dwallin

The data doesn't seem to add much support for this, outside of a blip because of the pandemic.

https://www.newyorkfed.org/research/college-labor-market#--:...

It’s lenticular but with 100 different possible angles, so there’s no eye tracking needed and it works with multiple viewers. The tradeoff is it seems you need to pump a lot of data in for all those views, and you probably need a pretty high resolution and brightness screen. There’s a good description in their docs:

https://docs.lookingglassfactory.com/keyconcepts/how-it-work...

I don't think it's entirely an architecture problem, there's a huge training set problem. Almost all the content you would train on assumes outside knowledge/facts, so the llm learns to store these in the model in order to be able to maximize it's completion ability. However, a lot of this assumed knowledge/generalizations are actually unhelpful/harmful for these sorts of cases.

If you wanted to create an llm that significantly improved on this you would probably need to make sure to massively clean up / reorganize your training data so that you always provide sufficient context and the llm is disincentived from baking in "facts". But I'm not sure this is tractable to do currently at the scale of data needed.

Making an excellent game isn't even enough these days, that's just table stakes. With so many games out there it's a market of lemons. You need to have a way of rising above the noise. For AAA, it's having budgets to make games on a bigger scale than everyone else. For mobile games, it's generally about leveraging a large war chest to fuel an advertising-driven margin game. For prestige/indie games you need something that people will want to talk/write about, or build a dedicated fanbase over time. For tiny games like flappy bird / wordle you need massive built-in virality and lots of luck.

That article about the golden ratio lost me with that horribly cropped image. It ignores all the things that actually matter when composing and cropping a shot, and gives some absurd advice that will actively lead people astray. The end result cuts off valuable content in favor of blurry background nonsense, leaving a random pencil end sticking up out of nowhere and a massively distracting board that is uncomfortably lined up with an outer edge. Learning some basic compositional techniques would be far more helpful both for the author and its readers.

They closed the ticket 12 minutes after that comment. It seems unlikely a coincidence. If someone wanted a reason to shut this down that comment gave them all the ammunition they needed. Don't forget, Google is not a singular entity and is made up of individuals with differing opinions, some of which are pro and some against jpegxl, but like any human social group, if you cause them to feel that one of their own is being attacked they are almost certain to close ranks.

To me (As an outside observer) the reason for the status change seemed pretty clear. There was no activity on the issue for a month and a half, and then someone made a new comment that personally attacked a google employee by name as a bad faith actor. Of course google is going to shut that down real quick.

I think the failures of iPadOS are really illuminating about the contrasting strengths of closed top-down design vs open stochastic exploration. Apple loves to craft an experience, and this design approach is great at taking large steps to get out of local maxima, and find new rich areas. Because they control the entire process they are theoretically able to make bigger moves and take larger design risk. For example, they could have never made the iPhone without this method of development. However any singular design vision is ALWAYS going to have blindspots that prevent it from reaching a better local optimal state. When it comes to figuring out these blindspots and moving towards a better local maxima, the best approach is having a lot of individuals experimenting widely.

For example, if outside developers had more capability to explore different ways of managing files on an iPad, it's likely that someone would stumble upon a really great way of handling it that gains traction, and that Apple could adopt more widely.

Yeah, I think this is a real possibility. Even without full-blown therapy, just building up people's confidence can end up helping with setting healthy habits and creating new opportunities for real life connection.

Yeah, this annoyed me about the paper. Clearly there are significant visual issues with patterns forming, but they just hand wave it away and basically say "according to all these common quality metrics, this method is superior". These patterns make it a non-starter for a ton of the use cases for blue noise, the authors should have been far more humble in making their case and addressing counterpoints.

The WHOLE point of the ad in the first place was to make people feel strong positive emotions toward Apple products. It turns out they misjudged and for many people it didn’t evoke the type of emotions they thought it would. It’s not like people are up in arms about a spec sheet.

I think you are being extremely irrational in expecting people to not feel passionately about random things. Companies spend insane amounts of money influencing consumer sentiment for good reasons.

Why get so bothered by other people being upset? Apple is going to be fine, you don’t need to worry on their behalf. No need to get your undergarments of choice in a twist. Good opportunity to step outside and get some fresh air.

If you trained an image generator, removing all instances of zebras from the training set, you could ask it to output images of a black and white striped horse and it would likely succeed. Then you could fine tune an image recognition model (also with all zebras removed from the training set) on the generated image set to associate it with the word zebra. If you then showed it a bunch of images of actual zebras there’s a really good chance it would succeed.

Placement within all steam games is not really a useful metric, there is an extremely long tail of games on steam that make basically nothing.

Your question doesn’t make sense if you read it directly, why are you asking which door to pick if you already know the door? It’s what you call a “trick” question, something humans are also bad at. It’s equally plausible (and arguably more useful for general purposes) for the model to assume that you mistyped and for eg. forgot to type “don’t” between I and know.

Ironically, by linking to it you’ve probably vastly boosted it’s rank on Google.

This is not two citizens having a private conversation. These are actions and conversations taken on behalf of a corporation. Corporations are given tons of privileges, but in exchange they have additional responsibilities that individuals do not have. If employees cause the company to violate those responsibilities the employees are generally not held directly responsible (unless they were directly violating a law). It’s the company itself who is held liable for the actions of its employees in this case.

Maybe it would be more accurate to say that the structure of a public corporation is inherently (and imo, intentionally) amoral? A defining characteristic of modern capitalism is the widespread laundering of responsibility.

Yeah, I think this link and title would be a better entry point into the topic.

I think you would be surprised about how much of a performance hit that would be over existing state of the art. The thing you are missing is that the human visual system evaluates existence of spaces in parallel, a single threaded algorithm would need to check every character to see if it's a space, in addition to checking the letters of the search string. Also that fails to generalize to languages without spaces, search strings with spaces, etc.

Also if you look at algorithms like Boyer-Moore, they effectively DO skip spaces, but do so in a manner that is language / content agnostic. (https://www-igm.univ-mlv.fr/~lecroq/string/node14.html)

Yeah this is the wrinkle. I do wonder though, whether there might be a situation reversal effect if you moved far enough in the other direction. If you instead had handoffs extremely frequently, say for example, every 6 hours, or perhaps a staggered system where there are people shifting off every 3 hours, you could/would build better process, habits, and culture potentially even to the point where it's no longer a significant liability, or even a benefit (fresh eyes).

It's somewhat similar to how having a handful of remote individuals in a mostly in-person company can lead to significant communication issues, but a fully/mostly remote company builds and leverages tools and processes to reduce these issues significantly.

I'd be curious to see how many studies controlled for relative emotional value of money, eg. by looking for differences in behavior for all groups at various values.

One could hypothesize that people might be more willing to share an unexpected "windfall", the perception of which could feasibly scale with one's emotional perception of wealth.

Of course we have higher Disposable Income in the US, it's a useless metric for these sorts of comparison purposes as it includes expenditures for things like healthcare, which in most other places is paid for by taxes (aka. not part of disposable income). A more useful metric would be Discretionary Income.

I really like this. I could see constructing a physical artifact for a specific case of this game.

According to another comment, the best puzzle with a high number of 6 is [1,6,3] with a minimum of 14 moves.

This would mean you have 6 total rods, and two gutters. The puzzle gutter with a length of 10, and the storage gutter with a length of 11.

If you want more visual symmetry a high number of 7 allows you potentially 2 gutters of length 14.

You can make a physical version by having one fixed height pole for each integer up to the max, and then an equal number of post holes. Eg. If you wanted to support a game up to the integer 10 you would need 10 posts with lengths from 1 to 10 units, and 10 post holes.

The biggest issue with this game is that there's no guarantee that any arbitrary starting state has a valid solution. A much needed improved would be a simple rule for guaranteed reversible starting formations (if one even exists).

What is with people writing purposely flawed examples?

If you want to convince a general audience that your approach is worth the additional cognitive overhead then compare it with code that is equivalently robust, not some low effort straw-man example.

I'll keep an eye out for that! The variations feature in Suno is a good example of what not to do here, as it effectively just makes another random iteration using existing settings.

I think the other missing pieces I've found are upscaling and stem splitting. While existing tool exist for splitting stems exist, my testing found that this didn't work well in practice (at least on Suno music), likely due to a combination of encoder-specific artifacts and the overall low sound quality. Existing upscaling approaches also faced similar issues.

My naive guess is that these are things that will benefit from being closely intertwined with the generation process. Eg when splitting up stems, you can use the diffusion model(s) to help jointly converge individual stems into reasonable standalone tracks.

I'm excited about the potential of these tools. I've definitely personally found uses cases for small independent game projects where a paying for musicians is far out of budget, and the style of music is not one I can execute on my own. But I'm not willing to sacrifice on quality of results to do so.

I think the problem here is the same one as the other current music generation services. Iteration is so important to creativity and right now you can't really properly iterate. In order to get the right song you just spray and pray and keep generating until one that is sufficient arrives or you give up. I know you hint at this being a future direction of development but in my opinion it's a key feature to take these services beyond toys.

I think it's better to think of the process of finding the right song as a search algorithm through the space of all possible songs. The current approach just uses a "pick a random point in a general area". Once we find something that is roughly correct we need something that lets us iteratively tweak the aspects that are not quite right, decreasing the search space and allowing us to iteratively take smaller and smaller steps in defined directions.

I know you only implied it but be careful with causation here. We know that there are situations where the free-market does not lead to ideal outcomes. Does it not seem likely for government to involve itself more in areas where the free market is most likely to fail?

Exploring the meaning behind the use of the word "Harmony" was the beginning of an interesting discussion but the last part of your comment unfortunately reads like thinly-veiled partisan jabs. You unnecessarily chose phrases particularly associated with the American left-wing when a more neutral approach would have been sufficient.

The idea of "Harmony" is strongly connected to collectivist ideology. In the west, mainstream political parties tend to embody particular aspects of collectivism, while parties that completely embrace or reject it are relegated to the fringe (eg. Communist or Libertarian parties). In the US, the right-wing tends to be highly collectivist when it comes to the military/law enforcement (eg. "Support our troops"), religious practice, and national identity (eg. "Patriots") while the left-wing tends to be more collectivist when it comes to government programs and policy (eg. Universal Healthcare).