seanmcdirmid

CSAM laws in many jurisdictions include computer generated or hand crafted media that do not pre-require CSA at all. The idea is that deviance leads to CSA, and by cracking down on CSAM, we are preventing CSA.

I don’t understand how healthcare can get even more expensive without the system just utterly collapsing. That being said, we could definitely justify higher salaries with greater productivity (via automation), but the market tends to reward the capitalists for that instead of the workers. We really have to figure that out soon.

You might want to go over the original Elliot/Hudak work? It’s a bit obtuse since it’s all in Haskell, but signals were never meant to be historical accumulators unless you set them up that way. Signals are continuous values, they don’t reveal discrete events like streams do, that’s the main distinction between them, your signal computations then can be viewed as a continuous value derived from other continuous values.

In early FRP systems, you could just plug in a t and compute a signal network (well, behavior) at any point in time, so the entire history of values needed to be saved. Modern FRP abondons first class time and focuses more on change propagation (you are told when your signal computation changes at some end point, otherwise no history is saved).

Anyways, it’s been a decade since I’ve done this stuff. Erik Miejer kind of muddied the waters with his observable stream work, and people started calling that FRP instead. I see the Jane street blog post is based on Evan’s work, I don’t remember much about it.

I’ve always implemented purely non-history preserving signals (Modern FRP) just because they are incredibly easy/efficient to implement and give you 99.9% of what you want and…first class time is just not useful unless you are doing simulations or something (I’ve never called my work FRO, just FRP inspired). It’s also easier to explain to people a continuous time function (the UI label changes as the values it is bound to change). I even have some work where I translate it down to WPF data binding in C# using the DLR to compile signal expressions.

The incremental computation landscape is pretty vast, but I always found that memoizing sub-trees of a computation and then doing a standard damage and repair algorithm to recompute subtrees that depend on changed values works well enough for most use cases. Again, not really related to FRP, but I somehow got involved in both worlds a couple of decades ago. You could try to define a function over the delta, but it’s incredibly dependent on the domain unless you restrict yourself to a differentiable programming language.

You aren’t thinking this through. You were just told that this wasn’t stacking but you keep insisting it is. N-versions is explicitly not stacking, and the same solutions that solve the Knight and Leveson issue for humans solves it for LLMs as well (they can be wrong about the same part of the specification, but will very unlikely be wrong about it in the same way, so you try to measure scalar rather than Boolean outcomes).

You can build some lightweight dependency graphs that flush out quickly. You don’t need to describe how the signal has changed, just that it might have changed, then flush dependencies on change (some listeners might be notified of a change they don’t care about anymore) and re-register them when they come back for fresh data again.

But incremental computation isn’t exactly functional reactive programming. They are different domains in practice that often get thrown together because the problem they address can overlap. Incremental computation explicitly derives a function that can operate on deltas, FRP might just use damage and repair instead.

This. The math of coincident errors is why we can get reliable AI, but it really is a matter of how the errors manifest and how they are detected.

I’ve written a bit about this and my approach is based on clean room and N-versions.

VLIW is being heavily used in some AI inference tasks (by Apple, AMD, Google, mainly in optimizing latency). It also is a nice way to do inference on edge devices that aren’t using constantly advancing models. It’s pretty much dead in general purpose computing though, and lacks the versatility of GPUs to go beyond inference.

You can reduce hallucinations with appropriate grounding checks. You can’t really get rid of them though, so feedback loops with verification are essential.

People busy building shippable things are generally too busy to blog about it. It is selection bias in its most classical meaning.

There must be some Alibaba posters who could clarify this. I think it’s like how Amazon pushed AWS, alibaba is hoping create a similar ecosystem. I wouldn’t conflate them with the CPC or other more niche players like deepseek.

In theory, rising energy prices would encourage more investment in capacity increases, that would then real a new equilibrium and so on. In theory of course, energy markets are so distorted now who knows. Having AI data centers bring their own power or power investments is probably going to be necessary.

US politicians are all over the map on this, but they aren’t really talking about Chinese AI much, it’s not as visible or tangible to most Americans like TikTok was.

I actually have long discussions with Gemini about this and have wound up download a bunch of different models for different things. There is no best, just fast but worse, slow but better, agentic or not, reasoning or not great at large contexts, better world knowledge, uncensored, etc…. It’s a bit daunting actually since there isn’t really a one size fits all model that you can just use for everything.

Queen has that already, although they seem to be moving away from local models unfortunately.

How does this work for RAG? Do they make it so the model doesn’t have that fact in their weights or do they make it not talk about it when it is included in context.

Ironically, Chinese models have the most uncensored versions available for download. Fairly sure they own the porn market.

Alibaba isn’t really the Chinese government though, or are you saying Americans will think that ever since Jack Ma was harmonized?

This isn’t a PRC specific thing either, TSMC is infamous for having PhDs doing night shift lab tech work for pennies (comparatively).

Engineers from Taiwan go to mainland China these days to earn more money. Taiwan was pretty brutal with personal sacrifice in its development as much or if more than the mainland. We could say similar about Korea, Japan, and Singapore as well. This is why Asia seems to be taking over the world now, but the people are about as happy as you’d expect.

Greenspan commission wasn’t a prediction. They used a previous pay as you go model that was simply broke at that time (and had been for awhile$, leading to the current pay for the future model. They then made predictions that the new system would be solvent way past 2033.

Goodhart’s law is the problem, not the metric itself. Also LLMs do not have any visual generation skills, so its idea of a pelican looks like purely linguistic, unlike diffusion models. That we get decent results at all from an LLM outputting SVG files of random things is just nuts to me.

The sad thing is that those people who need social security the most usually have it for the least amount of time. Obviously are priorities suck.

Ya, there are wide different outcomes to consider with respect to health. Also people should really be retiring earlier, not later, as automation picks up. The way things are progressing right now are just whack.

2033 is when all the bonds should have been redeemed and there is no money left to redeem, just contributions coming in and more payments going out. And yes, congress can do a straight injection from the general fund, it’s just even more deficit spending at that point.

The date used to be later but COVID drove it forward a bit since they got rid of contributions for some people for a couple of years I think? Generous cost of living increases have also caused the date to creep forward, but I don’t think it has ever moved backward.

The math has been constant, if you were reading reports saying it would hit the breaking point in the 90s or 2000s, you are either misremembering, confusing it with something else, or willfully distorting facts. Anyways, if you make claims like that you should take advantage of internet linking features and just provide your evidence directly vs a “I remember reading” anecdote.

There was a point where social security went from surplus to deficit in 2010. That wasn’t insolvency though since the federal government owed them for the previous surpluses they borrowed. 2033 is when the federal government no longer has to pay back money to SS and it is truly insolvent.

If nothing is done it’s a mathematical fact. The problem is that there are no plans to do anything yet, each administration and congress seeing it as too politically volatile to deal with, which means we will have a last minute suboptimal solution.

Obviously there are solutions. Maybe if America survives that long we might see one implemented.

I tied it depends on your bias? California is pretty capitalistic, the social safety net is still pretty light, elected leaders still look out for business interests rather than the people, and the people would stop electing them if they did otherwise. California only looks left wing compared to Texas, to the much of the world California looks pretty right wing, especially if you didn’t know it just got worse from there. Capitalists don’t flock to California because of the socialism.

I’m all for funding schools at the state or even national level and eliminating district inequality based on the price of housing. However, prop 13 creates huge unfair distortions in the tax base, where some people are paying super low taxes and other people are paying super high taxes, for the same property. Old millionaires who bought their home decades ago get a huge tax break and a struggling young family who just bought basically subsidize them. It’s the ultimate of ick. It’s definitely a great way to push the birth rate down.

Not that I would call California a left wing state, but prop 13 came about during Reagan, and it has been incredibly sticky afterwards even if it isn’t something that would pass today.

Also, schools are funded by the state rather than via local property taxes mainly because of prop 13. It’s odd by American standards, but probably more closely resembles funding models of European countries more than other states do. But I guess that is just the nuance you were talking about?

That is manager/executive/manager speak, real people don’t speak like that (unless they are in the aforementioned roles).

If you have access to the detector, you can formulate a generative solution that avoids being flagged. Which gets me wondering why don’t model providers do that? There must be something about that that destroys semantic weights somehow.

It’s hard to say. If they use photvalic instead of pure solar, they can top off a lack of sun with electricity from other sources to keep the hot water warm (outside of winter when it’s all natural gas based anyways). Ome would think they could do a better job with personal heat pumps.