of my ass.” LLMs have picked up the bad habit of trying to give an answer when no answer can be given from scientists, who overall don’t say “I don
250.
by p-e-w 6mo ago
Serious question: What exactly do they love America for? I just don’t get it. Seems like in every way that matters to the common people, the US is at best mediocre. Could it be that they secretly subscribe to a different version of the sa
251.
by p-e-w 6mo ago
Stuff like this is far above the capabilities of today’s top AIs. It’ll produce something , sure, but it won’t actually work, and making it work takes as much effort as building it from scratch.
252.
by p-e-w 6mo ago
Because U.S. English is everything, and people want it that way, and any concern about cultural imperialism that would immediately be brought up in any other context magically doesn’t apply when Americans could possibly be offended. Everyon
253.
by p-e-w 6mo ago
Firefox has many weaknesses, but I never once thought “man, that thing is slow”. It isn’t, and chasing benchmark numbers is a waste of effort. A better security model or deeper customizability would be far more valuable.
254.
by p-e-w 6mo ago
At some point, a derivative idea becomes so different from the original one that it’s a novel idea in essence. Just like SMS is ultimately a derivative of cuneiform tablets, and yet it isn’t in any meaningful sense.
255.
by p-e-w 6mo ago
The claim that you need to disable JavaScript to be secure is bullshit anyway, otherwise disabling JavaScript would be standard practice in any security-critical environment such as high-level government offices, which it most certainly isn
256.
by p-e-w 6mo ago
You’re asking why people are being “dramatic” about an automated system that can do what highly specialized experts get paid hundreds of thousands of dollars to do? It’s just fascinating to see how AI’s accomplishments are being systematica
257.
by p-e-w 6mo ago
It made me realize that “important events” are meaningless empathy theater, with the goal of desensitizing people until they care more about random folks dying on the other side of the planet than about the person sitting next to them on th
258.
by p-e-w 6mo ago
Sorrow for people you don’t know? That’s a category error, and soundbites like that are a big part of the reason why many people now find it difficult to feel genuine sorrow.
259.
by p-e-w 6mo ago
About 15 years ago, I saw on the news in Europe that 8 students had been shot at some random high school in a random US state. Having been conditioned by my environment to perceive such events as important, I turned to my friend and said “M
260.
by p-e-w 6mo ago
Is that true? If so, that’s not a good sign. I remember how impressed I was by ZoneAlarm in the early 2000s asking permission for itself to connect to the Internet, using the exact same dialogue it presented for any other program, with no
261.
by p-e-w 6mo ago
It’s unfortunate that brutalism has become synonymous with “crumbling concrete”. That was certainly not the intention of the brutalist architects, but rather a side effect of the poor quality of the (sometimes experimental) concrete mixture
262.
by p-e-w 6mo ago
People only notice now because the “right” kind of people are suddenly affected. Just like the invasion of Ukraine became the most important topic globally for years, and made everyone virtue signal about how important sovereignty supposedl
263.
by p-e-w 7mo ago
And this bug was reported eight years ago , with no serious attempt to fix it since.
264.
by p-e-w 7mo ago
I was shocked when IBM acquired Red Hat a few years ago. I had silently assumed at the time that Red Hat was far bigger than IBM nowadays, so the reverse would have made more sense to me.
265.
by p-e-w 7mo ago
Why would they go to the moon? They’re far too busy doing things that actually matter, such as slashing launch costs by 80% or more, while achieving the highest reliability of any launch system ever.
266.
by p-e-w 7mo ago
Strange that SpaceX doesn’t seem to be suffering from that limitation. Could it be that the real problem is pork barrel spending and government wastefulness?
267.
by p-e-w 7mo ago
Almost all of what makes spaceflight “cool” today is inherited excitement and nostalgia, most of it unearned by the current generation of space endeavors. Apollo was a humanity-defining undertaking. Repeating the same 60 years later with ou
268.
by p-e-w 7mo ago
> A 500k line codebase for an agent CLI proves one thing: making a probabilistic LLM behave deterministically is a massive state-management nightmare. Considering what the entire system ends up being capable of, 500k lines is about 0.001
269.
by p-e-w 7mo ago
> Across these variations, the overall result stays quite consistent: under certain conditions, ordinary people can be led to do harmful things. The pop culture version of what happened in those experiments is “regular people will admini
270.
by p-e-w 7mo ago
Many cloud products now continuously send themselves the input you type while you are typing it, to squeeze the maximum possible amount of data from your interactions. I don’t know whether ChatGPT is one of those products, but if it is, t

More ›