4 ms·

To me it's a useful reminder that those tools are nothing but text generating algorithms, optimised to produce compelling answers, irrespective whether they are

by ezst 1y ago

To me it's a useful reminder that those tools are nothing but text generating algorithms, optimised to produce compelling answers, irrespective whether they are truthful or not, having no concept of what's factual, and completely missing the ability to give up when asked for impossible or unreasonable answers outside of their training data. In essence, they are only adequate in niche situations (like creative writing, marketing, placeholder during iterative design, …) where there's no such social contract and assumption that people operate in good faith and do their best diligence not to deceive others. Pretending otherwise, not pushing back when LLMs are clearly used outside of those contexts, or dressing them into what they are not (thinking machines, search engines, knowledge archives, …) is doing the work of useful idiots defending tech oligarchs and data thieves against their own interests. And yeah, I get it, naysayers are annoying. Doesn't mean they are wrong or their voices shouldn't be heard at a time where the legality and ethics of all this are being debated.

infecto 1y ago
I tried reading your comment but it was a little difficult, there was a lot of hints of ChatGPT fluff. Ultimately saying a LLM is only useful in niche situations is a tell of the boat passing you by. I am not advocating that they solve all problems but they are definitely more useful than simply “niche” workflows.
- ezst 1y ago
  > there was a lot of hints of ChatGPT fluff. What does that mean? Are you implying that I used a LLM to formulate my comment? A comment about the dangers of LLMs for civic and respectful discourse? > Ultimately saying a LLM is only useful in niche situations is a fact. LLMs are not something new or obscure, we have plenty of open-source/weights implementations keeping in check the outlandish marketing promises of the big players and their "SOTA" models. > …is a tell of the boat passing you by. is an opinion, giving an emotional twist to something that really doesn't deserve one. All it shows is that you bought into LLMs being more than what they really are, and welcome the (admittedly cool and tempting) marketing above the truth and your critical and independent thinking on the subject.
  - infecto 1y ago
    You’re stating opinion as fact while accusing others of being emotional. Ironically, your own comment reads like a reaction to hype, not a sober assessment of capability. Saying LLMs are “only useful in niche situations” is not a fact—it’s a perspective, just like mine. We’re all operating off personal anecdotes and incomplete information. The difference is I acknowledge that. We can debate usefulness all day, but reducing the opposing view to “you fell for marketing” isn’t critical thinking. And yes your reply unfortunately read exactly like ChatGPT would respond which I found both amusing but also hard to read as the logic was overly verbose.
    - ezst 1y ago
      > We’re all operating off personal anecdotes and incomplete information. The difference is I acknowledge that. I'm sorry but I can't assume that your comment comes in good faith. You are the one making the exceptional claims here, not me. The burden of the proof falls on you. From my side, I've provided clear use-cases where LLMs are adequate, and large categories of use-cases where they are provably not (misguided usage in those by a number of people is no counter-argument). The extent of your argumentation is annoyingly nihilist "we don't know everything about LLMs, so I'm entitled to believe anything I want". > We can debate usefulness all day, but reducing the opposing view to “you fell for marketing” isn’t critical thinking. The reason why I'm pulling marketing into this is because that's the reason why people generally believe that LLMs are more capable than they really are. Placarding LLMs everywhere and over-selling their abilities is a coordinated effort by a very few immensely rich tech companies. Not by Machine Learning scientists or Computational Neuroscientists who have been mapping this same space for decades. > And yes your reply unfortunately read exactly like ChatGPT would respond Then I suppose it's a compliment, taken as someone whose native language obviously isn't English.
      - infecto 1y ago
        Still struggling to find anything in your wall of words that actually backs up your claims. As I always say—YMMV. On my side, we’ve seen real areas of value from LLMs. Maybe in your field or experience, that value doesn’t exist. Fair enough. But presenting your observations as objective fact is a stretch. Nice chatting, but it’s clear you’re out of productive things to say.
Brendinooo 1y ago
>In essence, they are only adequate in niche situations (like creative writing, marketing, placeholder during iterative design, …) Placeholders, sure. But I'm surprised you don't know about the utility it has in a programming context. I was kicking the tires on Claude for a personal thing the other day - "given this data and this output I want, write a utility function" - and it quickly gave me something that gave me exactly what I needed on the first try. I also had an extended session recently where it really helped me with the "naming is hard" problem. Naming is hard and it's nice to have a rubber duck that has a certain understanding of your code and is able to throw a bunch of ideas at you quickly. I could go on, but hopefully you get the point. And I'll emphasize your "placeholder" point: sometimes it's not even about the solution it gives, it's about the process of articulating the problem and getting something out there. A lot of times the first iteration of anything is the hardest to complete, and it can really cut down the time to achieve that.
- ezst 1y ago
  As I said, I don't reject LLMs for every situation (what you describe would probably fit the "iterative design" category). But as soon as this personal affair meets the wider world, inevitably becomes too big and transient to be thoroughly understood, and accelerates the rate of minimally-supervised code reviews sent to very human peers ; is where I have a bigger issue: individual convenience shouldn't incur more work and effort to others, nor become an increased risk to stakeholders. It is in that sense that I see LLMs challenge "social contracts": their use is infectious (probably also in the GPL-sense, be warned) and should be agreed upon by all parties upfront (do you want to be on the receiving end of a software that you pay good money to have developed while knowing that its developers have little understanding of its workings? How many times cheaper is cheap enough to pay-off the amount of technical and institutional debt/risk? Can LLMs effectively become that cheap?).

by ezst 1y ago

infecto 1y ago

I tried reading your comment but it was a little difficult, there was a lot of hints of ChatGPT fluff. Ultimately saying a LLM is only useful in niche situations is a tell of the boat passing you by. I am not advocating that they solve all problems but they are definitely more useful than simply “niche” workflows.

ezst 1y ago
> there was a lot of hints of ChatGPT fluff. What does that mean? Are you implying that I used a LLM to formulate my comment? A comment about the dangers of LLMs for civic and respectful discourse? > Ultimately saying a LLM is only useful in niche situations is a fact. LLMs are not something new or obscure, we have plenty of open-source/weights implementations keeping in check the outlandish marketing promises of the big players and their "SOTA" models. > …is a tell of the boat passing you by. is an opinion, giving an emotional twist to something that really doesn't deserve one. All it shows is that you bought into LLMs being more than what they really are, and welcome the (admittedly cool and tempting) marketing above the truth and your critical and independent thinking on the subject.
- infecto 1y ago
  You’re stating opinion as fact while accusing others of being emotional. Ironically, your own comment reads like a reaction to hype, not a sober assessment of capability. Saying LLMs are “only useful in niche situations” is not a fact—it’s a perspective, just like mine. We’re all operating off personal anecdotes and incomplete information. The difference is I acknowledge that. We can debate usefulness all day, but reducing the opposing view to “you fell for marketing” isn’t critical thinking. And yes your reply unfortunately read exactly like ChatGPT would respond which I found both amusing but also hard to read as the logic was overly verbose.
  - ezst 1y ago
    > We’re all operating off personal anecdotes and incomplete information. The difference is I acknowledge that. I'm sorry but I can't assume that your comment comes in good faith. You are the one making the exceptional claims here, not me. The burden of the proof falls on you. From my side, I've provided clear use-cases where LLMs are adequate, and large categories of use-cases where they are provably not (misguided usage in those by a number of people is no counter-argument). The extent of your argumentation is annoyingly nihilist "we don't know everything about LLMs, so I'm entitled to believe anything I want". > We can debate usefulness all day, but reducing the opposing view to “you fell for marketing” isn’t critical thinking. The reason why I'm pulling marketing into this is because that's the reason why people generally believe that LLMs are more capable than they really are. Placarding LLMs everywhere and over-selling their abilities is a coordinated effort by a very few immensely rich tech companies. Not by Machine Learning scientists or Computational Neuroscientists who have been mapping this same space for decades. > And yes your reply unfortunately read exactly like ChatGPT would respond Then I suppose it's a compliment, taken as someone whose native language obviously isn't English.
    - infecto 1y ago
      Still struggling to find anything in your wall of words that actually backs up your claims. As I always say—YMMV. On my side, we’ve seen real areas of value from LLMs. Maybe in your field or experience, that value doesn’t exist. Fair enough. But presenting your observations as objective fact is a stretch. Nice chatting, but it’s clear you’re out of productive things to say.

Brendinooo 1y ago

>In essence, they are only adequate in niche situations (like creative writing, marketing, placeholder during iterative design, …) Placeholders, sure. But I'm surprised you don't know about the utility it has in a programming context. I was kicking the tires on Claude for a personal thing the other day - "given this data and this output I want, write a utility function" - and it quickly gave me something that gave me exactly what I needed on the first try. I also had an extended session recently where it really helped me with the "naming is hard" problem. Naming is hard and it's nice to have a rubber duck that has a certain understanding of your code and is able to throw a bunch of ideas at you quickly. I could go on, but hopefully you get the point. And I'll emphasize your "placeholder" point: sometimes it's not even about the solution it gives, it's about the process of articulating the problem and getting something out there. A lot of times the first iteration of anything is the hardest to complete, and it can really cut down the time to achieve that.

ezst 1y ago
As I said, I don't reject LLMs for every situation (what you describe would probably fit the "iterative design" category). But as soon as this personal affair meets the wider world, inevitably becomes too big and transient to be thoroughly understood, and accelerates the rate of minimally-supervised code reviews sent to very human peers ; is where I have a bigger issue: individual convenience shouldn't incur more work and effort to others, nor become an increased risk to stakeholders. It is in that sense that I see LLMs challenge "social contracts": their use is infectious (probably also in the GPL-sense, be warned) and should be agreed upon by all parties upfront (do you want to be on the receiving end of a software that you pay good money to have developed while knowing that its developers have little understanding of its workings? How many times cheaper is cheap enough to pay-off the amount of technical and institutional debt/risk? Can LLMs effectively become that cheap?).