4 ms·
> An Honest Review > I do admit that this approach immediately triggered my contrarian side and made me very defiant of any AI tool. Makes sense. > While thi
by lnenad 2mo ago
> An Honest Review
> I do admit that this approach immediately triggered my contrarian side and made me very defiant of any AI tool.
Makes sense.
> While this could be partially remedied by always asking for a primary source or citation, I dislike the idea that one has to add magical incantations to their queries to get the right results. It’s a good laugh to make fun of “make no mistake” memes, until you start having to consider similar things seriously.
Man I am genuinely stumped at the obvious lack of desire to use something in a way it's supposed to be used. LLMs are tools and like with any tool it's on us to use it properly, not hitting a screw with a hammer and saying that hammers are a very stupid tool.
- Sharlin 2mo agoIt's more like a hammer that never drives nails straight unless you hold it in a specific uncomfortable manner. You shouldn't be required to tell an information retrieval tool to actually retrieve information rather than making it up!
- lnenad 2mo ago> It's more like a hammer that never drives nails straight unless you hold it in a specific uncomfortable manner. I mean sure, pick any analogy, still doesn't change the fact that there are right uses and wrong uses; right ways to use it and wrong ways to use it. > information retrieval tool How is an LLM this? It's a non-deterministic word generator.
- joshribakoff 2mo agoAn LLM is not what the public was “sold”, they were “sold” a personified agent. That is the gap.
- lnenad 2mo agoWe're not talking about "the public". The person writing the article is a very experienced software engineer. There shouldn't be a gap in understanding.
- rglover 2mo ago> You shouldn't be required to tell an information retrieval tool to actually retrieve information rather than making it up! And therein lies the problem: perception. LLMs are treated like information retrieval tools but in reality are probability machines that return plausible/mostly accurate information. They're not searching, they're inferring and then guessing. That the guesses are often quite good means that we can easily fool ourselves with whatever it spits out. We call it hallucination but it's a feature, not a bug.
- K0balt 2mo agoTo be fair, hammers are bad screwdrivers. They are faster at driving screws and require less finesse to do so, but are really bad at taking them out, often break the screw, and often result in a fastening that superficially looks correct but fails to perform well under stress. Two new types of screw, the ring-shank nonhelical fastener and the rivet, however, are changing everything. Soon, using screwdrivers will be rare except in special circumstances.
- lnenad 2mo agoI don't really see your point. You're still using the screwdriver and it's up to you to use it when it makes sense, see if it's doing it's job properly, fix issues it creates etc... If it's not doing a good job for your particular screw don't use it. But don't say it's a horrible tool since you choose to not use it properly. There is a middle ground between CEOisms "LLMs are literally Jesus" and people similar to this blog's author "LLMs are mostly trash."
- joshribakoff 2mo agoPart of the entire premise of the article is that many people aren’t being given a choice, so this is a really weird deflection. I disagree about the author not recognizing a middle ground.
- lnenad 2mo ago> Part of the entire premise of the article is that many people aren’t being given a choice Yeah, "part", but that's not the part I wrote about?
- K0balt 2mo ago[dead]
- duskdozer 2mo agoIt just seems silly to need to beg in English for things like that. What's the entire space of things you need to think about? Can you be sure the specific way you asked would be correct, or do you also have to add "don't lie about following this instruction:"? Why not have an interface `ai --make-no-mistakes --dont-make-things-up`? I don't want to have to talk to the computer like it's people.
- lnenad 2mo agoBut English is literally the interface. It's the *same* as `--make-no-mistakes`, it's just your preference compared to directly telling it to the AI which is ok. That doesn't change the fact the using the tool correctly != using the tool.
- Lwerewolf 2mo agoYou don't have to. Chatgpt-sol-high already does that for me. Could be the extra instructions or the base system prompt or whatever, point is - it already does it.
- inigyou 2mo agoHow are they supposed to be used? The providers promised us we could just say what we wanted and get it.
- righthand 2mo agoNot providing a source is how Llms get caught in sycophancy and hallucinations. The “tool” needs improvement, especially one that claims it’s intelligence. How would you react if you found out after schooling everything you learned has half-hallucinated because your teacher made it up and there is no text book or historical log of facts? You had assumed this whole time you were learning facts…now what?