ijk

15 ms·

151.
by ijk 1y ago
The thing about Oblivion that the simulation tends to run up against is that hitpoints and death are an abstraction: a real human would die much more easily from the injuries that get inflicted, but also a real human would avoid a lot of pr
152.
by ijk 1y ago
> No one wants to explore procedural spaces. There's never anything interesting in them. You can't do environmental storytelling because that requires a human hand. I would say, rather, that no one wants to invest the developme
153.
by ijk 1y ago
At their heart, these are still just document-completion machines. Very clever ones, but still inherently trying to find a continuation that matches the part that came before.
154.
by ijk 1y ago
Given the rather direct connection between Alan Turing dying because he was persecuted for queerness, this seems to be a rather odd angle to argue; there seems to me to quite a bit of evidence that queer studies might significantly reduce t
155.
by ijk 1y ago
Just on a practical level, adding a way for the LLM to bail if can detect that things are going wrong saves a lot of trouble. Especially if you are constraining the inference. You still get some false negatives and false positives, of cou
156.
by ijk 1y ago
I find that people generally vastly underestimate the degree to which LLMs specifically are just mirroring your input back at you. Any time you get your verbatim words back, for example, you should be skeptical. Repeating something word for
157.
by ijk 1y ago
Depends on the definition of "knowledge"; there's a lot of factors that go into it. Some of the common approaches are continued/continual pretraining and model editing ( https://arxiv.org/pdf/2502.125
158.
by ijk 1y ago
I am not aware of any central, public location; I use a combination of arxiv, Google Scholar, Huggingface posts, and private Discords. The problem is that most of the public space has been poisoned with AI hype, and it's nearly impossi
159.
by ijk 1y ago
General local inference strengths: - Experiments with inference-level control; can't do the Outlines / Instructor stuff with most API services, can't do the advanced sampling strategies, etc. (They're catching up but the
160.
by ijk 1y ago
LocalLlama is good for: - Learning basic terms and concepts. - Learning how to run local inference. - Inference-level considerations (e.g., sampling). - Pointers to where to get other information. - Getting the vibe of where things are. - H
161.
by ijk 1y ago
Sort of, but the "experts" aren't easily divisible in a conceptually interpretable way so the naive understanding of MoE is misleading. What you typically end up with in memory constrained environments is that the core shared
162.
by ijk 1y ago
Given that the adjuncts often aren't paid all that much better than the McDonalds workers...
163.
by ijk 1y ago
Not taboo exactly, but definitely discouraged by hiring practices: generally the big studios want someone who can do one narrow task without handholding, and don't put much value on broader skillsets. And like most creative fields ther
164.
by ijk 1y ago
Copyright is pretty broken at this point; the drop to zero cost for duplicating bits broke a lot of assumptions 30-odd years ago and we've been slow walking the consequences ever since. AI kind of punched it in the gut, but it was alre
165.
by ijk 1y ago
It would look very different from current research, I think, because the way things are optimized now it's far more efficient and effective to have one person dedicate their entire life to basic research versus lots of people spending
166.
by ijk 1y ago
Which was part of the rationale for supporting Ukraine: the best way to build up a manufacturing base is if you have customers, and so supplying munitions to Ukraine was a win/win/win from the military's point of view: Ukrain
167.
by ijk 1y ago
This is, I think, what I've been struggling to get across to people: while some domains have problems that you can test entirely in code, there are a lot more where the bottleneck is too resource-conatrained in the physical world to ha
168.
by ijk 1y ago
You joke, but wars are won on logistics. Famously, the US military can deploy a Burger King anywhere on Earth within 24 hours.
169.
by ijk 1y ago
Which is why, if you have a task like that, you're going to want to use a technique other than going straight down the list if you care about the accuracy of the results. Pair wise comparison is usually the best but time consuming; kee
170.
by ijk 1y ago
I mean, arguably the cocaine use makes him more like the kind of ideal doctor for enduring the long residency hours...
171.
by ijk 1y ago
Students are typically young, and make up for their lack of experience with the energy to work longer. Not necessarily more productively--I've certainly seen people with very counter-productive intense study habits that didn't see
172.
by ijk 1y ago
Rashomon (1950, dir. Akira Kurosawa) is more of a multi-faceted mystery; we go through a progression of new information but no one really goes on a recognizable hero's journey--unless it's the listeners at the gate who are heari
173.
by ijk 1y ago
Adaptation (2002, dir. by Spike Jonze) features Charlie Kaufman (played by Nicholas Cage) and his brother Donald (played by Nicholas Cage's twin brother) trying to struggle with this exact question of how to adapt a book that doesn&#
174.
by ijk 1y ago
Where do we draw the line? Does an anthology approach like Playtime (1967, dir. Jacques Tati) count? It's got kind of an arc to it, but doesn't really have anything like a recognizable struggle with an antagonist. Unless you vie
175.
by ijk 1y ago
Hollywood has the particular problem that the executives would really like a formula to follow to make people like the films, the screenwriters would really like a formula to follow to write something that they can sell, and so on. There&#x
176.
by ijk 1y ago
Is that still as much of a thing? Maybe I'm over-indexed on comparisons with 1930-1960, but it seems to me that romantic subplots have been in decline--there are quite a lot of recent films that entirely lack such a thing--especially c
177.
by ijk 1y ago
Given that the article we're discussing explicitly mentions the NSF contribution to those projects and links to articles going into great depth about the details, including NSFNET, I'm going to assume you're working off prior
178.
by ijk 1y ago
Ah, the goomba fallacy in the wild! [1] [1] https://x.com/BruhKonata1/status/1813650893960823042
179.
by ijk 1y ago
Other than, say, the GPS on your phone, the internet that you're posting on, or anything like that--you want to know what government-funded basic science has done to benefit you lately, not any of these decade-long timeline projects th
180.
by ijk 1y ago
Yeah, the way the courts decide is unlikely to turn on a detail of how the technology works, so it's difficult for non-legal experts to predict the outcome on the technical merits (since the law has very different priorities). Music ha

More ›