Live data from Hacker News

I genuinely don't understand why some people are still bullish about LLMs

twitter.com

501–510 of 1001 posts

Re: I genuinely don't understand why some people are still bullish about LLMs

#501

Earlier quoted context omitted.

That’s fair but it’s totally different use cases than the linked post discusses.

The click baited title is “I genuinely don’t understand how some people are still bullish about LLM”. I guess the author can understand now? When something was impossible only 3 years ago, barely worked 2 years ago, but works well now, there are very good reasons to be bullish, I suppose? The hypes cut both way.

> When something was impossible only 3 years ago, barely worked 2 years ago, but works well now

Are you talking of what exactly? What are you stating works well now and did not years ago? Claude as a milestone of code writing?

Also in that case, if there are current apparent successes coming from a realm of tentative responses, we would need proof that the unreliable has become reliable. The observer will say "they were tentative before, they often look tentative now, why should we think they will pass the threshold to a radical change".

Re: I genuinely don't understand why some people are still bullish about LLMs

#502

I get so confused on this. I play around, test, and mess with LLMs all the time and they are miraculous. Just amazing, doing things we dreamed about for decades. I mean, I can ask for obscure things with subtle nuance where I misspell words and mess up my question and it figures it out. It talks to me like a person. It generates really cool images. It helps me write code. And just tons of other stuff that astounds me…

> Wah, it can't write code like a Senior engineer with 20 years of experience! No, that's not my problem with it. My problem with it is that inbuilt into the models of all LLMs is that they'll fabricate a lot. What's worse, people are treating them as authoritative. Sure, sometimes it produces useful code. And often, it'll simply call the "doTheHardPart()" method. I've even caught it literally writing the wrong algor…

People fabricated a lot too. Yesterday I spent far less time fixing issues in the far more complex and larger changes Claude Code managed to churn out than what the junior developer I worked with needed. Sometimes it's the reverse. But with my time factored in, working with Claude Code is generally more productive for me than working with a junior. The only reason I still work with a junior dev is as an investment into teaching him.

Claude is cheaper, faster, produces better code.

Re: I genuinely don't understand why some people are still bullish about LLMs

#503
LLMs are an incredible technological breakthrough which even their creators are using completely incorrectly - in my view, half because they want to make a lot of money, and half because they are enchanted by the allure of their own achivement and are desperate to generalize it. The ability to generate human-style language, art, and other media dynamically on demand, based on a prompt communicated in natural language, is an astonishing feat. It's immensely useful on its own.

But even its creators, who acknowledge it is not AGI, are trying to use it as if it were. They want to sell you LLMs as "AI" writ large, that is, they want you to use it as your research assistant, your secretary, your lawyer, your doctor, and so on and so forth. LLMs on their own simply cannot do those tasks. They is great for other uses: troubleshooting, assisting with creativity and ideation, prototyping concepts of the same, and correlating lots of information, so long as a human then verifies the results.

LLMs right now are flour, sugar, and salt, mixed in a bowl and sold as a cake. Because they have no reasoning capability, only rote generation via prediction, LLMs cannot process contextual information the way required for them to be trustworthy or reliable for the tasks people are trying to use them for. No amount of creative prompting can resolve this totally. (I'll note that I just read the recent Anthropic paper, which uses terms like "AI biology" and "concept" to imply that the AI has reasoning capacity - but I think these are misused terms. An LLM's "concept" of something bears no referent to the real world, only a set of weights to other related concepts.)

What LLMs need is some sort of intelligent data store, tuned for their intended purpose, that can generate programmatic answers for the LLMs to decipher and present. Even then, their tendency to hallucinate makes things tough - they might imagine the user requested something they didn't, for instance. I don't have a clear solution to this problem. I suspect whoever does will have solved a much bigger, more complex than the already massive one that LLMs have solved, and if they are able to do so, will have brought us much much closer to AGI.

I am tired of seeing every company under the sun claim otherwise to make a buck.

Re: I genuinely don't understand why some people are still bullish about LLMs

#504

Earlier quoted context omitted.

> What's worse, people are treating them as authoritative. … I've both seen online and heard people quote LLM output as if it were authoritative. Thats not an LLM problem. But indeed quite bothersome. Dont tell me what Chatgpt told you. Tell me what you know. Maybe you got it from ChatGPT and verified it. Great. But my jaw kind of drops when people cite an LLM and just assume it’s correct.

Is it worse or better than quoting a Twitter comment and taking that as it were authoritative?

It remains proportional to the earned prestige of the speaker.

Re: I genuinely don't understand why some people are still bullish about LLMs

#505

Earlier quoted context omitted.

> inbuilt into the models of all LLMs is that they'll fabricate a lot. Still the elephant in the room. We need an AI technology that can output "don't know" when appropriate. How's that coming along?

How many companies train on data that contains 'i don't know' responses. Have you ever talked with a toddler / young child? You need to explicitly teach children to not bull shit. At least I needed to teach mine.

AI is not a toddler. It's not human. It fails in ways that are not well understood and sometimes in an unpredictable manner.

Re: I genuinely don't understand why some people are still bullish about LLMs

#506

She's a scientist. Most of the people on here are writing software which is essentially reinventing the wheel over and over. Of course you have a different experience of LLMs.

"Reinventing the wheel over and over" has been a trillion dollar business. Maybe it won't be for long.

Is that "bullish on LLMs" or not?

Re: I genuinely don't understand why some people are still bullish about LLMs

#507

Earlier quoted context omitted.

Tesla is valued based on the hope that it'll be the first to full self-driving cars. I don't think stock markets need to make sense, you invest in things that if true, could have huge growth, that's why LLM is being invested in, because alternatives will make you some ROI, but if LLM do break through major disruption in even a handful of large markets, your ROI will be huge.

Waymo is already full self-driving cars, doesn't look like people buy google stocks on this sentiment. I personally think Tesla is positioned well for EV first world compared to other brands. But Chinese companies are catching up.

"Waymo is already full self-driving cars"

they are neither FSD nor having cars at the moment

Re: I genuinely don't understand why some people are still bullish about LLMs

#508

I get so confused on this. I play around, test, and mess with LLMs all the time and they are miraculous. Just amazing, doing things we dreamed about for decades. I mean, I can ask for obscure things with subtle nuance where I misspell words and mess up my question and it figures it out. It talks to me like a person. It generates really cool images. It helps me write code. And just tons of other stuff that astounds me…

> Wah, it can't write code like a Senior engineer with 20 years of experience! No, that's not my problem with it. My problem with it is that inbuilt into the models of all LLMs is that they'll fabricate a lot. What's worse, people are treating them as authoritative. Sure, sometimes it produces useful code. And often, it'll simply call the "doTheHardPart()" method. I've even caught it literally writing the wrong algor…

I was part of preparing an offer a few weeks ago. The customer prepared a lot of documents for us - maybe 100 pages on total. Boss insisted on using chatgpt to summarize this stuff and read only the summary. I did a loner, slower, reading and cought on some topics chatgpt outright dropped. Our offer was based on the summary - and fell through because we missed these nuances. But hey, boss did not read as much as previously...

Re: I genuinely don't understand why some people are still bullish about LLMs

#509
A bunch of comments here seem to be missing a point : the author is (at least was ?) a scientist.

Her primarily work interest is in the truth, not the statically plausible.

Her point is that using LLM to generate truth is pointless, and that people should stop advertising llms as "intelligent", since, to a scientist, being "intelligent" and being "dead wrong" are polar opposite.

Other use cases have feedback loops - it does not matter so much if Claude spuits wrong code, provided you have a compiler and automated tests.

Scientists _are_ acting as compilers to check truth. And they rely on truths compiled by other scientists, just like your programs rely on code written by other people.

What if I tell you that, from now on, any third party library that you call will _stastically_ work 76% of the time, and I have no clue what it does is the remaining X % ?(I don't know what X is, I haven't hast chatgpt yet.)

In the meantime, I still have to see a headline "AI X discovered life-changing new Y on its own" (the closest thing I know of is alpha fold, which I both know is apparently "changing the world of scientists", and yet feel has "not changed the world of your average joe, so far" - emphasis on the "so far" ) ; but I've already seen at least one headline of "dumb mistake made because an AI hallucinated".

I suppose we have to hope the trend will revert at some point ? Hope, on a Friday...

Re: I genuinely don't understand why some people are still bullish about LLMs

#510
post #254

Earlier quoted context omitted.

Remember the quote that IBM thought there would be a total market for maybe 10 or 15 computer computers in the entire world? They were giant, and expensive, and very limited in application.

A popular myth, it seems to be made-up from a way-less-interesting statement about a single specific model of computer during a 1953 stockholder meeting: > IBM had developed a paper plan for such a machine and took this paper plan across the country to some 20 concerns that we thought could use such a machine. I would like to tell you that the [IBM 701] machine rents for between $12,000 and $18,000 a month, so it was…

Did you really think that people on HN, who I assume are very informed, wouldn't fall for popular disbelief? How can this happen?
Post reply on HN