Agree with this. LLMs have all sorts of abuse cases, but also have real use cases. It’s a shame to miss out on the latter because of anger at the former. I know I’ve found them helpful in learning things and getting a feeling for a field. Just gotta keep your wits about you. - ThreadSky

seanmcarroll.bsky.social • 71 days ago

Agree with this. LLMs have all sorts of abuse cases, but also have real use cases. It’s a shame to miss out on the latter because of anger at the former. I know I’ve found them helpful in learning things and getting a feeling for a field. Just gotta keep your wits about you.

Reposted from Mark Cuban

I have to admit that I've been surprised at the level of hate at AI LLMs.

There is finally a tool that allows those without access to an advanced education to self learn, anywhere , anytime. To get answers to questions, (even if sometimes wrong) , that they would otherwise never have access to

Comments

400trix.bsky.social•71 days ago

Name one that can’t be performed more efficiently by other means.

Just because people you deem smart tell you something doesn’t mean you should believe them.

bweatherson.bsky.social•71 days ago

I’ve found them really helpful with coding. That’s a special case because you can easily check the work. And you need to because they aren’t 100% reliable. But even after checking it’s quicker than I would be, and often they can fix their own mistakes when you tell it what went wrong.

dsquareddigest.bsky.social•71 days ago

I think a lot of brains have been broken by the counterintuitive fact that although it's more difficult to write decent code than adequate prose, it's easier to predictively generate decent code from an internet corpus than adequate prose

alexjs.dev•71 days ago

And it's sometimes better to use generated code. The closer your code is to the widely accepted "answer" to a coding problem, the less time future maintainers will waste distracted by your own idiosyncrasies.

If the test passes, it's probably not worse than human code.

roylonergan.bsky.social•71 days ago

That’s mostly ok for unprofessional fields where there are limited internal/external standards. But I was talking to a senior bank IT guy about it recently and it really hits the juniors’ bonuses if they produce code that they can’t explain to their boss.

bweatherson.bsky.social•71 days ago

Oh I agree. It’s like having an ok undergrad assisting you. That’s pretty good for most of us, but I wouldn’t want that undergrad running my bank’s IT.

mitycheese.bsky.social•71 days ago

It's true they can be useful, but they also provide black box answers. With traditional search, you can evaluate the sources you are sent to. An LLM just tells you "the answer."

mitycheese.bsky.social•71 days ago

But a lot of the backslash is about being forced to use them. MS just raised the price of Office because it now has AI, wasted or not.

The cost/value proposition doesn't work well for a lot of people.

jmwallach.bsky.social•71 days ago

I have used them for about 30 seconds but find the idea of using them for learning about something I am not familiar with dubious given that they have invented case law when lawyers try them.

turbineseaplane.bsky.social•71 days ago

Agree John - I'm concerned about how we get a handle on the actual quality or accuracy or output as we go along. There are way too many cases of really bad output to be putting much faith into it

400trix.bsky.social•71 days ago

You can’t, because accuracy isn’t an axis of LLM performance. All they do is combine tokens into the shape of a response to the prompt. They encode no real knowledge.

turbineseaplane.bsky.social•71 days ago

That doesn't sound great overall 😕

jmwallach.bsky.social•71 days ago

I had one write a sonnet about not fucking Elon even for a horse and it came up with something better than I could...

turbineseaplane.bsky.social•71 days ago

🤣

jctenton.bsky.social•71 days ago

yeah i played with them to see what was happening to my students and they're just constantly wrong. it was about my field so i could spot it, but it made it clear to me that they're useless for learning anything since how would you know when it's spinning bullshit?

aarwood46.bsky.social•71 days ago

I’m not thrilled about who’s programming them and who’s making money off them.
The interests of the owners of these systems are not mine, so why would I trust them?

brandonjitsu.bsky.social•71 days ago

I'd maybe agree with you, but my new laptop came with an "AI" key that replaced right ctrl, and it pisses me off viscerally every time I reach for it by accident.

jcmaas.bsky.social•71 days ago

I suppose it’s similar to any kind of knowledge, you can never take it at face value because of fallibility.

trixie-jones.bsky.social•71 days ago

I use it for brainstorming, not as an authoritative source. I’ve been asking about climate change, how to save our republic, and how to build back better if we can’t. “Regional power centers” (nation-states?) is one possible outcome. China doesn’t want to occupy, just wants trade advantages. 💥

straighttalk317.bsky.social•71 days ago

He’s pivoting from his original statement, which is that an uneducated person with use of an LLM will outperform an educated one at work. That’s the anger source and the whole issue: Mark’s ilk want artificial intelligence to replace workers and couch it like we’re all unfairly fighting progress.

straighttalk317.bsky.social•71 days ago

…he and the investor class fundamentally do not understand LLMs or want to, in my experience working at a start-up as an SME in the field applying the tech. No one wanted to hear the practical applications, they wanted to hear lies and magic—not just my bosses but the people we were pitching too!

straighttalk317.bsky.social•71 days ago

… it was absurd, and when I tried to logically explain why what we were trying to push was wrong, like ethically even (because of the application) I was fired in a hot second.

lassebrasse.bsky.social•71 days ago

Everything new is either negligible or in a superposition of good-bad. LLM:s are not negligible.

moltvikh.bsky.social•71 days ago

Have they made you a better physicist than you would have been without them?

un1crom.bsky.social•71 days ago

yup. if we treat them like any text/book/paper... and work with them socratically it all goes better. it is confusing for many because the "knowledge/data/content" seems to be emanating from the same source as any "computation". Decoupling these "modalities" or being explicit with them helps.

allthingsphysics.bsky.social•71 days ago

Indeed! They’re by no means perfect, but for some situations they are really quite amazing. Why throw out the baby with the bath water?

toddnelson.bsky.social•71 days ago

What’s the real use case for information that’s correct some of the time and wildly inaccurate some of the time? I don’t get it.

lguap0.bsky.social•71 days ago

You just described the internet.

toddnelson.bsky.social•71 days ago

Exactly. Anything trained on it is beyond useless.

lguap0.bsky.social•71 days ago

Not sure I’d say the information on the internet is useless. But you can believe whatever you like.

boolesvoid.bsky.social•71 days ago

Also worth noting that a lot of hate in that thread was directed towardsark Cuban and the billionaire class. THEN, MC made it about LLMs. He's shifting the conversation away from himself.

richardyoung.bsky.social•71 days ago

I do wonder, though, about whether most people have the discipline to treat LLM output as an "80% solution" that *demands* additional work. It seems in the field, people are just... lazy.

zcorbain.bsky.social•71 days ago

LLMs have places where they function well. Abuse will happen, that's human nature. Regardless, AI will always lack imagination and creativity, and the ability to distinguish fact from fiction, but it will infinitely regurgitate derivatives, w/o an exploration of language or ideas.

rickwojcik.bsky.social•71 days ago

One problem with them is that people tend to attribute thinking skills to them. Basically, all they do is extract and summarize a large body of training text. Natural intelligence is grounded in interactions with reality, not just word associations. People need to be wary of trusting them.

neildaviesart.bsky.social•71 days ago

No concerns about the vast amounts of energy and water being consumed by server farms powering all this?

timothyrice.org•71 days ago

It is interesting to me that nobody has these concerns about all the other server farms that exist. I've never heard someone critique the amount of energy that Youtube uses, for instance.

neildaviesart.bsky.social•71 days ago

True, I guess AI is the hot button issue and has brought attention to that sort of thing where it has been lacking.

seanmcarroll.bsky.social•71 days ago

Perfectly reasonable concerns. But it’s possible to consider those aspects separately from the question of what LLMs do is actually useful.

claytonmoraga.bsky.social•71 days ago

Khan Academy is better than anything we will get out of AI. It is reviewed content that doesn't rely on the user being able to discern false or deceptive content.

alexjs.dev•71 days ago

The point stands that Khan Academy misses all the stuff that experts haven't reviewed. I don't see how AI is much different than Wikipedia. As long as you know you shouldn't take it 100% seriously, it's a tool that can be great for learning something you otherwise wouldn't.

alexjs.dev•71 days ago

It's like saying a chef's meal is way better than a poptart. Of course it is, but people still justify buying poptarts.

roylonergan.bsky.social•71 days ago

But do you want a pop-tart in the middle of something you rely on in your business?

alexjs.dev•71 days ago

That sounds more like a management question. If the boss says no AI and I'm hired for different skills, then I'll do that. If the boss just wants it done fast with passing tests, a poptart is probably fine.

roylonergan.bsky.social•71 days ago

Management, quality, and standards. LLMs are fine in a field where all the legacy is crap anyway. But in a regulated business you’re hired to write software that can be maintained by others, won’t fuck up, won’t get you too many fines from regulators, and - especially - won’t damage the franchise.

windcatcher.bsky.social•70 days ago

It's a good thing we investigated fire and discovered how to manipulate it to our advantage in the infancy of our existence. Fire had abuse potential, IIRC (it's hard to recall my formative years prior to inventing the wheel. They're just a blur to me.)

duxinbelltown.bsky.social•71 days ago

I've tried using ChatBots to "discuss" philosophy. What I've noticed is that they are designed to be "flattering" and "agreeable". Presumably to increase engagement. I think it's much better to use Wikipedia and books by the philosophers. These are just as available on the Internet as the ChatBots.

Comments

Posting Rules

Reply