Hacker News Reader

Comment by bryanrasmussen | Hacker News Reader

>LLM isn't like calculator or microwave oven or other technology before. (Like OP's article) But it really is.

I would not buy a calculator that hallucinated wrong answers part of the time. Or a microwave oven that told you it grilled the chicken but it didn't and you have to die from Salmonella poisoning.

View on HN

The microwave analogy is good. I still use it, even though it often makes half my food scalding hot while the other half remains fridge cold.

You should set the microwave to much lower power and let it heat for much longer, so that the heat gets to transfer evenly across the mass of the food. It even says so in the instruction manual. If you blast with full power, leave the food for at least 2 minutes after it's heated for the heat to balance out across the food (again, it's in the manual).

or turn the food over, or move it to a different position inside the microwave -- the way microwave works is that it heats up the food unevenly (there's a wave involved).

Yes, my point was that microwaves are advertised as a 'throw your lunch in and get it warm in 1-2 minutes' appliance, but kinda like an LLM, they require some manual effort to do it well (or decently, depending on your standards).

Like:

1- Put it on the edge of the plate, not in the middle

2- Check every X seconds and give it a stir

3- Don't put metal in

4- Don't put sealed things in

5- Adjust time for wetness

6- Probably don't put in dry things? (I believe you needed water for a microwave to really work? Not sure, haven't tried heating a cup of flour or making a caramel in the microwave)

7- Consider that some things heat weirdly, for example bread heats stupid quick and then turns into stone equally as quick once you take it out.

...

I don’t really think repositioning it has a direct effect. An indirect effect of moving it around is that you turn the microwave off for around 30 seconds or more in order to do it. The reason some parts increase in heat faster is that they have higher concentrations of water; allowing the water to stop boiling and all of the heat to spread through is the magic.

(I’ve heard the fans that you hear are there to reflect the micro waves and make them bounce all over the place but I don’t know if that’s true. Regardless, most models have a spinning plate which will constantly reposition the food as it cooks.)

The fan you hear is to keep the microwave generator cool. It's outside the part of the microwave where the microwaves go.

Older microwaves had a fan-like metal stirrer inside the cooking box, that would continuously re-randomize where the waves went. This has been out of fashion for several decades.

> The reason some parts increase in heat faster is that they have higher concentrations of water;

Composition is part of it, but it isn’t the whole story. A microwave oven is a resonant cavity. There are standing electromagnetic waves in there, in several different modes. They have peaks and nulls. That’s why many microwaves have a rotating plate. It physically moves the food relative to the standing waves.

Microwave's can't do an oven's job. They heat the center first and spread out.

If you want a compromose, invest in an air fryer. It has the opposite problem, so cook 80% in an air fryer and let the microwave finish the center.

You would if you were able to do basic mental maths and you learned to engage and run basic sanity checks. That's still much faster than grabbing the slide rule. (And it's not like people are infallible)

Obviously if one product hallucinated and one doesn't it's a no brainer (cough Intel FPUs). But in a world where the only calculators were available hallucinated at the 0.5% level you'd probably have one in your pocket still.

And obviously if the calculator hallucinated at the 90% of the time for a task which could otherwise be automated you'd just use that approach.

I've seen my accountant's fingers flawlessly fly using a calculator to track expenses down to the penny. Few people have those mental skills even in the days before calculators - either mechanical or digital.

Slide rule are good for only a couple of digits of precision. That's why shopkeepers used abacuses not slide rules.

I have a hard time understanding your hypothetical. What does it mean to hallucinate at the 0.5% level? That repeating the same question has a 0.5% chance of giving the wrong answer but otherwise it's precise? In that case you can repeat the calculation a few times to get high certainty. Or that even if you repeat the same calculation 100 times and choose the most frequent response then there's still a 0.5% chance of it being the wrong one?

Or that values can be consistently off by within 0.5% (like you might get from linear interpolation)? In that case you are a bit better than a slide rule for estimating, but not accurate enough for accounting purposes, to name one.

Does this hypothetical calculator handle just plus, minus, multiply, and divide? Or everything that a TI 84 can handle? Or everything that WolframAlpha can handle?

If you had a slide rule and knew how to use it, when would you pay $40/month for that calculator service?

Slide rules were used in astronomy, engineering, and aviation. You could get them more accurate than 2 decimal places.

"A couple" does not always mean two.

https://en.wiktionary.org/wiki/couple - "(informal) a small number"

FWIW, "Maximum accuracy for standard linear slide rules is about three decimal significant digits" - https://en.wikipedia.org/wiki/Slide_rule

While yes, "Astronomical work also required precise computations, and, in 19th-century Germany, a steel slide rule about two meters long was used at one observatory. It had a microscope attached, giving it accuracy to six decimal places" (same Wikipedia page), remember that this thread is about calculating devices one might carry in one's pocket, have on one's self, or otherwise be able to "grab".

(There's a scene in a pre-WWII SF story where the astrogators on a large interstellar FTL spacecraft use a multi-meter long slide rule with a microscope to read the vernier scale. I can't remember the story.)

My experience is that I can easily get two digits, but while I'm close to the full three digits, I rarely achieve it, so I wouldn't say you get three decimal digits from a slide rule of the sort I thought was relevant.

I'm a novice at slide rules, so to double-check I consulted archive.org and found "The slide rule: a practical manual" at https://archive.org/details/sliderulepractic00pickrich/page/...

> With the ordinary slide rule, the accuracy obtainable will largely depend upon the precision of the scale spacings, the length of the rule, the speed of working, and the aptitude of the operator. With the lower scales it is generally assumed that the readings are accurate to within 0.5 per cent. ; but with a smooth-working slide the practised user can work to within 0.25 per cent

That's between 2 and 3 digits. You wouldn't do your bookkeeping with it.

"a couple" always means two. "A few" always means three. That wiki is wrong.

If someone says, there were a couple [of] people there, I would not expect there to have been two, specifically.

Feel free to check some print dictionaries.

New Merriam-Webster dictionary, 1989, def. 4 "an indefinite small number" - https://archive.org/details/newmerriamwebste00spri/page/180/...

Pocket Oxford English dictionary, 2005, def. 3 "(informal) an indefinite small number" - https://archive.org/details/pocketoxfordengl0000unse_p5e4/pa...

The Random House college dictionary, 1975, def. 6, "a couple of, (Informal) a small number of, a few"

> Slide rule are good for only a couple of digits of precision. That's why shopkeepers used abacuses not slide rules.

Shopkeepers did integer math, not decimal. They had no need for a slide rule, an abacus is faster at integer math, a slide rule is used for dealing with real numbers.

Yes ... Isn't that my point? I meant it as an example of how neither "basic mental maths and ... sanity checks" nor is a calculator with a 0.5% error rate are appropriate.

I mean that it's not a difference of accuracy, it's a difference of domain. A calculator & a slide rule have the same domain, an abacus and Mayan Quipu have the same domain. The calculator & abacus are faster than the slide rule & Quipu. Similarly a typewriter has the same accuracy as handwriting a document (worse if a pencil is used since typewriter's can't easily correct mistakes), but typing took over book writing because it was faster, not because it was more accurate.

I lead with the example of my accountant using a calculator to show it has a different domain than a slide rule, or simple mental techniques.

I would buy a calculator that could help me break down the problem and show my work though, that's the hardest part. I can always double check the numbers, and I would get partial credit for a miscalculation with the right process, but if I can't figure out how to represent the problem mathematically, I'm cooked.

Even if it is wrong half the time, and even when it gives you the right answer the work it shows isn't correct?

Just an odd aside that occurred to me: Would you buy a calculator that hallucinates wrong answers part of the time, but gets enough correct answers and "partial credit" to earn you a certificate for being competent in math?

Yes, people would buy it and work around the wrong answers, if getting a non-hallucinating calculator was off the table.

The people who care about certificates, but not quality, are usually pretty bad in their jobs anyway. Nobody cared what certificates I have or not, and I know several people who had to lie about their certificates - that they don’t have some when in reality they have - to get their current jobs. For example, they signal to me that you like to waste your time for performative acts, instead of doing your job. Also, it’s usually pretty bad to work for bosses who substantially value these.

So should AI also indicate this to me? That the job will suck, and there would be bad coworkers around me in the job?

Microwave oven does quite unexpected things when you cook a shelled egg or a dish on metal plate.

We teach our kids about microwave oven safety for this reason.

On the contrary, those things are quite predictable. Once you know those issues exist, you can reliably avoid them. But with LLMs you can't reliably avoid hallucinations. The unreliability is baked into the very nature of the tool.

Except that we know that it does that when you put those things in, so they aren't "quite unexpected".

My grandma did not have a microwave oven because she didn't see the point of it.

I'm in my 40s and don't have a microwave oven because I don't see the point of it... when I lived in a rented apartment, I got gifted one because how could I not have one? I tried it for a few days and just didn't find it useful. When I bought my own apartment and renovated the kitchen, I didn't bother to install one.

The use is heating up single portions of leftovers.

Which can be done in an induction cooker almost as fast, with the result tasting better, and without the need of a specific appliance that takes up considerable space.

I guess, if you have one of those. Vastly more expensive and more involved to install, especially when renting. I’ve never used one because I’ve never been at a place with one.

When I rented I had a standard ceramic hob and still didn't see the point... sure, you gain some time, but it's maybe 5 minutes of unattended time where you can often be doing something else, vs. much worse taste. But I understand that with slow cookers it can make sense for other people. With induction I think it's outright pointless.

As an anecdote, in my country there is a very popular brand of supermarket pizzas, Casa Tarradellas. I never buy them but a friend of mine used to eat them really frequently. So once he shows up at my house with one, and I say OK, I'm going to heat it. I serve it, he tries a bite and is totally blown away. He says "What did you do? I've been eating these pizzas for years and they never taste like this, this is amazing, the best Casa Tarradellas pizza I've ever had".

The answer was that he used the microwave and I had heated it in the regular oven...

> vs. much worse taste

I have never had that issue when heating stuff up. Your pizza example is not reheating (and generally you never want to reheat anything that’s supposed to be crispy in the microwave; though not on the stove top either).

Reheat pizza on a stovetop covered on low heat. Better than the first time it was cooked. Yw

I prefer convection ovens for that.

Ordinary ovens also do that alright. Takes 20-30 minutes instead of 2-3, but that just trains your delayed reward system a little. Also, don't use a plastic container.

They do a lot more things though, which microwaves don't. Pizza, for example, has to be cooked properly, not with a microwave. If I can only have one, I'll take the mini conventional oven.

The best tasting (and also quickest) corn on the cob is done in microwave. Cook the entire unpeeled ear for 5 minutes on high, cut the butt end about 1 inch in and pull out corn out of the husk. Butter, salt, enjoy!

Maybe the real question isn’t whether the microwave is useful, but whether she wanted what it offered. That seems to apply to a lot of tech debates today too.

I don’t own a microwave because I don’t mind the trade offs of other tools that do the same job. But I don’t go around telling people who find microwaves useful that they are bringing about the end of cooking and should feel bad because of it.

My microwave regularly doesn’t cool things as the instructions describe. I’ve learned to pay attention.

Do you use a GPS? That sometimes gets the route wrong, but overall gets you to where you want to go in less traffic than if you didn't use it? And occasionally really delights you with new routes?

(thanks Rory Sutherland for this analogy)

Rarely. I feel lost when I use GPS to get places.

Alec Watson of Technology Connections points out that GPS routing defaults to minimizing time, even when that may not the most logical way to get somewhere.

His commentary, which starts at https://youtu.be/QEJpZjg8GuA?t=1804 , is an example of his larger argument about the complacency of letting automation do things for you.

His example is a Google Maps routing which saves one minute by going a long way to use a fast expressway (plus $1 toll), rather than more direct but slower state routes and surface streets. It optimizes one variable - time - of the many variables which might be important to you - wear&tear, toll costs, and the delight of knowing more about what's going on in the neighborhood.

His makes the point that he is not calling for a return to paper maps, but rather to reject automation complacency, which I'll interpret as letting the GPS figure everything out for you.

We've all heard stories of people depending on their GPS too much then ending up stuck on a forest road, or in a lake, or other place which requires rescue - what's the equivalent failure mode with a calculator?

OK I don't think I'm going to persuade you if you don't use GPS. Buy 95% of the population do.

Pardon? I said I use GPS.

I'm also aware of the failure modes with GPS complacency, including its incomplete knowledge of the variables that I find important.

And that's with something that makes mistakes far less often than LLMs and related technology.

Which is why I don't think that your mention of GPS use is a strong counter-example to bryanrasmussen's comment against using hallucinating devices.

you seemed to have cast off the most important factor in the conversation of "I can check a GPS's work" for some reason.

If you drive into a lake or anything like that it's your own fault not the GPS. It doesn't control the car it just tells you directions. And if you know the area well enough to make judgements like the other things you mentioned, you don't need gps. Gps is specifically for when you don't know where to go.

I use it all the time, pretty much zero issues.

I view this thread as part of bryanrasmussen comment "I would not buy a calculator that hallucinated wrong answers part of the time" with zpeti pointing out that people use a GPS despite how it gets the route sometimes.

I don't know how to respond to your comment in that context.

Do you double-check your calculator all the time, to ensure it's not giving you the wrong answer?

As to Alec Watson's commentary about GPS, how do you know the area well enough to make judgements if you always follow the GPS routing which avoids the neighborhood?

I don't generally keep the entire thread in mind when reading and responding to comments.

If I spend a lot of time in an area I learn it and don't need gps to navigate it, however I might use gps just to find specific addresses as I don't usually memorize every street name. I also usually find that Google maps chooses perfectly sensible routes anyway, I don't see much point in trying to second guess it. Oh maybe I can save a minute or two or save a few kilometers by avoiding a highway, honestly who cares? I certainly don't. It will usually offer multiple route alternatives anyway, your ideal route or something close to it is probably among them.

I don't, I use static maps and aerial photos, and sometimes satellite photos. I also don't have a microwave oven, in part because they are highly unreliable depending on where certain molecules ended up in the bucket in the freezer and so on.

However, I do have a pressure cooker and a rice cooker that gets a lot of use. They're extremely reliable and don't use much electricity and I can schedule what they do, which is bulk cooking without me having to care about it while it happens.

The success rate and failure mode matters. Gps/maps you can look at before driving and confirm it's not completely insane quite easily, and if it takes a suboptimal route you still get to your destination.

If an LLM hallucinates and you don't know better, it can be bad. Hopefully people are double checking things that really matter, but some things are a little harder to fact check.

Wait, do people generally use GPS routes to go places? I find GPS to be great to locate myself when I can't figure it out from landmarks, but I very rarely use it to select a route – I can just look at the map in the same app and figure out a route on my own.

I can't remember a whole route just from looking at the map once. Also maps generally don't show whether streets are one way and things like that.

I just type the address into Google maps, or place a pin manually, then hit the start button. It'll tell me every step of the way. Keep right at the fork. In a hundred meters, turn left. Turn left. Take the second exit in the roundabout. Your destination is a hundred meters ahead on the right.

It's great and it works almost flawlessly. Even better if you have a passenger able to keep an eye on it for those times when it isn't flawless.

> Also maps generally don't show whether streets are one way and things like that.

Citation needed.

I used GPS routing today to cycle to from a station to a place I never been to before. Used it before for driving on highways in a weird place. Its very helfull when you dont know the area very well

It's a mix. If I don't know the area well (like if I'm going downtown), the GPS is a big boon.

If I do know the area, I can simply find the area and then drive straight there. There's still value in my mind of being aware of your environment.

If the GPS routinely hallucinated the existence of places to go, not just occasionally erroneous or out-of-date data, but literally putting streets where they never existed or taking me to the nearest branch of a bank when that never existed, it would be unreliable enough that I would probably only trust maps made the old fashioned way.

Reply on news.ycombinator.com