I still don’t really get this argument/excuse for why it’s acceptable that LLMs hallucinate. These tools are meant to support us, but we end up with two parties who are, as you say, prone to “hallucination” and it becomes a situation of the blind leading the blind. Ideally in these scenarios there’s at least one party with a definitive or deterministic view so the other party (i.e. us) at least has some trust in the information they’re receiving and any decisions they make off the back of it.
For these types of problems (i.e. most problems in the real world), the "definitive or deterministic" isn't really possible. An unreliable party you can throw at the problem from a hundred thousand directions simultaneously and for cheap, is still useful.
"The airplane wing broke and fell off during flight"
"Well humans break their leg too!"
It is just a mindlessly stupid response and a giant category error.
The way an airplane wing and a human limb is not at all the same category.
There is even another layer to this that comparing LLMs to the brain might be wrong because the mereological fallacy is attributing the brain "thinks" vs the person/system as a whole thinks.
You are right that the wing/leg comparison is often lazy rhetoric: we hold engineered systems to different failure standards for good reason.
But you are misusing the mereological fallacy. It does not dismiss LLM/brain comparisons: it actually strengthens them. If the brain does not "think" (the person does), then LLMs do not "think" either. Both are subsystems in larger systems. That is not a category error; it is a structural similarity.
This does not excuse LLM limitations - rimeice's concern about two unreliable parties is valid. But dismissing comparisons as "category errors" without examining which properties are being compared is just as lazy as the wing/leg response.
People, when tasked with a job, often get it right. I've been blessed by working with many great people who really do an amazing job of generally succeeding to get things right -- or at least, right-enough.
But in any line of work: Sometimes people fuck it up. Sometimes, they forget important steps. Sometimes, they're sure they did it one way when instead they did it some other way and fix it themselves. Sometimes, they even say they did the job and did it as-prescribed and actually believe themselves, when they've done neither -- and they're perplexed when they're shown this. They "hallucinate" and do dumb things for reasons that aren't real.
And sometimes, they just make shit up and lie. They know they're lying and they lie anyway, doubling-down over and over again.
Sometimes they even go all spastic and deliberately throw monkey wrenches into the works, just because they feel something that makes them think that this kind of willfully-destructive action benefits them.
All employees suck some of the time. They each have their own issues. And all employees are expensive to hire, and expensive to fire, and expensive to keep going. But some of their outputs are useful, so we employ people anyway. (And we're human; even the very best of us are going to make mistakes.)
LLMs are not so different in this way, as a general construct. They can get things right. They can also make shit up. They can skip steps. The can lie, and double-down on those lies. They hallucinate.
LLMs suck. All of them. They all fucking suck. They aren't even good at sucking, and they persist at doing it anyway.
(But some of their outputs are useful, and LLMs generally cost a lot less to make use of than people do, so here we are.)
I don’t get the comparison. It would be like saying it’s okay if an excel formula gives me different outcomes everytime with the same arguments, sometimes right, but mostly wrong.
As far as I can tell (as someone who worked on the early foundation of this tech at Google for 10 years) making up “shit” then using your force of will to make it true is a huge part of the construction of reality with intelligence.
Will to reality through forecasting possible worlds is one of our two primary functions.
Big community of people who motorbike around the world non-stop. It’s definitely possible to prepare beforehand and actually more admin getting a vehicle through borders.
Biking is faster, you can arrange for all visas for 6 months in advance but not for years. Even for 6 months to have them all approved with no gaps requires either a lot of luck or a very strong passport or both.
Yeah, it used to be that you could get a visa from the local embassy of the country you were currently in. These days, not so much. There are a lot more obstacles to long duration travel now--there are not enough long duration travelers for the system to be set up for it.
Basically yes, the useful models need a modernish GPU to get inference running at a usable speed. You can get smaller parameter models 3b/7b running on older laptops, it just won’t produce output at a useful speed.
I like this. I’ve been working on something similar. Good spec’ing is critical to getting good output and I suspect a lot of the “I’ve already got plan mode” comments are from technical HN folk, who do know the right questions to ask and know what good looks like. But as the success of Lovable shows there are millions of people out there who clearly want to build apps but don’t have the technical chops to do so and clearly don’t know what a good spec looks like. My experience of “plan mode” is that it won’t serve these people. I’d be keen to connect and share ideas around this. My email in my profile.
Loads of similar products out there, but non that did all of: open source code with attested releases, recorded mic and system audio to work with any meeting app and used Apple Intelligence for private summarisation. In beta, and also just released a experimental version with self hosted Ollama support.
Just a shame they spent so long skimping on iPhone memory. The tail-end of support for 4gb and 6gb handsets is going to push that compute barrier pretty low.
Eh, maybe a bit, but those era devices also have much lower memory bandwidth. I suspect that the utility of client models will rule out those devices for other reasons than memory.
Not really? The A11 Bionic chip that shipped with the iPhone X has 3gb of 30gb/s memory. That's plenty fast for small LLMs if they'll fit in memory, it's only ~1/3rd of the M1's memory speed and it only gets faster on the LPDDR5 handsets.
A big part of Apple's chip design philosophy was investing in memory controller hardware to take advantage of the iOS runtime better. They just didn't foresee any technologies beside GC that could potentially inflate memory consumption.
Bots could be crawlers gathering data to periodically be used as raw training data or the requests could just be from a web search agent of some form like ChatGPT finding latest news stories on topic X for example. I don’t know if robots.txt can distinguish between the two types of bot request or whether LLM providers even adhere to either.
I'm undecided on this, initially I was on the “this is bad, we’re outsourcing our thinking” bandwagon, now after using AI for lots of different types of tasks for a while now, I feel like generally I’ve learnt so much, so much more quickly. Would I recall it all without my new crutch? Maybe not, but I may not have learnt it in the first place without it.
Some people benefit from the relaxing effects of a little bit. It helped humanity get through ages of unsafe hygiene by acting as a sanitizer and preservative.
For some people, it is a crutch that inhibits developing safe coping mechanisms for anxiety.
For others it becomes an addiction so severe, they literally risk death if they don't get some due to withdrawal, and death by cirrhosis if they keep up with their consumption. They literally cannot live without it or with it, unless they gradually taper off over days.
My point isn't that AI addiction will kill you, but that what might be beneficial might also become a debilitating mental crutch.
It makes calories cheaper, it’s tasty, and in some circumstances (e.g. endurance sports or backpacking) is materially enhances what an ordinary person can achieve. But if you raise a child on it, to where it’s what they reach for by default, they’re fucked.
It comes down to how you use it, whether you're just getting an answer and moving on, or if you're getting an answer and then increasing your understanding on why that's the correct answer.
I was building a little roguelike-ish sort of game for myself to test my understanding of Raylib. I was using as few external resources as possible outside of the cheatsheet for functions, including avoiding AI initially.
I ran into my first issue when trying to determine line of sight. I was naively simply calculating a line along the grid and tagging cells for vision if they didn't hit a solid object, but this caused very inconsistent sight. I tried a number of things on my own and realized I had to research.
All of the search results I found used Raycasting, but I wanted to see if my original idea had merit, and didn't want to do Raycasting. Finally, I gave up my search and gave copilot a function to fill in, and it used Bresenham's Line Algorithm. It was exactly what I was looking for, and also, taught me why my approach didn't work consistently because there's a small margin of error when calculating a line across a grid that Bresenham accounts for.
Most people, however, won't take interest in why the AI answer might work. So while it can be a great learning tool, it can definitely be used in a brainless sort of way.
This reminds me of my experience using computer-assisted mathematical proof systems, where the computer's proof search pointed me at the Cantor–Schröder–Bernstein theorem, giving me a great deal of insight into the problem I was trying to solve.
That system, of course, doesn't rely on generative AI at all: all contributions to the system are appropriately attributed, etc. I wonder if a similar system could be designed for software?
Idk, depends on the situation. Is he a student trying to show stuff on a resume? Is he a professional trying to sell a product? Is he a researcher trying to report findings? A startup trying to land a pitch?
The value isn't objective and very much depends on end goals.People seem to trounce out the "make games, not engines" without realizing that engine programmers still do exist.
It was just a small personal test of skill with no purpose or stakes. Not even really with intent to make a real game, just a slice of something that resembled a game to see how far I could get without help. Then, once I got as far as I could, research and see how I could do better.
Discussing this in terms of anecdotes of whether people will use these tools to learn, or as mental crutches.. seems to be the wrong framing.
Stepping back - the way fundamental technology gets adopted by populations always has a distribution between those that leverage it as a tool, and those that enjoy it as a luxury.
When the internet blew up, the population of people that consumed web services dwarfed the population of people that became web developers. Before that when the microcomputer revolution was happening, there were once again an order of magnitude more users than developers.
Even old tech - such as written language - has this property. The number of readers dwarfs the number of writers. And even within the set of all "writers", if you were to investigate most text produced, you'd find that the vast majority of it is falls into that long tail of insipid banter, gossip, diaries, fanfiction, grocery lists, overwrought teenage love letters, etc.
The ultimate consequences of this tech will depend on the interplay between those two groups - the tool wielders and the product enjoyers - and how that manifests for this particular technology in this particular set of world circumstances.
> The number of readers dwarfs the number of writers.
That's a great observation!
'Literacy' is defined as the ability to both read and write. People as a rule can write, even if it isn't a novel worth publishing they do have the ability to encode a text on a piece of paper. It's a matter of quality rather than ability (at least, in most developed countries, though even there there are still people who can not read or write).
So think that you could fine-tune that observation to 'there is a limited number of people that provide most of the writings'. Observing for instance Wikipedia or any bookstore would seem to confirm that. If you take HN as your sample base then there too it holds true. If this goes for one of our oldest technologies it should not be surprising that on a forum dedicated to creating businesses and writing the ability to both read and write are taken for granted. But they shouldn't be.
The same goes for any other tech: the number of people using electronics dwarfs the number of circuit designers, the number of people using buildings dwarfs architects and so on, all the way down to food consumption and farmers or fishers.
Effectively this says: 'we tend to specialize' because specialization allows each to do what they are best at. Heinlein's universal person ('specialization is for insects') is an outlier, not the norm, and probably sucks at most of the things they claim to have ability for.
> Heinlein's universal person ('specialization is for insects') is an outlier, not the norm, and probably sucks at most of the things they claim to have ability for.
This is quoted elsewhere in this thread (https://news.ycombinator.com/item?id=45482479). Most of the things are stuff that you will be doing at some point in your life, that are socially expected from every human at part of human life or things you do daily. It also only says you should be able to do it, it does not need to be good; but should the case arise, that you are required to do it, you should be able to deal with it.
>those that leverage it as a tool, and those that enjoy it as a luxury.
Well the current vision right now seems to be for the readers to scroll AI TikTok and for writers to produce AI memes. I'm not sure who really benefits here.
That's my primary problem as of now. It's not necessary used as some luxury tool or some means of entertainment. It's effectively trying to outsource knowledge itself. Using ChatGPT as a Google substitute has consequences to readers, and using it to cut corners for writers has even worse consequences. I don't think we've had tech like this that can argued as dangerous on both sides of the aisle simultaneously.
> I don't think we've had tech like this that can argued as dangerous on both sides of the aisle simultaneously.
On the contrary, all tech is like this. It is just the first time that the knowledge workers producing the tech are directly affected so they see first hand the effects of their labor. That really is the only thing that is different.
I really hate this dismissal of "well it's affecting YOU now so now it's an issue". I'm not just "knowledge worker" and have grown up seeing the dangers of the internet, social media, invasion of privacy, and radicalization through seemingly benign channels. I've witnessed wars, unrest, and oppressions throughout all stages of my life.
So let's not just handwave it as "nothing special" and actually demonstrate why this isn't special. Most other forms of technological progress have shown obvious benefits to producers and consumers. Someone is always harmed in the short term, yes. But society's given them ways to either retire or seek new work if needed. I'm not seeing that here.
> I really hate this dismissal of "well it's affecting YOU now so now it's an issue". I'm not just "knowledge worker" and have grown up seeing the dangers of the internet, social media, invasion of privacy, and radicalization through seemingly benign channels. I've witnessed wars, unrest, and oppressions throughout all stages of my life.
Sorry, but my comment wasn't about you in particular. It was about the tech domain in general. I know absolutely nothing about you so I would not presume to make any statements about you in that sense.
> But society's given them ways to either retire or seek new work if needed. I'm not seeing that here.
No, not really. For the most part they became destitute and at some point they died.
What you are not seeing is that this is the end stage of technological progress, the point at which suddenly a large fraction of the people is superfluous to the people in charge. Historically such excess has been dealt with by wars.
>For the most part they became destitute and at some point they died
Having opportunity doesn't mean they will seize it. I will concede that if you are disrupted and in your 50's (not old enough to retire, and where it becomes difficult to be re-hired unless you're management) you get hit especially hard.
But it's hard to see the current landscape of jobs now and suggest that boomers/older GenX had nothing to fall back on when these things happen. These generations chided millennials and Gen Z for being "too proud to work a grill". Nowadays you're not even getting an interview at McDonald's after submitting hundreds of applications. That's not an environment that let's you "bounce back" after a setback.
>Historically such excess has been dealt with by wars.
Indeed. We seem to be approaching that point, and it's already broken out in several places. When all other channels are exhausted, humans simply seek to overthrow the ones orchestrating their oppression.
In this case that isn't AI. At least not yet. But it's a symptom of how they've done this.
> Having opportunity doesn't mean they will seize it.
Well, in that sense everybody has opportunity. But I know quite a few people who definitely would not survive their line of employment shutting down. A lot of them have invested decades in their careers and have life complications, responsibilities and expenses that stop them from simply 'seizing opportunity'. For them it would be the end of the line, hopefully social security would catch them but if not then I have no idea how they would make it.
Generally speaking, yes. I do understand not everyone has the opportunities for life circumstances beyond their control. So I don't want to belittle that.
But speaking in macroeconomics, most people have the capacity to readjust if needed. I had to do so these last few years (and yes, am thankful I am "able bodied" and have a family/friend network to help me out when at my lowest points). And the market really sucks, but I eventually found some things. Some related to my career, some not.
But I'm 30. In the worst worst cases, I have time and energy to pivot. The opportunities out there are dreadful all around, though.
Right. It doesn't matter how smart you still are if the majority of society turns into Idiocracy. Second, we're all at risk of blind spots in estimating how disciplined we're being about using the shortcut machine the right way. Smart people like me, you, grandparent aren't immune to that.
> the “this is bad, we’re outsourcing our thinking”
> Would I recall it all without my new crutch? Maybe not
This just seems like you’ve shifted your definition of “learning” to no longer include being able to remember things. Like “outsourcing your thinking isn’t bad if you simply expect less from your brain” isn’t a ringing endorsement for language models
That’s the problem, I think: Using AI will make some people stupider overall, it will make other people smarter overall, and it will make many people stupider in some ways and smarter in other ways.
It would have been nice if the author had not overgeneralized so much:
> That’s the problem, I think: Using AI will make some people stupider overall, it will make other people smarter overall, and it will make many people stupider in some ways and smarter in other ways.
And then:
> It would have been nice if the author had not overgeneralized so much
But you just fell into the exact same trap. The effect on any individual is a reflection of that person's ability in many ways and on an individual level it may be all of those things depending on context. That's what is so problematic: you don't know to a fine degree what level of competence you have relative to the AI you are interacting with so for any given level of competence there are things that you will miss when processing an AI's output. The more competent you are the better you are able to use it. But people turn to AI when they are not competent and that is the problem, not that when they are competent they can use it effectively. And despite all of the disclaimers that is exactly the dream that the AI peddlers are selling you. 'Your brain on steroids'. But with the caveat that they don't know anything about your brain other than what can be inferred from your prompts.
A good teacher will be able to spot their own errors, here the pupil is supposed to be continuously on the looking for utter nonsense the teacher utters with great confidence. And the closer it gets to being good at some stuff the more leeway it will get for the nonsense as well.
Cheap, _limitless_ energy from fusion could solve almost every geopolitical/environmental issue we face today. Europe is acutely aware of this at the moment and it's why China and America are investing mega bucks. We will eventually run out of finite energy sources. Even if we do capture the max capacity possible from renewables with 100% efficiency, our energy consumption rates increasing at current rates will eventually exceed this max capacity. Those rates are accelerating. We really have no choice.
There is zero reason to assume that fusion power will ever be the cheapest source of energy. At the very least, you have to deal with a sizeable vacuum chamber, big magnets to control the plasma and massive neutron flux (turning your fusion plant into radioactive waste over time), none of which is cheap.
I'd say limitless energy from fusion plants is about as likely as e-scooters getting replaced by hoverboards. Maybe next millenium.
That date means nothing though. We have yet to figure out how to run a fusion reactor for any meaningful period of time and we haven't figured out how to do it profitably.
Setting a date for when one opens is just a pipe dream, they don't know how to get there yet.
> I like fusion, really. I’ve talked to some of luminaries that work in the field, they’re great people. I love the technology and the physics behind it.
> But fusion as a power source is never going to happen. Not because it can’t, because it won’t. Because no matter how hard you try, it’s always going to cost more than the solutions we already have.
Deepmind are working on solving the plasma control issue at the moment, I suspect they're probably using a bit of AI.... and I wouldn't put it past them to crack it.
This is the thing with AI: We can always come up with a new architecture with different inputs & outputs to solve lots of problems that couldn't be solved before.
People equating AI with other single-problem-solving technologies are clearly not seeing the bigger picture.
Time travel was the most important invention of the 1800s too, but that goes to show how bad resolving the temporal paradox issue is, now that entire history is gone.
but people say that AI will spit out that fusion reactor, ergo AI investment is prior in the ordo investimendi or whatever it would be called (by an AI)
Why would it be too cheap to meter? You're still heating up water and putting it through a turbine. We've been doing that for ages (just different sources of energy for the heating up part) and we still meter energy because these things cost money and need lots of maintenance.
As we have more and more solars, we see rises for being connected to the grid more and more while electricity stays relatively cheap. Fusion won't change that, somebody has to pay for the guy reconnecting cables after a storm
I understand the spray and pray sentiment, but I cannot recommend enough hustling connections at the companies you’re targeting. Channel your inner SDR and stalk the company, the people you’d work with and the best guess at who would be your boss. Clearly don’t be creepy, but connecting on LinkedIn, mentioning you’ve applied, generally seeing what these people are up to is 100% free and you might even spot an event they’re attending that you could “bump in to them”. It’s so hard for someone not to do some sort of follow up if you’ve met them face to face and had a pleasant conversation such as recommending the hiring manager follow up with you etc. Having a direct connection means you can also prod someone every couple of weeks if you haven’t heard anything from your initial application. Most of the time people are embarrassed their company is being slow on processing good applicants. It’s hard but it really makes you stand out from the pure spray and prayers.
I assure you if i had done this, I would not have a job right now. This may have worked in the past, but not in this job market. It's a numbers game now, and when you've only got a few months until you're homeless you do not have the time to mess around.
Honestly, I disagree, my partner just did exactly this. 50+ “cold” applications with a single first round interview. Tracked down where the head of design of one company was going to be at an event, cornered her, chatted for 10mins and was then fast tracked through an interview process and had the job signed 2 weeks later. I know that is a sample of 1, but I’ve spent most of my time as the hirer, going the extra mile to stand out does make a difference even more so in this f’d up AI infected hiring world we’re now in.
Certainly, some companies just aren't hiring today so even knowing the right person probably doesn't help. But my (few) jobs over the past 25 years--including one right after 9/11--came through personal connections and didn't even involve a response to a job posting. In fact, in the last case, the job position was created for me. I'm a bit sorry I never checked to see if there was an associated posting on the jobs site.
But I've certainly been lucky. A few dodged bullets.
again, if i did this i would not have a job right now. This is my lived experience. This is what I had to do. It is not what I wanted to have done. That's why I wrote my post, because someone out there might really be struggling with this job market and need to hear about it.
I'm not saying taking a more personal approach doesn't work. Sure, if you've got an event you can go to or you've got a network you can tap, do that. But neither were options for me. Nobody in my network was hiring, and the one event i could have attended (where relevant people would be) sold out within a few hours.
Fair, proof is in the pudding and you scored a job with your approach. So congrats and 100% not knocking that, hopefully my comment provides a bit of perspective from the other side of the fence.
Your sample size leaves much to be desired. I lean towards the takes that you're indistinguishable AI fluff with no real value. If you could have offered more value and you had more people you had worked with in the past that thought highly of your work you wouldn't have struggled nearly as much. I'm not sure why your former colleagues won't stick their neck out for you. But I do think that reflects more on you than the current hiring environment. It's in vogue to blame others for your own problems right now, so I understand where you're at and what you're struggling with to some degree. But it's not the immigrant's fault.
yes well... it's a story. I was semi-retired before, living off my portfolio. But between the new presidency messing up the market and some family expenses that popped up I was put in an unexpected corner. Life be like that sometimes.
