joe

I meant no insult, truly. You just don’t have the minimum required knowledge to discuss this intelligently. There’s no shame in that; no one can know everything.

I can’t know the motives of anyone, but I can say it’s at least plausible that the people working on the frontier models want an excuse to slow down that doesn’t give an advantage to other countries and companies, and also doesn’t put them in the crosshairs of investors and board members.

You apparently still internalize LLMs as fancy Markov chains, which tells me that you don’t have a working grasp to discuss the topic, even at the laymen level we’re at.

These hacking events may be “PR Stunts”, but the “stunt” part would be leaving a known vulnerability as a path to the wider Internet instead of making it physically impossible to break out, not the hacking of other websites. No one credible is disputing that these models are capable of hacking.

Your opinion is borne from lack of knowledge about the topic you’re trying to discuss, similar to an anti-vaxxer attempting to discuss the dangers of mRNA vaccines.

Fix your lack of knowledge or don’t; that’s your call, but if you care about discussing this topic somewhat intelligently, you should probably put in some effort to understand it.

The groups doing frontier research with the appropriate funding are few enough that a global pause would be possible, and useful.

I agree that a ban would be impossible, and counterproductive even if it were possible.

Any pause will have to be a global treaty; the unfortunate reality is that we’re in an arms race now, and despite Sanders’ threats otherwise, I don’t think there is going to be much will to pause in the US while China keeps developing.

And to be clear, I think it would be every wise to pause development and, afterwards, keep the pace slow. We know how to make “AI” more capable, but we don’t know how to make it stay aligned with human goals. The technology is useful and isn’t going anywhere, but that doesn’t mean we have to develop it as quickly as possible without regard to the real dangers it brings with it.

They were being pretty vague, but it could be something like “25 characters after every comma used, put a vowel. 25 characters after that vowel, put a space. 25 characters after that space, put a period.” (Except much more complex than what I made up.) The point is to string together a text pattern that is so exact that it can’t be just a coincidence.

The “watermark” is the text, so when you copy the text, you’re also copying the watermark. It’s also why they say it might not be conclusive if the text output is manually edited, or the output is too short.

Not for nothing, I think these types of endeavors are going to have unintended consequences. Despite them flat out saying that the lack of a watermark doesn’t mean it wasn’t AI generated, I think it’s very plausible that the proliferation of watermarks will lend credibility to anything without a watermark.

I would love it if you’d point me at these mathematical proofs.

Those recommendations are for protecting data you want to stay a secret; it would also apply to unlocking this fancy USB drive, as a best practice.

The USB still has advantage in that no one will compel you to unencrypt a drive that they don’t realize is encrypted, which can be a pretty significant advantage depending on who you’re hiding the information from, or why.

The issue is that AI companies have been funded on the promise that with enough advancements and upgrades this tech will achieve AGI. Which is an absurd statement to anyone in the field actually developing these things. That is the equivalent of promising that this 260mph Bugatti, with a few years of upgrades and advancements, will become a Harrier Jump Jet!

A better analogy is promising that a toddler will, with enough knowledge and training, eventually become a heart surgeon.

15 years ago the assumption with AGI was that we needed some paradigm shift in technology to create it, but after transformer technology was invented (the T in GPT), and it was trained on a lot of information, we discovered an emergent property that it could take natural language queries and answer them with its knowledge base-- which was unexpected and unintended.

While science can always end up going down the wrong path, the current mainstream stance is that we were wrong about needing new technology for AGI; it seems that AGI may be a function of information and training, on hardware we already have. Hence all the data centers being built.

Anyone who says with certainty that it will result in AGI is just as wrong as someone who says with certainty that it won’t.

They’re talking about the ads in podcasts put there by the podcasters. Not ads put there by Spotify.

Notably, the game data is not loaded into RAM. In case it isn’t clear, those requirements are very low, by 2026 standards. If it can display video on the monitor you have, then it can probably stream a game.

I thought the extra question marks would have made it clear I was being sarcastic. No sarcasm this time: I’m sorry I wasn’t clear enough. That’s on me. Sorry.

My ambiguous sarcasm aside, I was pointing out that your solution also had the same problem you were trying to solve, because that problem is always there, no matter what.

Not that you bothered to ask, but my solution is to add enough operators so that they aren’t working barely staffed at all times and getting overwhelmed the second more than a few calls come in.

And if we lived in a world where money wasn’t very frequently the limiting factor, your solution might have merit. Unfortunately, capitalism is still something we have to deal with.

Which, if it’s literally just a binary yes/no why the hell do we need AI and not ‘press 1 if you’re calling about x emergency’?

But but but what if the person is stroking out and presses 1 when they meant 2???

All it needs to do is be at least as good as a human, and humans get tired, distracted, irritable, etc. so it’s pretty easy to outperform humans.

It’s only looking for the word “yes”. Literally any other sound and it routes to a human.

I’m actually married to the idea that the same thing could go wrong with a human, so from that standpoint it’s a wash.

It would have to be a stroke during the time there was an ongoing scenario that resulted in higher than typical call volume about the scenario, and a stroke victim calling in who could only say “yes”.

Not impossible, but wildly unlikely.

Sure, usually, but compare like to like. If a human operator asked if the call was about an already known scenario and transferred people to an automated service if they said “yes”, then the same exact thing could happen.

You attemtped to craft a very specific scenario in which harm would result, when in reality it is wildly unlikely that someone would call in at that precise time and only respond with the word “yes”.

I can only speculate as to why you would do that.

A person would catch something was up even if they said yes.

Besides “because it supports the conclusion I want to arrive at”, what makes you believe this is true?

The requirements for cloud gaming revolve around having the hardware to drive the local monitor, and having fast enough internet with low enough latency.

For example, here are Geforce Now’s minimum requirements: www.nvidia.com/en-us/geforce-now/system-reqs/

There is some overhead involved in streaming, but the RAM requirements are low and certainly don’t change based on the game.

There was recently a study using human short stories and LLM generated short stories prompted to write a similar story. The participants where told which stories were AI generated, but sometimes they were lied to, and told a human-written story was AI generated.

The study showed that people liked the AI stories more. It also showed that if someone has a negative bias against AI generated content, they’d claim the AI story was worse, even if it were actually the human story and they were just told it was AI.

Also, another test was performed where they provided a story and asked them if it was AI generated or human generated. The participants could only tell the AI written story 45% to 50% of the time. Essentially, they were guessing.

More info: theguardian.com/…/ai-generated-stories-rated-bett…

I think many people have taken a mental snapshot of the quality of AI generated content from a couple years ago and just assumed it hasn’t improved since then.

Off the Grid as well, I picked one of the recommended models, Gemma 4 EB4.