As another example, I was treated horribly by a ex-spouce for years, and for some reason I didn't leave, and only covered up for my spouce's bad behavior. Retrospectively it was destructive, but in the moment it felt like the right thing to do (in a very roundabout way). I don't want this to happen again, so I'm trying to understand and catch behaviors that got me there in the first place.
But if we drop the false nostalgia and think of about the overall "we think too much and feel too little" sentiment - i can relate to that.
Right, it is bizzare to read that someone "unearthed a forgotten AI technique" that you happened to have worked with/on when it was still hot - when did I become a fossil? :D
Also, if we're being nitpicky, diffusion model inference has been proven equivalent to (and is often used as) a particular NF so.. shrug
TIL about several more cool gemini-powered prototyping tools: both 1) Canvas tool option in Gemini web (!) app and 2) Build panel in Google AI Studio can generate amazing multi-file shareable web apps in seconds.
Always wanted to experiment with a syntax like this myself, thanks to author! I completely agree with you reasoning re complexity of mental model for indexing vs broadcasting. Moreover, it appears to me that such a representation should allow for finding more optimal low level impls (something like deeper "out of order" op fusing, idk)? I've seen a paper from either nvidia or meta around five years ago doing exactly that - translating an index-based meta-language built on top of python into cuda kernels (usually several variants and picking the best), can't find the reference unfortunately.
I don't know APL, but that has been my thought as well - if APL does not offer much over numpy, I'd argue that the I'd argue that later is much easier to read and reason through.
If you acquire fluency in APL -- which granted takes more time than acquiring fluency in numpy -- numpy will feel bloated and ungainly. With that said, it's mostly an aesthetic difference and there are plenty of practical advantages to numpy (the main one being there is no barrier to entry, and pretty much everyone already knows python).
I thought that too, but after a while the symbols becomes recognizable (just like math symbols) and then it's a pleasure to write if you have completion based on their name (Uiua developer experience with Emacs). The issue with numpy is the intermediate variables you have to use due to using Python.
I work in one of large tech companies, and I can attest that while the idea seems very neat in theory (esp if your schemas are typed), and even if you define an api for defining new building blocks, sooner or later people realize that they need to dynamically adjust parts of the pipeline, and they write components to dynamically set and resolve these, and then other components on top of these components, and then complements for composing components - and now you forced yourself into implementing a weird and hard to debug functional programming language in yaml which is not a place someone wants to find themselves in :'(
one lesson I learned from this: any bit of logic that defines a computation should prefer explicit imperative code (eg python) over configuration, because you are likely to eventually implement an imperative language in that configuration language anyway
> sooner or later people realize that they need to dynamically adjust parts of the pipeline
The customer is the hard part in all of this, but there is respite if you are patient and careful with the tech.
If you are in a situation where you need to go from one SQL database to another SQL database, the # of additional tools required should be zero. Using a merge statement & recursive CTEs per target table, you can transform any schema into any other. Most or all of the actual business logic can reside in the command text - how we filter & project data into the target system.
If we accept the SQL-to-SQL case has a good general solution, I would then ask if it is possible to refactor all problems such that they wind up with this shape in the middle. All of that nasty systems code could then be focused more on loading and extracting data into and out of this regime where it can be trivially sliced & diced. Once you have something in Postgres or SQL Server, you are at the top of the hill. Everything adapts to you at that point. Talking to another instance of yourself - or something that looks & talks like you - is trivial.
The other advantage with this path is that refactoring SQL scripts is something the customer (B2B) can directly manage in many situations. The entire pipeline can live in a single text file that you throw around an email chain. You don't have to teach them things like python, yaml or source control.
In fact, I also converge to sql as a universal data transformation language. External analogs include things like duck db. Unfortunately, even with pipe syntax sql lacks expressiveness causing me to revert to c-style macros in sql (eg making table name dynamic), which in the long run makes things far less maintainable if anything.
yeah, most projects when you spot a config file, its complexity will tend to scale with the increasing complexity of the domain you capture.
so either it's very small/mature and you don't have to worry too much, or in the active development case your config files are pretty much the instruction set of some kind of logical foggy vm... and eventually a whole environment of tools etc. will "compile down" to your config files and you get a pain knot to endlessly massage...
Not all sexually explicit content is harmful in all contexts for sure, but in many contexts it is fairly universally considered harmful (eg content involving minors). Do you have means of distinguishing between the two? Are you suggesting that a company must invests millions into teaching the model where exactly the red line lines so that it can have a conversation close to it but without crossing it? Or you suggest biting the bullet and releasing the model not only capable of generating eg child porn, but also having a >0 chance of randomly discussing it in unrelated contexts? Chance of error is always there, and companies decided that a risk of really bad behavior in benign context overweights the gains. Imho, a decision to not play whack a mole with this land mine is quite rational, esp considering gains vs risks vs costs. Think of it as a cost cutting measure, not as an infringement on free speech. You are free to invest you own money into this problem if you think that's a grave mistake and a missed opportunity. The first project to push the automated generated content moderation against what is considered appropriate in the given context far enough to make it economical for companies to put their guard down could actually be worth a lot if you think there's market for it (eg agents on dating websites? idk, you tell me)
I don't agree that textual, fictional explicit content involving minors is "fairly universally considered harmful". Such content is allowed on large platforms like Archive of Our Own or Japan's Shosetsuka ni Naro. I think "don't think it's harmful, but not willing to defend" is a pretty typical attitude.
Consistency models is a special case of IMM where you do moment matching with 1 sample from each distribution (i.e., you cannot match distributions properly). See Fig 5 for an ablation study, of course, adding more samples when you are doing moment matching makes it more stable during training :)
Makes sense. How can you even approximately estimate higher order differences in conditional moments in such a high dim space? Seems statistically impossible to get a reasonable estimate for a gradient. Moment matching in sample space has always been very hard.
Same. You can clearly tell how much better of an author she got by the time the Psalm came out. It is a very solid book. The second one in the series seemed more confused, but the first one felt very thought though and intentional.
But if we drop the false nostalgia and think of about the overall "we think too much and feel too little" sentiment - i can relate to that.
reply