My favorite thing about emerging tech is the interesting stuff that happens when domains overlap - in this case, next-gen nuclear power plants and AI/ML theverge.com/2023/9/26/2388995

@darkuncle

We learned nothing from bitcoin. Nothing.

At least this is only wildly wasteful rather than an actual ponzi scheme.

@Biggles so you think there’s no value to AI/ML, nor to next-gen modular nuclear reactors? 🤔

@darkuncle

I think current AI/ML has *negative* value for society - it, like bitcoin, causes more harm and potential harm than benefits. We have lots of things like that - heroin, for example. It has super niche uses where it might be beneficial, but on the whole it's a dangerous attractive nuisance.

For example - a big company might invest in nuclear power to waste on the fantasy, rather than for something *actually useful to society* like making said nuclear power safer and cheaper and easier to adopt as a replacement to fossil fuels, or investing in green wind and solar power. MS isn't looking at next gen nuclear power to benefit society - they mean to utilize it for yet another gpu-based boondoggle. In this case, when all is said and done we'll have additional nuclear waste and bad poetry and AI hallucinations and lies to show for it. I do hope the lies don't get people hurt or killed.

@Biggles MSFT adopting small modular nuclear to power their datacenters is a net societal benefit, regardless of whether that energy consumption is powering the training of AI models or some other compute workload.

@darkuncle

How do you figure? Not that MS was going to donate that money/time/effort to housing the homeless or universal healthcare, but what tangible benefit do you or I (or anyone!) get because MS powers it's current boondoggle via nuclear rather than say solar and batteries?

@Biggles they’re not powering it with fossil fuels, for one thing

@darkuncle ok, but - I'm not beating up nuns. Not being actively evil isn't worthy of praise. IMHO - MS is wasting energy on bullshit - the fact that it's carbon-zero is lovely but mostly irrelevant, unless you want to talk about greenwashing. A better carbon-zero solution is to not waste energy on the latest gpu-based-fad.

@Biggles MS is going to be powering datacenters regardless; that they are pursuing a next-gen way of doing so that is clean and sustainable is worth a golf clap, at least.

@Biggles I think there’s a lot more substance here (AI/ML broadly, and generative AI specifically) than you are giving credit for - just because there’s a lot of hype and FOMO doesn’t mean there aren’t legitimate use cases being overshadowed by all the froth.

Follow

@darkuncle

Maybe?

But again - heroin. It's damn handy for the "legitimate use cases". And they used to put it in all sorts of crazy shit. Until they realized how addictive it was. Now it's a controlled substance because it causes more harm than good.

Let's look at LLMs, today. You can chat with it! You can ask it to do your homework! And ... it's wrong (lots - plus it plagiarizes). Well that sucks. Good thing I didn't use it for anything important. Except - some people did. And still do. We are arguably still in the "fuck around" stage. I'm not excited about the "find out" bit.

LLMs today belong in research labs. Not as a " virtual assistant" in things the general public actually uses. We've already seen them ruin careers and offer up fatal advice - see arstechnica.com/information-te - do people actually have to die before we exercise restraint.

@Biggles yeah, we need better guardrails around use cases before we start getting widespread mass adoption and finding out the hard way where the sharp edges are

Sign in to participate in the conversation
CleverLibre Social

CleverLibre Social is an inclusive social instance for open discussion, learning, and community.
All cultures welcome.
Hate speech and harassment strictly forbidden.