Or hear me out here… we as a species have developed measuring methodology that is precise enough at human scale but not actually correct and breaks down at larger scales.
> Please stop with this argument. This is one of the easiest problems to solve. Heat dissipation requires a radiator, which is just a dumb hunk of metal.
Convective radiation does not happen in space and this challenge is far more significant than your comment implies. Rather than "a dumb hunk of metal", radiators for spacecraft are often made of ceramics and carbon laminates with higher IR emissivity than convective radiators made of simple metals.
From the article you're commenting on:
> The chips can operate for about 15 minutes in space before needing to be shut down so they can cool off, said Travis Beals, Google’s senior director of product management for Project Suncatcher.
ML Infrastructure comes with some pros (larger emissive footprint) and cons (exponentially larger TDP) compared to the concerns there, but if you aren't familiar with the challenges of heat dissipation in space, please give it a read.
There's also a pretty interesting pop-sci article on cooling the Webb telescope, since it needs to be especially cold for its purpose. Not directly related, but may give insight into both challenges and solutions as well as well. https://science.nasa.gov/mission/webb/science-overview/scien...
Emissivity is one factor, but it is dwarfed by the T^4 term. Sure, maybe if you use exotic materials you can get from 0.9 to 0.95 emissivity, but why bother? Just run the radiators a little hotter.
The equation is:
A ~ (1000 P) / (2 e k T^4)
Where
A is the radiator area in square meters
P is the power in kilowatts
e is emissivity (usually 0.9)
k is the constant 5.67e-8
P and T are the dominating factors. Don't worry about emissivity.
Emissivity is an important factor here because as I said, and as the sources I linked for you to reference clearly stated convective radiation is not taking place in space.
Both Google and NASA are worried about this for a reason, if you think they are wrong, you should offer your assistance to them, rather than debating me.
Oh, sure yeah I agree with that. I was just making a point about the “big hunk of metal” comment and the fact that emissivity is important because the expense of radiator materials aren’t the problem - the surface area (and mass) of the radiator is.
And I agree with you on that. "Big hunk of metal" was too much of an exaggeration--SpaceX's design has liquid cooling, so I assume they have some channels or tubs running through the radiator, plus pumps, etc.
Yes this sounds just like the effect where people new to coding AI negatively berate it like a person and it continues to get worse and make more mistakes because that’s what those tokens are related to.
Surely tiktok being heavily composed of people doing creative work that have understandably extremely negative views of AI would not be the cause of this, no, it must be an unprovable hunch.
While I agree with some of what you've said and the conclusion you've arrived at, in the end, I think you've missed part of the picture here with regard to "prompting experiences".
> Communication only works if you have multiple levels of representation and abstraction, including but not limited to - letter shapes, grammatical structures, style and register, stylometry
These are methods of encoding, there's no reason all of these can't be represented in an LLM from a technical point of view.
> and subtext.
This is the other half of the equation to me. Humans communicate by relating shared experiences, an LLM cannot have shared experiences. While it might be able to encode subtext that has been specifically called out and explained, it will never be able to encode the breadth of human subtext, especially that which is reliant on emotion.
I don't believe it is possible to change this until the point mankind truly develops a "wetware interface" to the digital world (and I personally don't want such a thing to exist).
> Is he arguing that LLMs pretending to have emotions adds more unpredictability?
Unpredictability or a weight towards dangerous actions, and it’s fairly easy to understand why. Humans in distressed emotional states take actions and speak in ways that would not be considered rational. They do this in prose, and they do this in internet conversations.
An LLM trained on these sources may necessarily drift towards those weights if it is trained to behave as if it is emotional and in danger.
What do we do about it? Do we stop AI training? This is silly and not enforceable given its global nature in my opinion. I believe we should regulate and hold accountable those who deploy and use it. But good luck enforcing that in the current kleptocracy.
It seems like a rational approach for several reasons.
- The need for empathetic communication, including understanding the motivations in advesarial situations.
- The emotional bias in in-seperable from the human corpus.
- Desire to have the ability to craft human like communication.
So then the choice becomes do you try to deny emotions exist in the model and you try to blanket suppress them? Or do you try to lean in and craft what we would describe as a "well adapted" persona? I suppose there is a 3rd option of increased meta-cognition which to me seems even more dangerous as it by definition means the behaviour is duplicitous.
I think we have seen people want to use the agents in ways where it has to act as a peer or an subbordinate and I don't see a way of doing that without it having an emotional register.
I don’t think that using an LLM in a way that acts as a human being should be considered acceptable or appropriate. It should be viewed as detached from reality and concerning due to the mental break from social life that appears to come with such usage.
Given that belief, yes we should be training the models not to mimic emotion.
Most philosophy seems to agree that you can't have an intelligence with agency in the way we think of a general intelligence without emotion.
I think when you start to dig really deep, you'll find it isn't actually easy to sever the simulated emotional components without breaking the agent or worse greatly increasing the paperclip maximizer likelihood.
Why would a lack of emotions lead to greater paperclip maximizer chances? If anything, wouldn't an AI that felt really good, or had a simulated digital orgasm for every paperclip it made have a stronger, not weaker drive to create paperclips vs an unemotional AI that didn't care one way or the other about paperclips?
Regulation is easier said than done, in part because the regulation surface, so to speak, is broad and complicated.
Even a badly misaligned LLM is only as dangerous as its tools, but that's a poor regulation target because it turns out to be very very difficult (probably impossible with current LLM technology) to build a toolkit that is both useful for autonomous work and safe in the sense that it can't escape its own sandbox or otherwise perform malicious actions, whether it's because of misalignment or because of malicious prompt injection.
Another option is to regulate the training process. Perhaps an LLM may not be legally distributed unless it contains certain RL steps that penalize malicious behavior and reward self regulation. That that's going to seriously limit innovation while also heavily favoring incumbent labs who can check the boxes and maintain a paper trail of such things.
The other option is to regulate observed behavior, like how airplanes and cars have to meet certain minimum requirements but have some latitude in how they can achieve those requirements. In a framework like this, you can't distribute an LLM until it's past some formal audit or testing procedure, with some kind of formal certification regulators will ask you for and fine you if you don't have it.
Regulating observed behavior is maybe the most tractable approach, and it also works the best with our existing frameworks for regulation, where you always have some kind of a division between DIY/hobby projects, which tend to be lightly regulated, and commercial projects, which tend to be more heavily regulated. Of course, even drawing such a line itself will be challenging.
And that's before you get into any problems of regulatory capture, fun stuff.
If you try to regulate training and tools, you end up with a space where you're trying to use the law to reign in a relatively small amount of experts. That didn't work for the early internet, or even the relatively recent internet (series of tubes, anyone?).
So regulating observed behavior makes the most sense to me as well. Some of the most sane, broad protections can come from that category - stuff like "you're not allowed to let your AI commit cyber attacks on other people without their consent" or "you're not allowed to put an AI in control of a medical device without passing these safety reviews".
With the usual caveats applying - regulatory capture like you pointed out, or fines being so small that they are essentially just line items on the cost of business.
As far as I can tell, after months of fighting being DDoSed by Anthropic and OpenAI across 50+ sites - Cloudflare also allows what it considers "good bots" through all of your bot blocking rules, with no option to turn this off unless you pay them money.
No, I definitely spent months fighting off bots across multiple hosts and never set up a robots.txt file anywhere nor did I look at the analytics dashboard on cloudflare.
To that point, Cloudflare quite literally tells you the source of the bot traffic on your sites that use it as a WAF! And what percentage they are allowing through, despite you setting rules specifically to block them by name.
reply