I know humans that hallucinate more

I can see that actually. I guess the post points out two things:

1. the dev being a a bit deceitful, downplaying their use of AI when clearly the project is like 90% AI
2. the community being frustrated by the lack of quality signals in the new world. Previously programming took a lot more effort, so just the existance of an app already meant that the dev was mildly competent. And at the very least it meant that the dev could fix and maintain the app when something broke. Now those trust signals are gone.

but I was so new to the game I didn’t even know what those words meant

There’s nothing wrong with being new to something. But you shouldn’t be new to rust and releasing an app. Normally by the time you’ve written such an app you would be familiar with rust, but vibe coding allows you to bypass that wall

Those papers are based on dense model architecture. The MoE architecture that I mentioned a few comments ago, does not follow the same laws. Architectural changes could push us past the wall.

Not to mention if we reach 90% accuracy (which was defined in those papers to be AGI level), there’s no reason we will need to keep making new models and training them. AGI is good enough. After that we improve inference performance and bring inference cost down.

Well if we use vague metrics like those, then anybody can claim anything. “more than 1/20 tokens wrong” what does wrong mean? One view is that a computer program is never wrong, it does exactly what the code says. Another view is that if the AI ends up at a verifiably incorrect answer (for example if you prompted it with a math question), then all the tokens it gave out were wrong. But then humans can be wrong too. Are humans 1/20 neurons wrong on average?

For comparison, Moore’s law uses well defined metrics like computations per second. That’s what made it a useful concept.

There are obviously more such cases under a forced-digital regime

I wouldn’t just dismiss it as “obvious” when that’s the topic of the entire discussion. You first have to prove that digital methods make it harder to boycott certain companies. We already discussed how offline access can still use third-party services when processing the data (like OCR services). And on the flip side, digital methods can avoid third party dependencies (like the government hosting their own email).

Indeed in the end forced-digital is just all about saving a buck at the expense of our freedom.

Maybe try to convince the other citizens to pay more taxes in order to maintain an offline system. Forcing everybody to pay for a system that that most don’t care about, isn’t very “free” either

My argument was that the models would get more powerful over time, even when controlling for size and energy usage. I wasn’t talking about waste

I’m no physicist, but afaik particle/wave duality was already a thing before quantum. So we were already representing the world as waves aka frequencies aka vibrations. But then we discovered that those waves were composed of smaller waves. The rabbit hole can go on ad infinitum, as we zoom in to these waves trying to figure out the equations behind them, but we keep finding more complexity, like a fractal.

Moore’s law continuing was due to a ton of different advancements and innovations, not just one. And yes it’s slowing down but it still went for 30+ years. If AI continues to improve at this rate for 30 years, hard to imagine how good it could get.

There’s been a ton of innovation in the space right now. Like MoE, which was only introduced like 2 years ago and now it’s everywhere. It’s hard to say what can happen when you have millions of engineers working on something.

in response to your first point: 24B models from this year are far better than 24B models from 3 years ago. Same model size, similar energy consumption, far better results.

30 years ago there was a lot of doubt that Moore’s law could continue the pace for so long. There was no evidence that PCs would continue improving, and yet they did.

So it’s really anybody’s guess whether or not AI will continue improving. But with the amount of money being poured into it I’m willing to bet it will.

Though it’s crazy that there can be such dramatic variation in bone structure and shape (eg humans vs horses vs bats), yet such homogeneity in the number of bones and their connections (arms connect to shoulders, which connect to spine, etc).

Perhaps the genetic code for the number of bones and their connections, is more resiliant to mutation?

My point is that if you truly want boycott power, even when dealing with mandatory government services, then you need to minimize third-party dependencies for those government services.

You rightfully recognize the difficulty of such a proposition. Let’s say the government outsourced prison food to www.pizza.com. Then those prisoners would not be able to boycott www.pizza.com.

So if you recognize that boycott power has to be waived in certain cases for practical reasons, then it is fairly reasonable that you might just have to use MS to use a government service. Offline access is very expensive to provide, and very few people want it, thus it’s not a practical alternative.

honestly, I wouldn’t be so hard on yourself. This stuff is way harder than people think. People don’t realize how hard it is to establish trust starting from none. Normally you type a website into Google, and Google has already done the work for determining which website is the legit one and which is the shady phishing site, and will filter out the shady site. This convenience does not exist for darknet sites, so you just have to establish trust yourself.

even asking for the true php keys from you right now is submitting to defeat

not necessarily. You can get the pgp keys from random strangers online. It’s just not the only source you should rely on. Get it from multiple sources and then verify if they are all the same. If they are, think to yourself how likely it is that all 3 sources are actually the same attacker giving you a fake key.

DM me if you’re actually interested in the pgp key and I’ll dig it up from my notes

I’ve heard this a lot but I don’t put too much weight to it. It’s too easy to forget that you had bills to pay and just think “if I had just quit my job everything would have been sunshine and rainbows”

You seem to have a strong distinction between direct vs internal systems. I think you underestimate how easy it is to add a facade and convert a direct dependency into an internal one. In software this is called “adding a layer of indirection”. For example, if the government wanted to hide the fact that they use MS for email, they can simply use email aliases. If you send an email to someservice@website.gov, you will have no idea what provider they are using (assuming that they take measures to hide it).

Minimal processing does not restrict /who/ may do the processing, just that excessive processing is banned

I was talking about minimizing third-party dependencies, not minimizing data abuse.

Without copyright, game creators work even harder to protect their games from copying…

Those same creators would work just as hard to work around laws that require offline access. Because as you just established, they want to protect their games from being copied, and online access is their technological measure for doing so. These game creators would just add an online leaderboard to their games and say it’s an essential component.

But if we go so far to abolish copyright, presumably other forms of creator compensation would become dominant. Forms that don’t rely on copy restrictions. After all, not all industries can just add DRM and technological measures to protect their creations. Photographers and digital artists also want to protect their works.

Though this is a big topic I’d rather not get into. I’ll just say that trying to force game companies to provide offline playability, is fairly unrealistic. I’ll wait to see how the “Stop Killing Games” initiative shakes out.

You can learn discrete at the very beginning. That’s the beauty of it. It’s so simple yet so difficult at the same time. Same with number theory.

if the attacker has physical access, you’ve already lost the battle of security

So remote attestation and confidential compute is all a lie? Many security experts would disagree with you

ime rideshare drivers are generally incredibly good at driving, so I trust their judgement. Are there stats that they get in more crashes?

that’s probably generalized too far and reaches into pseudo-science.

Just understand the concept of Turing-completeness, and the idea that many systems are Turing equivalent.

Well the innovations clearly must be good because people use them so much. Frankly the term “illegal” doesn’t carry much weight for me. Piracy is illegal, but I think copyright is far worse.