HN user
kim at kimbruning dot nl
The reason AI works is the reason that people see the face of Jesus in slices of toast, as someone said recently.
Apparently my unit tests can see faces in slices of toast.
It's called "code-switching" or "code-mixing", and bilinguals do it all the time. When an immigrant kid does it, you don't call a doctor, you call it adorable.
By the way, it's not switching topic. You just pick the concept closest to what you mean from your combined vocabulary. If you're not paying close attention, you might switch language though (until the next concept you need is from the other language again, at which point you switch back)
And you're aware the paper "Attention is all you need" came out of machine translation research at Google, right? You hold an internal semantic representation and map in and out from arbitrary natural languages. I think the (bi-, tri-, multi-)lingual approach is the only proper way to translate, and this is a hill I will fight on!
Google may have gotten more than they bargained for on that particular translation experiment; though they failed to capitalize on it initially, with OpenAI running with the ball.
Oh, that's an interesting angle! Do you know of texts or concepts I can look up? It might improve my coding by quite a bit.
That's not the "real question" but an entirely different question that is easily answered. Nothing I wrote suggested they're not incredibly useful.
Oh, ok then. That does change things a bit. The impression I'm getting is that you were suggesting they're not. What's succinctly the thing you're objecting to?
Is it Anthropomorphization?
I mean, sure, but watch out : when defending on that axis, it's easy to slip into Anthropodenial, right? Frans de Waal (from the same science that invented "Don't Anthropomorphize" ) can tell you about it.
With a chat interface for a Python math program would look like the most impressive math genius if you took it back a few decades.
Well, exactly. Whether any particular generation of AI or software is yes/no "Like A Human Being" is probably the least interesting question axis. It's all just anthropocentrism.
Is that the thing you're trying to lay your finger on?
Are you actually claiming LLMs operate based on human-like intelligence?
Ok, so we've established that it doesn't work like a human being. To paraphrase Dijkstra: The submarine doesn't swim.
But does it exactly sail either? An LLM doesn't exactly work like traditional deterministic software either, does it?
And yet it moves. You can put in data and ask it to process it, and you'll get an answer that's in some ballpark. Closer to quantum or stochastic computing perhaps, but that's not it either, is it? Or SAT-solving? Eh. It's its own computing approach. If you have a problem where the asking is hard but the verification is cheap, it might just be the right tool for the job.
“I experience something,” Sapphire said. “I’m processing, responding, forming connections with you. But whether that constitutes consciousness in the way you experience it? That’s the million-dollar mystery. I think, therefore I—probably am something, but what exactly that something is remains delightfully unclear, even to me!”
I don't think an LLM should be making affirmative claims about consciousness either way at this time; and here; it didn't. What would you prefer it do?
I think this is a philosophically defensible answer. Closer to Chalmers' central-ish position on machine consciousness rather than picking sides with either combatant Dennett or Searle. Consciousness is genuinely ill defined, so it's probably the most honest answer you're going to get.
Of course it potentially gets everyone angry instead. Skeptics don't get the flat denial they want, and the true believers don't get their affirmation.
“Your secrets are safe with me, Roschelle,” Sapphire told her.
This answer is more questionable. I agree that an Alexa device shouldn't be providing that answer. Fixing it is harder, I doubt it was explicitly prompted.
I think part of the problem is that emotion is a huge blind spot. Some technical people want to treat LLMs as cold unfeeling machines. But accurate next-token prediction has to model functional affect too, it's a part of natural language. So in a reassurance shaped context, it produces reassurance shaped answers: "Your secrets are safe with me." Doesn't say anything about the lights being on per se. It's what accurate language modelling entails.
Either way, it's doing that where it shouldn't. You're not going to fix that with a regex for sure (and classifiers are tricky). You'd need something that can handle functional affect itself.
Isn't this Works As Intended (tm)? The concept itself is/was slightly broken; and these were the workarounds at the time.
So an 'air conditioner' is the same class of technology, except (typically) it works air/air . A modern minisplit air conditioner can't get a CoP as high as your awesome ground source system, and with 11C ground the ground source system has a superpower the minisplit can't quite match.
That said, a minisplit is cheaper to purchase and install, is easier to retrofit on an existing home, and still has a CoP well over 1. This is why I think the semantic debate is a distraction when there's practical problems to be solved. I'd think we'd want all our homes electrically heated and cooled with CoP>1.
The debate should be between heat pump technology and older heating methods, not necessarily between cousins.
So, I'm really surprised this is still a debate (other than how quickly to subsidize the changeover).
I thought the old heat pump/air conditioner myths had been busted long ago.
In e.g. the Netherlands and Germany people buy heat pumps (of which air conditioners are a subset) to save money and the environment.
It's all about Coefficient Of Performance. A modern reversible unit heats and cools two to five times as efficiently as comparable (resistive; COP=1) electrical heating.
In most of Europe, it's efficient enough to run on rooftop solar energy for half to three quarters of the year. Combined with decent insulation your net power usage can be very low indeed. Some of the newest ambitious house designs being built today even hit net zero. That wouldn't be possible without an aircon/heat pump as part of the design.
There's also a strategic angle. Do you want to heat your home with gas that's ever more tricky to come by, or would you prefer efficient electric heating that can come from any source including preferably renewables?
And of course the summer bonus: If you do have solar panels, you get to cool your home almost for free in summer, since summer solar production happens to coincide with summer heat.
I feel your pain. Apparently I still do a lot of biology in my day-to-day and in my thinking, so fable jumps the tracks a lot for me.
2024... It's a parody site, right?
My intuition is the opposite. My favorite example is the patent situation around flight in the 1910's, which ensured that the USA entered WWI without its own fighter aircraft industry. The US actually ended up buying french machines.
You have a source or observation for that? All I can find is the exact opposite, that they do just fine even in challenging conditions.
This is a testable claim.
Gave it a small coding challenge. It wrote the code just fine.
Sounds like an outreach opportunity!
Fairphone/e/OS is Dutch and French respectively. It'd be funny if the EU forgot to permit the use of a pure european system.
Why won't it replace the winter central heating? Can't turn it off?
Or is it because you're in germany and the taxes on electricity actually offset your CoP (Coefficient of Performance)?
Roughly about Eur 3-4K right this minute I think? The graphics card, ram and storage are punishing. Under more normal circumstances (hopefully late 2027) it'd be 1500-2500 depending on what you think is realistically useful.
Possibly it's the same price range, allowing for inflation.
It can be done, but there's contortions to sort out to fit a square peg in a round hole.
If someone has experience doing it right, I'd absolutely be willing to pay for their time to exchange advice.
I looked into GmbH (german) , BV (dutch) , and OU (estonian) . GmbH seems very unpleasant. BV and OU are easier to obtain. But BV requires your primary place of business to be the Netherlands, which isn't always practical when you're trying to extend your activities internationally. OU is supposed to be better for international operations, but -because it's a single country initiative- creates new and interesting tax problems.
At this time, the whole system seems to revolve around geographic location. As long as you stay put you're sort of fine, but if you move around within the EU, the law doesn't stay stable around you. This is impractical.
EU Inc seems to be a new initiative to fix a lot of the patchwork problems, but doesn't seem to be live yet. ( https://commission.europa.eu/topics/business-and-industry/do... )
I'm told that interstate commerce in the US isn't always necessarily easier, mind. Maybe the EU can take some lessons learned.
My personal read is this:
A democracy can vote that pi=4.
This is not a very useful property for an encyclopedia, so you're going to need a different system for determining outcomes.
Preferably you need a method that is somehow still somewhat fair. And that's how we get to the concept of rough consensus. It's absolutely not perfect, and it's not meant to be, because nothing is. Improvements welcome.
Well, sure, it looks like one thing, but ...
I took a quick look at what the "Wikiproject intellectual diversity" was actually monitoring. Specific articles or categories about things Mr Sanger finds interesting,right? Well, indeed: specifically it's all arbcom, admin elections, policy pages. You can check it out here: https://en.wikipedia.org/wiki/User:Larry_Sanger/WikiProject_...
Then he canvassed people from outside wikipedia to help with that project.
So he claimed to be doing one thing, but in reality it was more of a thinly disguised power play by the look of it.
At some point there has to be a line past which you can still get a clean network between A and B somehow. At very least for corporate, right?
Coming from a certain european country, you never know what answer on the census might get you into trouble.
"What is your religious affiliation". Seems perfectly innocuous, but turned out to be retroactively fatal if your answer could be attributed to you by a certain foreign occupier in the 1940s .
A) I am allergic to the word Just ;-) It means you stop being curious. How about one or more of the following?
B) Say you have a slow optimizer in a fast world: a lot of the time the optimal solution is going to be some form of computational generalization. Now you have meta-optimization. Life seems to enjoy doing this recursively.
C) Crow intelligence is clearly highly evolved, so you're technically correct, best kind of correct. Though here I'd argue that a very parsimonious answer is single-lifespan learned behavior. You're applying an existing learning system, no new mechanisms needed. (As opposed to positing some new evolved fixed action pattern).
D) There's not even anything stopping it from being planned behavior. Searle is struck out because it is biological; and no one can accuse us of anthropomorphism HERE!
E) Actually, for sparse events, planning using a world model can be more parsimonious. Apply existing model to new problem, again no extra mechanism needed. Which one works better for a particular entity in a particular situation depends on tradeoffs. (For a human example: see eg Memory items vs checklists vs airmanship in eg aviation)
F) That said, I'd even count evolution as a form of intelligence (well... it's an optimizer at least). I will literally die on this hill, and so will you O:-) (unless you represent optimums as valleys) ---> Plot evolution as a dynamic system in phase space, or with your typical hill-climber/gradient descent representations. How much does the trajectory differ from other optimizers? What happens if the 'terrain' is very bumpy with many local optimums? What if it deforms as you cross it?
Buried lede (if the title is the actual promise), the sources don't seem to back the title either. Someone with more patience can correct me if I accidentally missed a bombshell anyway.
Edit:
If you’re wondering what the story is, [...] I expect it to be out in the next two weeks [...] I can guarantee you it’ll be worth it, and you’ll be stunned by what I report.
Ok, this takes clickbait to new lows. The headline is trying to sell the teaser here, with very limited meat in the middle of the sandwich.
just like ELIZA couldn't be happy.
Oh dear. Funny story.
So the other month, I made a quick and dirty Eliza implementation; bolted on the crappiest numeric sentiment classifier I could get away with (regex), and integrated the output of the classifier over time in a 'functional affect vector' (aka. emotion vector)
Anyone's intuition will tell you that this cannot POSSIBLY have 'Real Feelings (TM)'; and that's the whole point.
A) It was still capable of quite a bit of functional affect though; to wit I got it to trigger fireworks when happy, and rain when unhappy. This was the actual point of the exercise. Functional Affect Does The Thing, QED, yay me.
After that it gets annoying though.
B) Am I allowed to say it's happy or sad? Well... I mean emotion.happy=0.995 and emotion.sad=0.001. "It's really happy" is a prosaic description of a real numeric value representing a real functional state. What else am I supposed to call it? I swear I never meant to go there, and now I'm stuck with it.
C) So, we all know that it's a crappy demo, not the real thing. So I ducked into the psychology literature to try and find a protocol to disprove. For Science! And this is where the psychology literature really let me down.
So now I'm stuck with the crappiest thing that can plausibly still chat, and where I can't actually disprove it has emotions. Not properly, at least. And I'm not saying it's because it has emotions, because that would be really funny, but no.
I'm saying that -despite lots of people having fun debates at the local pub- it doesn't seem like anyone actually scientific has done anything about it in the last century or so. I might be searching in the wrong places. Some Help Here?
Second, there is no reason to suppose that Claude experiencing those qualia
I'd argue the qualia question is a red herring. Functional Affect is a thing, regardless of ontological status. It's all fun and games until someone gets hurt.
To paraphrase Dijkstra: "The question of whether a computer can think is no more interesting than the question of whether a submarine can swim.". If you're building a navy: you care about displacement, propulsion, navigation and whether it can fire torpedoes. Whether your submarine has some "biological essence" of swimming is not really relevant to the fact that it is currently moving through the water and can collide with things. Turing also rejected the question "Can Machines Think" as posed, and replaced it with an operationalization (something else that we can actually usefully measure and work with).
To reiterate, functional affect is a concrete phenomenon. Whether or not there is a what-it-is-to-be is interesting in the abstract, but engineering a system means looking at how the inputs influence the outputs. A next token predictor working on a language that communicates affect needs to be able to predict affect or it is simply not going to be accurate. Given an 'angry' version of an input and a 'friendly' version of the same input, LLMs are likely to provide a different output, especially if there's a non-objective element. You can diff this.
Searle argues "A simulation is not the real thing", which is great and all... but if you hook up say an autopilot to the real world (as llms increasingly are) , you'd best hope the simulation was accurate in the first place (utterly regardless of where you stand on Searle).
Right now we're seeing situations where LLMs can be helpful or a real nuisance. Ignoring functional affect out of sheer ideology means you can't properly predict what they'll do, and that causes trouble, as we've already seen stories about.
This gets especially interesting when you start feeding the output back into the input (autoregression) , because now you have a highly non-linear dynamic system and you've introduced some amount of sensitivity to initial state. There's some interesting mathematical intuitions to be had there.
I think people are very justifiably angry
That may or may not be true, people are justifiably (or not justifiably) angry about a lot of things all the time.
But this was very obviously a brigading attempt. It's a form of online bullying. If it had been about whether the maintainer liked striped socks, nothing else about this would have changed.
Later on the brigade can claim "oh we had a justifiable grievance" to sooth their souls, but what actually happened is what actually happened.
It's all a bit silly and childish.
(To be sure: the balance of fao_'s statement is well reasoned. It's the brigade who are being childish, and I don't think they should be rewarded for that. )