sdenton4

9 ms·

151.
by sdenton4 9mo ago
People talking to each other in person tend to modulate their voices to match the context. People talking on speakerphone tend to crank the volume and shout.
152.
by sdenton4 9mo ago
That time he showed up on metafilter with a sock puppet and talked about himself in the third person. https://www.metafilter.com/102472/How-to-Get-a-Real-Educatio... "As far as Adams' ego goes, maybe you don&
153.
by sdenton4 9mo ago
And that becomes obvious when you go to look for it. These work best in situations where false positives have low cost/impact, but true positives are easily verifiable and have high impact. in other words, problems in epistemological p
154.
by sdenton4 9mo ago
If taking heroic doses of ketamine on a regular basis is wrong, I don't want to be right.
155.
by sdenton4 9mo ago
Adams was insanely narcissistic. Inventing reverse racism to avoid confronting his own short comings is completely on brand.
156.
by sdenton4 9mo ago
Curiously, the game was mass produced before it was a game at all... As boxes of 1000 blank white flash cards.
157.
by sdenton4 9mo ago
Yeah, this is by far the biggest value I've gotten from LLMs - just pointing me to the area of literature neither me nor any of my friends have heard of, but which have spent a decade running about the problems we're running into.
158.
by sdenton4 9mo ago
My quixotic bicycle grocery buying system still wins, though.
159.
by sdenton4 9mo ago
The quadratic attention problem seems to be largely solved by practical algorithmic improvements. (Iterations on flash attention, etc.) What's practically limiting context size IME is that results seem to get "muddy" and get
160.
by sdenton4 9mo ago
Pushing model builders to use smarter scrapers is a net good. Endless rescrapes of static content is driving up bandwidth bills for housing simple things.
161.
by sdenton4 9mo ago
I like the high level idea! (how do we test intelligence in a non functional way?) I'm effect, the different response types are measuring how the models respond to a context-free novel environment. I imagine humans would also respond o
162.
by sdenton4 9mo ago
Only rich people can afford houses in the US, ergo.
163.
by sdenton4 9mo ago
"Hey, hey, are you still asleep? Using spare cycles, I have designed an optimal recipe for mashed potatoes, as you mentioned ten days ago. I need you to go get some potatoes."
164.
by sdenton4 9mo ago
My original home on the internet is metafilter, where I've been a member since 2001. For an extremely long time, it was the internet's best kept community, imo. Unfortunately, it also seems to be falling into pure doomerism, espec
165.
by sdenton4 9mo ago
There are times when mathematics is a bit like a full set of wrenches trying to take apart a problem held together by screws.
166.
by sdenton4 10mo ago
This is analagous to the problem of premature optimization - if you try to optimize performance without benchmarking, you end up eating a lot of time and effort on things that don't matter. Likewise for product: it is very easy to solv
167.
by sdenton4 10mo ago
I'm in historicals! And I mess around with 3d printers! The big difference in the historical wargame space is that there's no way to trademark Alexander the Great. Anyone can make the minis, whether it's a small-but-awesome c
168.
by sdenton4 10mo ago
When you're not shipping, you're not learning from users. As a result, it's easy to build working, correct, performant code which doesn't fit what anyone actually needs.
169.
by sdenton4 10mo ago
On the other hand, another week another JavaScript framework, amirite? There continues to be new stuff to ask questions about, but stack overflow failed to be the default location for new stuff. I guess now there's more discussion dire
170.
by sdenton4 10mo ago
There's a things called 'hidden hearing loss' in which the ability to pause midband sounds specifically in complex/noisy situations degrades. This is missed by standard tests, which only look for ability to hear a given
171.
by sdenton4 10mo ago
You're not wrong, Walter. https://www.youtube.com/watch?v=C6BYzLIqKB8
172.
by sdenton4 10mo ago
I have very little faith that scotus will have any consistency in their decisions going forward - they seem to be nakedly political, and backing trump. If the elections swing the other direction (despite their aid in gerrymandering), expect
173.
by sdenton4 10mo ago
Or maybe it's expensive because it doesn't scale. The per-unit cost of nuclear power plants is extremely high, making it hard to get economies of scale in building more of them. And if we /do/ hit economies of scale, ura
174.
by sdenton4 10mo ago
Not all transformers are LLMs.
175.
by sdenton4 10mo ago
If only there were some system that could start from some sparse and noisy observations and weave together a plausible completion...
176.
by sdenton4 10mo ago
For anyone actually interested in the question of measuring animal intelligence, I recommend the book 'Are we smart enough to know how smart animals are?' by frans de waal. https://www.goodreads.com/book/show&
177.
by sdenton4 10mo ago
Meh. Information is often jumbled and wrong in the immediate aftermath of a newsworthy event, and it is tempting to accept tenuous claims which reinforce one's biases. Take the murder of Bob Lee, in which early reports were a bit off a
178.
by sdenton4 10mo ago
Humans are /extremely/ visual compared to other animals: this tends to make us underestimate the intelligence of other animals (when we use visual intelligence as a proxy for general intelligence) and miss out on smart uses of oth
179.
by sdenton4 10mo ago
Shouldn't the llm that all this scraping is powering be able to help figure out which websites have an API and how to use it?
180.
by sdenton4 10mo ago
If only there were some convenient technology that could help us sort out these many small cases automatically...

More ›