> Yeah, I sometimes use AI for questions like "is it possible to do [x] using library [y] and if so, how?" and have received mostly solid answers.
In my experience most LLMs are going to answer this with some form of "Absolutely!" and then propose a square-peg-into-a-round-hole way to do it that is likely suboptimal vs using a different library that is far more suited to your problem if you didn't guess the right fit library to begin with.
The sycophancy problem is still very real even when the topic is entirely technical.
Gemini is (in my experience) the least likely to lead you astray in these situations but its still a significant problem even there.
IME this has been significantly reduced in newer models like 4.5 Opus and to a lesser extent Sonnet, but agree it's still sort of bad- mainly because the question you're posing is bad.
if you ask a human this the answer can also often be "yes [if we torture the library]", because software development is magic and magic is the realm of imagination.
much better prompt: "is this library designed to solve this problem" or "how can we solve this problem? i am considering using this library to do so, is that realistic?"
> Even before LLMs people were asking Google search questions rather than looking for keyword matches
Google gets some of the blame for this by way of how useless Google search became for doing keyword searches over the years. Keyword searches have been terrible for many years, even if you use all the old tricks like quotations and specific operators.
Even if the reason for this is because non-tech people were already trying to use Google in the way that it thinks it optimized for, I'd argue they could have done a better job keeping things working well with keyword searches by training the user with better UI/UX.
(Though at the end of the day, I subscribe to the theory that Google let search get bad for everyone on purpose because once you have monopoly status you show more ads by having a not-great but better-than-nothing search engine than a great one).
Wholly anecdotal, but as a 52 year old nearly-lifelong caffeine (ab)user I quit this year and the withdrawal period was horrendous -- not for the headaches everyone knows about (they were bad but only lasted a couple of days) but for the somewhat extended depression/anhedonia which I had never really experienced before.
I was worried during that stretch of time that maybe the caffeine had been masking some underlying depression I already had, but a couple of weeks in it passed, so I think my brain just needed to rebalance itself to the new caffeine-free reality.
I'm glad I quit (less anxiety, better sleep, I'm finding it a lot easier to eat healthy while not buzzed on caffeine all the time, and the depressive episode was temporary) but going through that makes it pretty easy for me to believe caffeine can have rather powerful effects in this area.
I think a lot of the difficulty in quitting can be mitigated by slowly titrating down the dose over a month or two instead of quitting cold turkey.
But your experience mirrors mine in going cold turkey which I think demonstrates that caffeine can cause both physical chemical dependence, and psychological addiction.
Speaking as someone who used to buy them regularly to support a PC gaming hobby stretching back to the original glQuake -- GPUs were on average very reasonably priced prior to the crypto boom that preceded the AI boom.
So its technically not AI "ruining everything" here, but there was a nice, long before-time of reasonable pricing.
> AI is just a tool, like most other technologies, it can be used for good and bad.
The same could be said of social media for which I think the aggregate bad has been far greater than the aggregate good (though there has certainly been some good sprinkled in there).
I think the same is likely to be true of "AI" in terms of the negative impact it will have on the humanistic side of people and society over the next decade or so.
However like social media before it I don't know how useful it will be to try to avoid it. We'll all be drastically impacted by it through network effects whether we individually choose to participate or not and practically speaking those of us who still need to participate in society and commerce are going to have to deal with it, though that doesn't mean we have to be happy about it.
A crowd of people continually rooting against their best interests isn't exactly what's needed for the solidarity that people claim is a boon from social media. Its not as bad as other websites out there, but I've see these flags several times on older forums.
It won't be as hard as you think for HN to slip into the very thing they mock Instagram of today for being.
Uh huh, that's always how it starts. "Well you're in the minority, majority prevails".
Yup, story of my life. I have on fact had a dozen different times where I chose not to jump off the cliff with peers. How little I realized back then how rare that quality is.
But you got your answer, feel free to follow the crowd. I already have migrations ready. Again, not my first time.
Virtually everywhere I've ever worked has had an unwritten but widely understood informal policy of placing a multiple on predicted effort for both new code/features and bug fixing to account for Hofstadter's law.
People continued to use the internet during and after the very real ".com" bubble burst.
LLMs can be useful tools in the right situations and the valuations for companies involved with them can be wildly and irrationally inflated at the same time.
Values and creator-side issues aside, YouTube is just awful to use in its natural state just from a user experience perspective.
Way too many ads per minute of content watched, the ads are all extremely low quality and a lot of them are just outright scams these days.
You can solve this to some degree (on some devices) using adblockers, but YouTube has been going out of its way over the past year making this as difficult as possible.
And there are non-ad issues as well, eg. the algorithm absolutely sucks at discovering new content.
If the problem as stated is "Performing an LLM query at newly inflated cost $X is an iffy value proposition because I'm not sure if it will give me a correct answer" then I don't see how "use a tool that keeps generating queries until it gets it right" (which seems like it is basically what you are advocating for) is the solution.
I mean, yeah, the result will be more correct answers than if you just made one-off queries to the LLM, but the costs spiral out of control even faster because the agent is going to be generating more costly queries to reach that answer.
Which isn't the same as saying LLMs and related technology aren't useful... they are.
But as you mentioned the financials don't make sense today, and even worse than that, I'm not sure how they could get the financials to make sense because no player in the space on the software side has a real moat to speak of, and I don't believe its possible to make one.
People have preferences over which LLM does better at job $XYZ, but I don't think the differences would stand up to large price changes. LLM A might feel like its a bit better of a coding model than LLM B, but if LLM A suddenly cost 2x-3x, most people are going to jump to LLM B.
If they manage to price fix and all jump in price, I think the amount of people using them would drop off a cliff.
And I see the ultimate end result years from now (when the corporate LLM providers might, in a normal market, finally start benefiting from a cross section of economies of scale and their own optimizations) being that most people will be able to get by using local models for "free" (sans some relatively small buy-in cost, and whatever electricity they use).
In my experience most LLMs are going to answer this with some form of "Absolutely!" and then propose a square-peg-into-a-round-hole way to do it that is likely suboptimal vs using a different library that is far more suited to your problem if you didn't guess the right fit library to begin with.
The sycophancy problem is still very real even when the topic is entirely technical.
Gemini is (in my experience) the least likely to lead you astray in these situations but its still a significant problem even there.
reply