11 ms·
It's also worth considering that past some threshold, it may be very difficult for us as users to discern which model is better. I don't think thats what's goi
by jablongo 1y ago
It's also worth considering that past some threshold, it may be very difficult for us as users to discern which model is better. I don't think thats what's going on here, but we should be ready for it. For example, if you are an ELO 1000 chess player would you yourself be able to tell if Magnus Carlson or another grandmaster were better by playing them individually? To the extent that our AGI/SI metrics are based on human judgement the cluster effect that they create may be an illusion.
- Wowfunhappy 1y ago> For example, if you are an ELO 1000 chess player would you yourself be able to tell if Magnus Carlson or another grandmaster were better by playing them individually? No, but I wouldn't be able to tell you what the player did wrong in general. By contrast, the shortcomings of today's LLMs seem pretty obvious to me.
- make3 1y agothe argument is for in the future, not now
- runarberg 1y agoThe future had us abandon traditional currency in favor of bitcoin, it had digital artists being able to sell NFTs for their work, it had supersonic jet travel, self driving or even flying cars. It had population centers on the moon, mines on asteroids, fusion power plants, etc. I think large language models have the same future as supersonic jet travel. It’s usefulness will fail to realize, with traditional models being good enough but for a fraction of the price, while some startups keep trying to push this technology but meanwhile consumers keep rejecting it.
- Mackena 1y ago[flagged]
- germandiago 1y agoI also have the intuition that something like this is the most likely outcome.
- eru 1y agoEven if models keep stagnating at roughly the current state of the art (with only minor gains), we are still working through the massive economic changes they will bring. Unlike supersonic passenger jet travel, which is possible and happened, but never had much of an impact on the wider economy, because it never caught on.
- rusk 1y agoCost was what brought supersonic down. Comparatively speaking, it may be the cost/benefit curve that will decide the limit of this generation of technology. It seems to me the stuff we are looking at now is massively subsidised by exuberant private investment. The way these things go, there will come a point where investors want to see a return, and that will be a decider on wether the wheels keep spinning in the data centre. That said, supersonic flight is yet very much a thing in military circles …
- eru 1y agoYes, cost is important. Very important. AI is a bit like railways in the 19th century: once you train the model (= once you put down the track), actually running the inference (= running your trains) is comparatively cheap. Even if the companies later go bankrupt and investors lose interest, the trained models are still there (= the rails stay in place). That was reasonably common in the US: some promising company would get British (and German etc) investors to put up money to lay down tracks. Later the American company would go bust, but the rails stayed in America.
- runarberg 1y agoI think there is a fundamental difference though. In the 19th century when you had a rail line between two places it pretty much established the only means of transport between those places. Unless there was a river or a canal in place, the alternative was pretty much walking (or maybe a horse and a carriage). The large language models are not that much better than a single artist / programmer / technical writer (in fact they are significantly worse) working for a couple of hours. Modern tools do indeed increase the productivity of workers to the extent where AI generated content is not worth it in most (all?) industries (unless you are very cheap; but then maybe your workers will organize against you). If we want to keep the railway analogy, training an AI model in 2025 is like building a railway line in 2025 where there is already a highway, and the highway is already sufficient for the traffic it gets, and won’t require expansion in the foreseeable future.
- ants_everywhere 1y agoI've seen this take a lot, but I don't know why because it's extremely divorced from reality. Demand for AI is insanely high. They can't make chips fast enough to meet customer demand. The energy industry is transforming to try to meet the demand. Whomever is telling you that consumers are rejecting it is lying to you, and you should honestly probably reevaluate where you get your information. Because it's not serving you well.
- refactor_master 1y agoI think the above post has a fair point. Demand for chatbot customer service in various forms is surely "insanely high" - but demand from whom? Because I don't recall any end-user ever asking for it. No, instead it'll be the new calculator that you can use to lazy-draft an email on your 1.5 hour Ryanair economy flight to the South. Both unthinkable luxuries just decades ago, but neither of which have transformed humanity profoundly.
- ants_everywhere 1y agoThis is just the same argument. If you believe demand for AI is low then you should be able to verify that with market data. Currently market data is showing a very high demand for AI. These arguments come down to "thumbs down to AI". If people just said that it would at least be an honest argument. But pretending that consumers don't want LLMs when they're some of the most popular apps in the history of mankind is not a defensible position
- runarberg 1y agoI‘m not sure this works in reverse. If demand is indeed high, you could show that with market data. But if you have marked data e.g. showing high valuation of AI companies, or x many requests over some period, that doesn’t mean necessarily that demand is high. In other words, marked data is necessary but not sufficient to prove your claim. Reasons for market data seemingly showing high demand without there actually being one include: Market manipulation (including marketing campaigns), artificial or inflated demand, forced usage, hype, etc. As an example NFTs, Bitcoin, and supersonic jet travel all had “an insane market data” which seemed at the time to show that there was a huge demand for these things. My prediction is that we are in the early Concord era of supersonic jet travel and Boeing is racing to catch up to the promise of this technology. Except that in an unregulated market such as the current tech market, we have forgone all the safety and security measures and the Concord has made its first passenger flight in 1969 (as opposed to 1976), with tons of fan fare and all flights fully booked months in advance. Note that in the 1960 it was market forecasts had the demand for Concord to build 350 airplanes by 1980, and at the time the first prototypes were flying they had 74 options. Only 20 were every built for passenger flight.
- School-Cotton 1y agoLLMs are already extremely useful today
- xwolfi 1y agoAny sort of argument ?
- School-Cotton 1y agoPersonal experience: I use them.
- ngcazz 1y agoA future where anything has to be paid (but it's crypto) doesn't sound futuristic to me at all.
- alfalfasprout 1y agoActually, chess commentators do this all the time. They have the luxury of consulting with others, and discussing + analyzing freely. Even without the use of an engine.
- komaromy 1y agoAnyone more than ~300 points below the players can only contribute to the discussion in a superficial capacity though
- nopinsight 1y agoAu contraire, AlphaGo made several “counterintuitive” moves that professional Go players thought were mistakes during the play, but turned out to be great strategic moves in hindsight. The (in)ability to recognize a strange move’s brilliance might depend on the complexity of the game. The real world is much more complex than any board game. https://en.m.wikipedia.org/wiki/AlphaGo_versus_Lee_Sedol https://en.m.wikipedia.org/wiki/AlphaGo_versus_Lee_Sedol
- tonyhart7 1y agoI think its doable tbh, if you pour enough resources (smart people,energy,compute power etc) like the entire planet resources of course we can have AGI (damned if we don't) because we put so much, it better works but the problem we cant do that right because its so expensive, AGI is not matter of if but when but even then it always about the cost
- jsrozner 1y agoThere may be philosophical (i.e. fundamental) challenges to AGI. Consider, e.g., Godel's Incompleteness Theorem. Though Scott Aaronson argues this does not matter (see e.g., youtube video, "How Much Math Is Knowable?"). There would also seem to be limits to the computation of potentially chaotic systems. And in general, verifying physical theories has required the carrying out of actual physical experiment. Even if we were to build a fully reasoning model, "pondering" is not always sufficient.
- tsunamifury 1y agoThis is the F1 vs 911 car problem. A 911 is just as fast as an f1 car to 60 (sometimes even faster) but an f1 is better at super high performance envelope above 150 in tight turns. An average driver evaluating both would have a very hard time finding the f1s superior utility
- torginus 1y agoBut he would find both cars lacking when doing regular car things (the F1 moreso than the 911).
- tsunamifury 1y agoFine whatever replace it with a Tesla. Jesus pedantic enough?
- WD-42 1y agoUnless one of them forgets to have a steering wheel, or shifts to reverse when put in neutral. LLMs still make major mistakes, comparing them to sports cars is a bit much.
- tsunamifury 1y agoThis take is extremely ridiculous.
- DoctorOetker 1y agowe could run some tests to first find out if comparative performance tests can be conjured: one can intentionally use a recent and a much older model to figure out if the tests are reliable, and in which domains it is reliable. one can compute a models joint probability for a sequence and compare how likely each model finds the same sequence. we could ask both to start talking about a subject, but alternatingly each can emit a token. look again at how the dumber and smarter models judge the resulting sentence does the smart one tend to pull up the quality of the resulting text, or does it tend to get dragged down more towards the dumber participant? given enough such tests to "identify the dummy vs smart one" and verifying them on common agreement (as an extreme word2vec vs transformer) to assess the quality of the test, regardless of domain. on the assumption that such or similar tests allow us to indicate the smarter one, i.e. assuming we find plenty such tests, we can demand model makers publish open weights so that we can publically verify performance agreements. Another idea is self-consistency tests: a single forward inference of context size say 2048 tokens (just an example) is effectively predicting the conditional 2-gram, 3-gram, 4-gram probabilities on the input tokens. so each output token distribution is predicted on the preceding inputs, so there are 2048 input tokens and 2048 output tokens, the position 1 output token is the predicted token vector (logit vector really) that is estimated to follow the given position 1 input vector, and the position 2 output vector is the prediction following the first 2 input vectors etc. and the last vector is the predicted next token following all the 2048 input tokens. p(t_(i+1) | t_1 =a, t_2=b, ..., t_i=z). But that is just one way the next token can be predicted using the network: another approach would be to use RMAD gradient descent, but keeping model weights fixed, and only considering the last say 512 input vectors as variable, how well did the last 512 predicted forward prediction output vectors match the gradient descent best joint probability output vectors? This could be added as a loss term during training as well, as a form of regularization, which turns it into a kind of Energy Based Model roughly.
- DoctorOetker 1y agoLets call this branch of research unsupervised testing
- torginus 1y agowhich is a thing with humans as well - I had a colleague with certified 150+ IQ, and other than moments of scary smart insight, he was not a superman or anything, he was surprisingly ordinary. Not to bring him down, he was a great guy, but I'd argue many of his good qualities had nothing to do with how smart he was.
- Quarrelsome 1y agoI've always figured that tanglible "intelligence" which leads to more effective decision making is just a better appreciation of one's own stupidity.
- ethbr1 1y ago+1. Being exceptionally intelligent doesn't always catch unknown unknowns. (Sometimes, but not always)
- cgio 1y agoThat would be an extreme criterion for exceptional intelligence, akin to asking for there to be no unknowns.
- ethbr1 1y agoIt's an observation that being smarter in the things you do know isn't everything.
- Quarrelsome 1y agoperhaps the argument is simply that "exceptional intelligence" is just being better at accepting how little you know, and being better at dealing with uncertainty. Both respecting it and attempting to mitigate against it. I find some of the smartest people I know are careful about expressing certainty.
- alluro2 1y agoI'm in the same 150+ group. I really think it doesn't mean much on its own. While I am able to breeze through some things and find some connections sometimes that elude some of the other people, it's not that much different than all the other people doing the same at other occasions. I am still very much average in large majority of every-day activities, held back by childhood experiences, resulting coping mechanisms etc, like we all are. Learning from experience (hopefully not always your own), working well with others, and being able to persevere when things are tough, demotivational or boring, trumps raw intelligence easily, IMO.
- Mackena 1y ago[flagged]
- tbrownaw 1y ago> It's also worth considering that past some threshold, it may be very difficult for us as users to discern which model is better. Even if they've saturated the distinguishable quality for tasks they can both do, I'd expect a gap in what tasks they're able to do.
- ohelno 1y ago> it may be very difficult for us as users to discern which model is better But one thing will stay consistent with LLMs for some time to come: they are programmed to produce output that looks acceptable, but they all unintentionally tend toward deception. You can iterate on that over and over, but there will always be some point where it will fail, and the weight of that failure will only increase as it deceives better. Some things that seemed safe enough: Hindenburg, Titanic, Deepwater Horizon, Chernobyl, Challenger, Fukushima, Boeing 737 MAX.
- robjellinghaus 1y agoHindenburg indeed killed hydrogen blimps. Of everything else on your list, the disaster was in the minority. The space shuttle was the most lethal other item -- there are lots of cruise ships, oil rigs, nuke plants, and jet planes that have not blown up. So what analogy with AI are you trying to make? The straightforward one would be that there will be some toxic and dangerous LLMs (cough Grok cough), but that there will be many others that do their jobs as designed, and that LLMs in general will be a common technology going forward.
- jodrellblank 1y agoDon’t malign the beautiful Zeppelins :( Titanic - people have been boating for two thousand years, and it was run into an iceberg in a place where icebergs were known to be, killing >1500 people. Hindenburg was an aircraft design of the 1920s, very early in flying history, was one of the most famous air disasters and biggest fireballs and still most people survived(!), killing 36. Decades later people were still suggesting sabotage was the cause. It’s not a fair comparison, an early aircraft against a late boat. Its predecessor the Graf Zeppelin[1] was one of the best flying vehicles of its era by safety and miles traveled, look at its achievements compared to aeroplanes of that time period. Nothing at the time could do that and was any other aircraft that safe? If airships had the eighty more years that aeroplanes have put into safety, my guess is that a gondola with hydrogen lift bags dozens of meters above it could be - would be - as safe as a jumbo jet with 60,000 gallons of jet fuel in the wings. Hindenburg killed 36 people 80 years ago, aeroplane crashes have killed 500+ people as recently as 2014. Wasn’t Challenger known to be unsafe? (Feynman inquiry?). And the 737 MAX was Boeing skirting safety regulations to save money. [1] https://en.wikipedia.org/wiki/LZ_127_Graf_Zeppelin https://en.wikipedia.org/wiki/LZ_127_Graf_Zeppelin
- spot5010 1y agoMy guess is that more than the raw capabilities of a model, users would be drawn more to the model's personality. A "better" model would then be one that can closely adopt the nuances that a user likes. This is a largely uninformed guess, let's see if it holds up well with time.
- jv22222 1y agoI think Musk puts it well when he says the ultimate test is can they help improve the real world.
- snthpy 1y agoThat's a great point. Thanks.
- flir 1y ago> For example, if you are an ELO 1000 chess player would you yourself be able to tell if Magnus Carlson or another grandmaster were better by playing them individually? Yes, because I'd get them to play each other?
- cylemons 1y agoHe specifically said play them individually.
- artursapek 1y agoWe’re judging them with benchmarks, not our own intuitions.
- andrepd 1y agoI could certainly tell if they played ??-level blunders, which LLMs do all the time.
- blackkettle 1y agoIt's even more difficult because, while all the benchmarks provide some kind of 'averaged' performance metric for comparison, in my experience most users have pretty specific regular use cases, and pretty specific personal background knowledge. For instance I have a background in ML, 15 years experience in full stack programming, and primarily use LLMs for generating interface prototypes for new product concepts. We use a lot of react and chakraui for that, and I consistently get the best results out of Gemini pro for that. I tried all the available options and settled on that as the best for me and my use case. It's not the best for marketing boilerplate, or probably a million other use cases, but for me, in this particular niche it's clearly the best. Beyond that the benchmarks are irrelevant.
- unsupp0rted 1y agoYou don't have to be even good at chess to be able to tell when a game is won or lost, most of the time. I don't need to understand how the AI made the app I asked for or cured my cancer, but it'll be pretty obvious when the app seems to work and the cancer seems to be gone. I mean, I want to understand how, but I don't need to understand how, in order to benefit from it. Obviously understanding the details would help me evaluate the quality of the solution, but that's an afterthought.