Llamamoe

The problem with Wikipedia is that quality varies greatly across fields. Technology or software engineering are great, physics varies, but e.g. even articles about the most basic math concepts are completely incomprehensible.

I can often ask an AI to give me an easily understandable three paragraph overview of a concept that on wikipedia has a hundred paragraph article requiring multiple prerequisite textbooks to even start to comprehend. And that's for basic math concepts like multiplication or exponentiation.

Healthcare is one of the most critical fields to our collective wellbeing while simultaneously being one of the most problematic ones, with doctors routinely disregarding, misdiagnosing, and otherwise failing to help or even harming patients, especially anyone who isn't a typical case of a typical issue.

I have more sympathy for almost every other profession affected than doctors.

Ties or not, it's true that the label is imprecise and that not everything that's been highly processed is equal.

For example I'd argue that wheat flour and subsequently bread or pasta are in fact very highly processed, and carb bombs to boot, but nowhere near as bad as some other stuff out there.

Or take soy protein nuggets. As ultraprocessed as it gets, but probably healthy if anything.

It's a really complex topic and I feel lkke any sweeping generalizations are bound to be misleading in some ways or others.

That's just a tool you rent and cannot own or make/improve/repair yourself.

I'm not sure who the "you" in the article is, but I find both the article and the summary void of any attempts to explain anything intuitively, extremely math-dense, and essentially incomprehensible.

There is no guarantee that hardware will remain accessible if tech corporations continue pricing people out of the market. Having a PC powerful enough to run genuinely useful local models is already a privileged position that many people in the world do not have.

While AI inference is far less dependent on large scale organization than most other forms of automation, it is not intrinsically independent the way e.g. being able to get a 3D printer is - at least until they ban those on grounds of potentially facilitating gun fabrication or such.

What broad shifts in political leanings are there with age other than the overarching "well, I got mine, time to pull up the ladder" shift to conservatism?

And I wish I shared your optimism about the future AI enables. Without some massive societal transformations, it's on its way to doing what all prior automation did - disproportionately benefit the oligarchy as they use it to capture more and more of the wealth of our civilization by disempowering people who have increasingly less and less leverage.

Also, I genuinely think that a WW3 might not even be top 3 of our biggest worries between the constant war against privacy and AI that would make it possible to finally realize a truly all-seeing, irreversibly authoritarian dystopian state, or even just models purpose-built to create content and adjust algorithms after end-to-end "how does this affect a person's worldviews?" training. That's basically mind-control.

Or even "only" almost all wealth and power being concentrated in the hands of those who control AI and resulting automation.

Nevermind the extremely real extinction risk.

War would definitely be a far worse short-term tragedy, but it's inherently polarizing against itself and doesn't carry the risk of insidiously bringing about a potentially irreversible dystopian state in itself.

You are terrified because somewhere deep down, what you want is someone who will act on worldviews similar to your own, and age is a fairly substantial component of that.

Ultimately, perhaps there is a trend, but my observations simply do not match the premise of people monotonically accumulating wisdom as they age, and in fact there are certain types of gullibility that seem to increase with age, such as the willingness to disregard and disparage scientific or objective fact in favor of personal beliefs, and to prioritize social status over signals of expertise/intelligence when judging authority. And don't forget that 80%+ of successful scams are against older people.

I think there are almost definitely tradeoffs between cognitive flexibility versus cumulative lifetime experiences, and we shouldn't discount the former - just look at how utterly dysregulated tech industries are because of out-of-touch representatives who listen to elites over experts in the field.

In this context, what does "closest" mean? Most similar token embeddings? Most similar Q embedding to prior K embeddings? Most similar positional embedding?

Are you, in essence, implying that the Key-Query attention mechanism is basically a neural function approximation of the Nadaraya-Watson kernel or something very like it?

The core of this issue is that under capitalism, automation results in disempowerment and wealth concentration instead of collective benefit from improved efficiency.

Even AI - at least until some undefined singularity point - would have been an enormous boon for the world if it wasn't all but guaranteed that it will be used to disempower those less capable of affording and employing it at an industrial scale.

Considering how unlikely even the most ideal world would be to cater to the exact conditions we've evolved to adapt to, I'd say "more habitable than earth" is impossible without extreme terraforming of a lottery ticket planet.

Successfully sued to result comparable to common people being fined a fiver for murder.

I don't think reality TV was ever that big of a phenomenon in most of the world. It's a very American type of show, tbh.

It doesn't have to be "not free speech" to be deleterious, that's an insane argument.

A large part of how human settle on opinions and beliefs is by inference from what and how often is said around them, calibrated around in-person interactions.

Social media warp this horribly both through algorithmically selecting what information to present users with, and through impersonal "big account->user" that remove the stakes normally present in in-person communication, which completely breaks our perception of how important statements actually are.

It doesn't, but it does filter out people clueless about the topic itself at least, which is a better baseline than a survey of random people who self-claim to have insight into the matter.

This is basically standard practice worldwide for most "sugar free" products. Most common sources are dates, apple juice concentrate, honey in my experience.

Products actually using erythritol, mannitol, or xylitol instead are a rarity.

> Interestingly backprop is limited by conditioning of the Hessian matrix in order to converge (differentiate correctly)

What does this mean?

Agreed. It was literally the wake up call for the fact that air quality standards are as important as soap and handwashinhg are, and yet... nothing happened.

And so the question of what effects getting rid of several-yearly respiratory infections would have on our health and the development of children remains open.

You could probably improvise something cheaper using the same filter arrangement, though now that you have it, sounds like the best thing to do might be just including an extra sacrificial initial filter to increase the lifespan of the proper ones.

Better "understanding of technology", probably not - early computing and internet required a lot more technical insight, and younger generations are notoriously tech illiterate.

Insight into how it effects society? Very possibly, at the least it's a different, more native perspective. Whether it is also broader or more correct, it's hard to say.

Much if fhe inspections and standards aren't "necessary for safety" but the direct product of paranoia and lobbying. Nuclear is brutally overregulated, far, far beyond what's necessary to ensure safety, and that's where most costs come from.

Encryption backdoors by definition mean everybody who wants it gets hands on your data. You can't make backdoors only the "good guys" can access.

And while, sure, adtech corpos maybe won't due to the bad PR, what about the Kremlin, criminal organizations, or your insane abusive ex?

This might have been plausible if this study came out of nowhere but there have been placebo controlled studies inducing/reversing autism-like symptoms in mice with fecal transplants and studies linking individual git microbiome metabolites to symptom severity in both animal models and humans.

Yeah the research is still preliminary but it's still pretty strong evidence that it's a research direction worth exploring.

TLDR of the article is that it's yet another person confusing "heritable" with "genetic". Genes aren't the only thing we inherit from out parents - and in fact there's strong preliminary evidence that the gut microbiome plays a major role in autism.

Heritable != genetic.

There are very strong reasons to suspect that gut microbiome at least plays a major role in at least a large portion of autism cases, complete with studies where we've been able to induce/reverse a subset of autism symptoms in mice, and even humans, as well as linking individual gut microbe metabolites to them. And ~80% of autistic people have gut problems.

There's no question in my mind that at least some of the propensity to develop autism is genetic, but calling the whole of it genetic feels ridiculous in the face of preliminary evidence to the contrary consistent with the massive increase in diagnosis rates beyond what can be explained by changing criteria.

Is a misaligned AI worse than a super - capable AI aligned with corporate interest? Because I can't really tell anymore.

You can smoothly interpolate between different projections, and the scale differences are large enough that it should be gradual enough not to notice much. Either way, globe view is always superior.

I believe that multimodal training increases the robustness of latent representations regardless of which modality is being processed.

In the same sense as pharmacology is a branch of physics. Technically, yes. Realistically the overlap is small and you don't need one for the other most of the time.

So like always, the people who get shafted the hardest are the chronically ill, the fatigued, the ADHD, or even the depressed.

Executive function has always been the beginning and the end of efficacy and capability.