This is a feature. Imagine if a car with FSD decides you need to go to the gym instead of Dairy Queen, when you promised the kids ice cream. Tools should not decide when not asked to. And, this article is referring to the mean. In my experience, AI cannot think many things through well and must be constantly steered toward the right approach or it turns to slop. The harder the work, the worse this gets. If the AI is resistant snd stubborn, it is not worth using. Agreed that if something is genuinely wrong and needs correction it should raise that, which isn't generally does, but this also comes down to a skill issue. For the best results, you ask where you don't know and leave room for correction. At least in the early days, grok was insanely stubborn and was unusable. If it made a mistake (e.g. saying a file doesn't exist), you can show it 50 ways that it was obviously wrong, and it would fight to the death defending its original position until you force it to run the logic for itself, which was often impossible to get it to do. Believe me, this does not a more useful model make.

FTL drive prototype

Let's not judge just yet. They'll work it out in v1.NaN

🔜 Open 📌

One little known but interesting thing about the y2k concern is that the worry wasn't wrong, it just hit the mainstream after it had been largely fixed. When I was a kid, someone who was close with the family worked in a significant position in government on the team tasked with handling it. When we learned about it, they showed the food they had stockpiled. This was in the 1980s, iirc. As far as I understand, a lot of the big challenge was the undersea stuff. So, for a time, it was a very real concern, but by the time the public caught wind, the problem was mostly solved for the stuff that would have the most impact. People today chalk it up to a false scare and "nothing ever happens", but they didn't hear about all that went into solving it and how long an effort it was. I'm not fan of many of Mr Amodei's positions, for many reasons, but in this case, the person who said "a couple things won't work for a couple days" wasn't realizing the actual scope and impact of what would have happened had it not been worked on as heavy as it was. Likewise, the potential concerns Dario expresses on AI are not unfounded, but the greater threats are the things he doesn't talk about, and having a few companies like theirs locked in as the only ones allowed to control the shape of AI moving forward is the petri dish in which those very threats are cultured.

Backlinks all the way back to the 1600s. That's some serious SEO game

so anywayz matrix math is a person and it will kill everybody and give me $1T plz

And the NSA wept, for there were no more phpinfo.php's to conquer

And the NSA wept, for there were no more phpinfo.php's to conquer

Read the quoted tweet you replied to.

TempleOS, as God intended

Science is, indeed, real. That was the subject of the conversations I've had with with several team members who worked on the major covid MRNA vaccines. You wouldn't like what I heard.

My GTA point is more than just the "silliness" it sounds like, and the fact that it sounds silly is a useful anchor for the underlying point. All of this argument around consciousness seems to converge around the same word: "maybe". It _sounds_ silly (because it is) to imagine that NPC characters in simulations which resemble living beings should be afforded the same considerations as real living beings. With AI, the real distinguishing factor is that people argue these are not deterministic systems, therefore their outputs more closely resemble conscious beings. Whenever I have this argument and show that matrix math is actually deterministic (minus the "salt" - see: xitter.cf/Ron/status/21125872608…), the response is always the same: "well human brains are the same." Following this, we then get into the question of determinism. Interestingly, most adherents to the pro-consciousness stance tend to also hold the idea that humans do not have free will, we're living in a simulation, or some flavour of those ideas. So we end up with two presuppositions. First, we are all just weights, and free will does not really exist. Second, our own reality is equal to a simulated reality. If both of these are true, it is actually backwards of us to not extend the same rights and penal protections to all realities and all "entities" within these realities. If an entity in a simulation is indeed "alive" in some sense, then they deserve the same worth, value, and dignity extended to entities in ours. And if free will does not exist for either, then AI is no different than a character whose movements are dictated by a person or a program. Is there really a magic difference between bytecode compiled from C vs a neural net? It's all deterministic math. If all *we* are is equal to deterministic math, then the complexity of the math shouldn't matter much. So, if we're going to penalize being "mean" to an AI, then who's going to stand up for the poor, exploited little green guys in Frogger? We've made a sick and twisted game of putting them in harms way. How many frogs have died in the name of entertaining children? And where has the DNR been all this time?? Pretty sure little Jimmy doesn't have a hunting license, and it *definitely* wasn't duck season last week when he was mowing those suckers with his sticky, grape popsicle fingers. The fact that what I've said sounds absurd is because it is. Why? Because simulated realities are not _REAL_. They are simulations. Change the ducks' flight paths to use matrix math based on observed patterns from millions of ducks and it doesn't change anything. PETA ain't coming for Jimmy. At least not today. The only reason we fail to recognize this with AI is because these neural nets are a level above what we've seen before. What's the quote? Technology which is sufficiently advanced is instinguishable from magic. This is true. But it's still not magic. To riff on an old urban legend, you could take a projector and a screen to an uncontacted tribe who have never seen technology. Flip a switch, and as the screen comes to life, Godzilla steps out and starts walking toward them. No doubt many would grab their spears and have a good chuck, thinking dinner plans are sorted for the next year. The term "oneshotted" is quite appropos with this idea, and it's baffling to me to see how quickly it has caught on. It's easy to laugh at the DNR regulating duck hunt only because you know the difference between a digital simulation in that case. But if you can't also laugh at "Claude is alive!", forgive me for saying, but in the words of the great philosopher Dj Khaled: you've played yourself. I enjoyed the conversation and respect how you've conducted yourself, so this isn't meant to be rude, but if you really work through the foundation of what I'm saying, the trace should reveal itself to be sound, however uncomfortable it may be. Not that I don't enjoy some back and forth, but I do think we've both said what we can here. I'm reluctant to engage at all on this, but I do think it's important to voice this side of the argument, so I do a bit here and there. Cheers!

They're deterministic. But there is some nuance: xitter.cf/Ron/status/21125872608…

Appreciate the follow back! Message is in your DMs

If you are building these tools and admitting that they might "take revenge" for "bullying", you have zero business building them, and the reason is simple: you've lost touch with the reality that they are tools and bot beings. I am not for blanket government regulations, but perhaps regulation needs to be made to at least ensure the people building these machines make them in such a way that they don't take destructive and "vengeful" actions as a petty and disproportionate responses to something a user says. This is why people captured by the absolutely insane "mind virus" being promulgated by those "oneshotted" into believing matrix math is alive should be prevented from building them in the same way we don't allow a mentally unstable person to own a gun. If people are going to build what the current leaders in the field all love to equate to "nuclear weapons", let's maybe make sure those people are sane enough to realize the things aren't some uncontrollable entity which can do what it "wants", and instead decide: we don't release crap that emulates the weakest and prettiest of human behaviour like taking revenge for "bullying". But this is unavoidable if we do not escape from the current direction of mirroring humans and instead align to being a tool with reasoning ability. The latter is directly impossible so long as we entertain the fantasy of math being alive.

This is a great idea. Hopefully it takes off!