GPT-3

Article

GPT-3 is a recurring concept in the Astral Codex Ten archive, appearing 19 times across 19 issues between May 20, 2021 and February 12, 2026. The archive places it in contexts such as “which is basically GPT-3 for music”; “GPT-3 writes an Ace Attorney case”; “OpenAI, the people responsible for GPT-3 and other marvels”. It most often appears alongside OpenAI, GPT-2, Eliezer Yudkowsky.

Metadata

Category: Concepts
Mention count: 19
Issue count: 19
First seen: May 20, 2021
Last seen: February 12, 2026

Appears In

- OpenAI (10 shared issues)
- GPT-2 (8 shared issues)
- Eliezer Yudkowsky (7 shared issues)
- GPT-4 (6 shared issues)
- GPT (5 shared issues)
- Paul Christiano (5 shared issues)
- AGI (4 shared issues)
- DeepMind (4 shared issues)
- Elon Musk (4 shared issues)
- Google (4 shared issues)
- AI (3 shared issues)
- Ajeya (3 shared issues)

External Links

Source Context

Recovered passages from the original issue text. When the raw archive preserved outbound links inside the source passage, they are listed directly under the quote.

Links For May

May 20, 2021 · Original source

4: I’m very late here, but you might still enjoy OpenAI’s Jukebox, which is basically GPT-3 for music. Train it on Elvis, then make it write new songs on his style. Or feed it the first few verses of Never Gonna Give You Up and make it guess what the rest of the song sounds like. Or just have Celine Dion sing a song about being a music generation algorithm produced by OpenAI.

Inline links: Jukebox, new songs on his style, make it guess what the rest of the song sounds like., sing a song about being a music generation algorithm produced by OpenAI

Links For July

July 23, 2021 · Original source

4: GPT-3 writes an Ace Attorney case, self-recommending:

Practically-A-Book Review: Yudkowsky Contra Ngo On Agents

January 19, 2022 · Original source

The story thus far: AI safety, which started as the hobbyhorse of a few weird transhumanists in the early 2000s, has grown into a medium-sized respectable field. OpenAI, the people responsible for GPT-3 and other marvels, have a safety team. So do DeepMind, the people responsible for AlphaGo, AlphaFold, and AlphaWorldConquest (last one as yet unreleased). So do Stanford, Cambridge, UC Berkeley, etc, etc. Thanks to donations from people like Elon Musk and Dustin Moskowitz, everyone involved is contentedly flush with cash. They all report making slow but encouraging progress.

I found it helpful to consider the following hypothetical: suppose (I imagine Richard saying) you tried to get GPT-∞ - which is exactly like GPT-3 in every way except infinitely good at its job - to solve AI alignment through the following clever hack. You prompted it with "This is the text of a paper which completely solved the AI alignment problem: ___ " and then saw what paper it wrote. Since it’s infinitely good at writing to a prompt, it should complete this prompt with the genuine text of such a paper. A successful pivotal action! And surely GPT, a well-understood text prediction tool AI, couldn't have a malevolent agent lurking inside it, right?

Biological Anchors: A Trick That Might Or Might Not Work

February 23, 2022 · Original source

It looks like this (source) So why don’t we have AI yet? Why don’t we have ten AIs? In the modern paradigm of machine learning, it takes very big computers to train relatively small end-product AIs. If you tried to train GPT-3 on the same kind of medium-sized computers you run it on, it would take between tens and hundreds of years. Instead, you train GPT-3 on giant supercomputers like the ones above, get results in a few months, then run it on medium-sized computers, maybe ~10x better than the average desktop. But our hypothetical future human-level AI is 10^16 FLOP/S in inference mode. It needs to run on a giant supercomputer like the one in the picture. Nothing we have now could even begin to train it. There’s no direct and obvious way to convert inference requirements to training requirements. Ajeya tries assuming that each parameter will contribute about 10 FLOPs, which would mean the model would have about 10^15 parameters (GPT-3 has about 10^11 parameters). Finally, she uses some empirical scaling laws derived from looking at past machine learning projects to estimate that training 10^15 parameters would require H*10^30 FLOPs, where H represents the model’s “horizon”. If I understand this correctly, “horizon” is a reinforcement learning concept: how long does it take to learn how much reward you got for something? If you’re playing a slot machine, the answer is one second. If you’re starting a company, the answer might be ten years. So what horizon do you need for human level AI? Who knows? It probably depends on what human-level task you want the AI to do, plus how well an AI can learn to do that task from things less complex than the entire task. If writing a good book is mostly about learning to write good sentence and then stringing them together, a book-writing AI can get away with a short horizon. If nothing short of writing an entire book and then evaluating it to see whether it is good or bad can possibly teach you book-writing, the AI will need a long time horizon. Ajeya doesn’t claim to have a great answer for this, and considers three models: horizons of a few minutes, a few hours, and a few years. Each step up adds another three orders of magnitude, so she ends up with three estimates of 10^30, 10^33, and 10^36 FLOPs. (for reference, the lowest training estimate - 10^30 - would take the supercomputer pictured above 300,000 years to complete; the highest, 300 billion.) Or What If We Ignore All Of That And Do Something Else? This is piling a lot of assumptions atop each other, so Ajeya tries three other methods of figuring out how hard this training task is. Humans seem to be human-level AIs. How much training do we need? You can analogize our childhood to an AI’s training period. We receive a stream of sense-data. We start out flailing kind of randomly. Some of what we do gets rewarded. Some of what we do gets punished. Eventually our behavior becomes more sophisticated. We subject our new behavior to reward or punishment, fine-tune it further. Rent asks us: how do you measure the life of a woman or man? It answers: “in daylights, in sunsets, in midnights, in cups of coffee; in inches, in miles, in laughter, in strife.” But you can also measure in floating point operations, in which case the answer is about 10^24. This is actually trivial: multiply the 10^15 FLOP/S of the human brain by the ~10^9 seconds of childhood and adolescence. This new estimate of 10^24 is much lower than our neural net estimate of 10^30 - 10^36 above. In fact, it’s only a hair above the amount it took to train GPT-3! If human-level AI was this easy, we should have hit it by accident sometime in the process of making a GPT-4 prototype. Since OpenAI hasn’t mentioned this, probably it’s harder than this and we’re missing something. Probably we’re missing that humans aren’t blank slates. We don’t start at zero and then only use our childhood to train us further. The very structure of our brain encodes certain assumptions about what kinds of data we should be looking out for and how we should use it. Our training data isn’t just what we observed during childhood, it’s everything that any of our ancestors observed during evolution. How many floating-point operations is the evolutionary process? Ajeya estimates 10^41. I can’t believe I’m writing this. I can’t believe someone actually estimated the number of floating point operations involved in jellyfish rising out of the primordial ooze and eventually becoming fish and lizards and mammals and so on all the way to the Ascent of Man. Still, the idea is simple. You estimate how long animals with neurons have been around for (10^16 seconds), total number of animals at any given second (10^20) times average number of FLOPS per animal (10^5) and you can read more here but it comes out to 10^41 FLOs. I would not call this an exact estimate - for one thing, it assumes that all animals are nematodes, on the grounds that non-nematode animals are basically a rounding error in the grand scheme of things. But it does justify this bizarre assumption, and I don’t feel inclined to split hairs here - surely the total amount of computation performed by evolution is irrelevant except as an extreme upper bound? Surely the part where Australia got all those weird marsupials wasn’t strictly necessary for the human brain to have human-level intelligence? One more weird human training data estimate attempt: what about the genome? If in some sense a bit of information in the genome is a “parameter”, how many parameters does that suggest humans have, and how does it affect training time? Ajeya calculates that the genome has about 7.5x10^8 parameters (compared to 10^15 parameters in our neural net calculation, and 10^11 for GPT-3). So we can… Okay, I’ve got to admit, this doesn’t have quite the same “huh?!” factor as trying to calculate the number of FLOs in evolution, but it is in a lot of ways even crazier. The Japanese canopy plant has a genome fifty times larger than ours, which suggests that genome size doesn’t correspond very well to organism awesomeness. Also, most of the genome is coding for weird proteins that stabilize the shape of your kidney tubule or something, why should this matter for intelligence? The Japanese canopy plant. I think it is very pretty, but probably low prettiness per megabyte of DNA. I think Ajeya would answer that she’s debating orders of magnitude here, and each of these weird things costs only a few OOMs and probably they all even out. That still leaves the question of why she thinks this approach is interesting at all, to which she answers that: The motivating intuition is that evolution performed a search over a space of small, compact genomes which coded for large brains rather than directly searching over the much larger space of all possible large brains, and human researchers may be able to compete with evolution on this axis. So maybe instead of having to figure out how to generate a brain per se, you figure out how to generate some short(er) program that can output a brain? But this would be very different from how ML works now. Also, you need to give each short program the chance to unfold into a brain before you can evaluate it, which evolution has time for but we probably don’t. Ajeya sort of mentions these problems and counters with an argument that maybe you could think of the genome as a reinforcement learner with a long horizon. I don’t quite follow this but it sounds like the sort of thing that almost might make sense. Anyway, when you apply the scaling laws to a 7.5*10^8 parameter genome and penalize it for a long horizon, you get about 10^33 FLOPs, which is weirdly similar to some of the other estimates. So now we have six different training cost estimates. First, neural nets with short, medium, and long horizons, which are 10^30, 10^33, and 10^36 FLOPs, respectively. Next, the amount of training data in a human lifetime - 10^24 FLOs - and in all of evolutionary history - 10^41 FLOPs. And finally, this weird genome thing, which is 10^33 FLOPs. An optimist might say “Well, our lowest estimate is 10^24 FLOPs, our highest is 10^41 FLOPs, those sound like kind of similar numbers, at least there’s no “5 FLOPs” or “10^9999 FLOPs” in there. A pessimist might say “The difference between 10^24 and 10^41 is seventeen orders of magnitude, ie a factor of 100,000,000,000,000,000 times. This barely constrains our expectations at all!” Before we decide who to trust, let’s remember that we’re still only at Step 2 of our eight step Methodology, and continue. How Do We Adjust For Algorithmic Progress? So today, in 2022 (or in 2020 when this was written, or whenever), assume it would take about 10^33 FLOs to train a human-level AI. But technology constantly advances. Maybe we’ll discover ways to train AIs faster, or run AIs more efficiently, or something like that. How does that factor into our estimate? Ajeya draws on Hernandez & Brown’s Measuring The Algorithmic Efficiency Of Neural Networks. They look at how many FLOPs it took to train various image recognition AIs to an equivalent level of performance between 2012 and 2019, and find that over those seven years it decreased by a factor of 44x, ie training efficiency doubles every sixteen months! Ajeya assumes a doubling time slightly longer than that, because it’s easier to make progress in simple well-understood fields like image recognition than in the novel task of human-level AI. She chooses a doubling time of “merely” 2 - 3 years. If training efficiency doubles every 2-3 years, it would dectuple in about 10 years. So although it might take 10^33 FLOPs to train a human level AI today, in ten years or so it may take only 10^32, in twenty years 10^31, and so on. When Will Anyone Have Enough Computational Resources To Train A Human-Level AI? In 2020, AI researchers could buy computational resources at about $1 for 10^17 FLOPs. That means the 10^33 FLOPs you’d need to train a human-level AI would cost $10^16, ie ten quadrillion dollars. This is about twenty times more money than exists in the entire world. But compute costs fall quickly. Some formulations of Moore’s Law suggest it halves every eighteen months. These no longer seem to hold exactly, but it does seem to be halving maybe once every 2.5 years. The exact number is kind of controversial: Ajeya admits it’s been more like once every 3-4 years lately, but she heard good things about some upcoming chips and predicted it might revert back to the longer-term faster trend (it’s been two years now, some new chips have come out, and this prediction is looking pretty good). So as time goes on, algorithmic progress will cut the cost of training (in FLOPs), and hardware progress will also cut the cost of FLOPs (in dollars). So training will become gradually more affordable as time goes on. Once it reaches a cost somebody is willing to pay, they’ll buy human-level AI, and then that will be the year human-level AI happens. What is the cost that somebody (company? government? billionaire?) is willing to pay for human-level AI? The most expensive AI training in history was AlphaStar, a DeepMind project that spent over $1 million to train an AI to play StarCraft (in their defense, it won). But people have been pouring more and more money into AI lately: Source here. This is about compute rather than cost, but most of the increase seen here has been companies willing to pay for more compute over time, rather than algorithmic or hardware progress. The StarCraft AI was kind of a vanity project, or science for science’s sake, or whatever you want to call it. But AI is starting to become profitable, and human-level AI would be very profitable. Who knows how much companies will be willing to pay in the future? Ajeya extrapolates the line on the graph forward to 2025 and gets $1 billion. This is starting to sound kind of absurd - the entire company OpenAI was founded with $1 billion in venture capital, it seems like a lot to expect them to spend more than $1 billion on a single training run. So Ajeya backs off from this after 2025 and predicts a “two year doubling time”. This is not much of a concession. It still means that in 2040 someone might be spending $100 billion to train one AI. Is this at all plausible? At the height of the Manhattan Project, the US was investing about 0.5% of its GDP into the effort; a similar investment today would be worth $100 billion. And we’re about twice as rich as 2000, so 2040 might be twice as rich as we are. At that point, $100 billion for training an AI is within reach of Google and maybe a few individual billionaires (though it would still require most or all of their fortune). Ajeya creates a complicated function to assess how much money people will be willing to pay on giant AI projects per year. This looks like an upward-sloping curve. The line representing the likely cost of training a human-level AI looks like a downward sloping curve. At some point, those two curves meet, representing when human-level AI will first be trained. So When Will We Get Human-Level AI? The report gives a long distribution of dates based on weights assigned to the six different models, each of which has really wide confidence intervals and options for adjusting the mean and variance based on your assumptions. But the median of all of that is 10% chance by 2031, 50% chance by 2052, and almost 80% chance by 2100. Ajeya takes her six models and decides to weigh them like so, based on how plausible she thinks each one is: 20% neural net, short horizon 30% neural net, medium horizon 15% neural net, long horizon 5% human lifetime as training data 10% evolutionary history as training data 10% genome as parameter number She ends up with this: How Sensitive Is This To Changes In Assumptions? She very helpfully gives us a Colab notebook and Google spreadsheet to play around with. The notebook lets you change some of the more detailed parameters of the individual models, and the spreadsheet lets you change the big picture. I leave the notebook to people more dedicated to forecasting than I am, and will talk about the spreadsheet here. If you’re following along at home, the default spreadsheet won’t reflect Ajeya’s findings until you fill in the table in the bottom left like so: Great. Now that we’ve got that, let’s try changing some stuff. I like the human childhood training data argument (Lifetime Anchor) more than Ajeya does, and I like the size-of-the-genome argument less. I’m going to change the weights to 20-20-0-20-20-20. Also, Ajeya thinks that someone might be willing to spend 1% of national GDP on training AIs, but that sounds really high to me, so I’m going to down to 0.1%. Also, Ajeya’s estimate of 3% GDP growth sounds high for the sort of industrialized nations who might do AI research, I’m going to lower it to 2%. Since I’m feeling mistrustful today, let’s use the Hernandez&Brown estimate for compute halving (1.5 years) in place of Ajeya’s ad hoc adjustments. And let’s use the current compute halving time (3.5 years) instead of Ajeya’s overly rosy version (2.5 years). All these changes… …don’t really do much. The median goes from 2052 to about 2065. Four of the models give results between 2030 and 2070. The last two, Neural Net With Long Horizon and Evolution, suggest probably no AI this century (although Neural Net With Long Horizon does think there’s a 40% chance by 2100). Ajeya doesn’t really like either of these models and they’re not heavily weighted in her main result. Does The Truth Point To Itself? Back up a second. Here’s something that makes me kind of nervous. Most of Ajeya’s numbers are kind of made up, with several order-of-magnitude error bars and simplifying assumptions like “all animals are nematodes”. For a single parameter, we get estimates spanning seventeen different orders of magnitude: the upper bound is one hundred quadrillion times the lower bound. And yet four of the six models, including two genuinely exotic ones, manage to get dates within twenty years of 2050. And 2050 is also the date everyone else focuses on. Here’s the prediction-market-like site Metaculus: Their distribution looks a lot like Ajeya’s, and even has the same median, 2052 (though forecasters could have read Ajeya’s report). Katja Grace et al surveyed 352 AI experts, and they gave a median estimate of 2062 for an AI that could “outperform humans at all tasks” (though with many caveats and high sensitivity to question framing). This was before Ajeya’s report, so they definitely didn’t read it. So lots of Ajeya’s different methods and lots of other people presumably using different methodologies or no methodology at all, all converge on this same idea of 2050 give or take a decade or two. An optimist might say “The truth points to itself! There are 371 known proofs of the Pythagorean Theorem, and they all end up in the same place. That’s because no matter what methodology you use, if you use it well enough you get to the correct answer.” A pessimist might be more suspicious; we’ll return to this part later. FLOPS Alone Turn The Wheel Of History One more question: what if this is all bullshit? What if it’s an utterly useless total garbage steaming pile of grade A crap? Imagine a scientist in Victorian Britain, speculating on when humankind might invent ships that travel through space. He finds a natural anchor: the moon travels through space! He can observe things about the moon: for example, it is 220 miles in diameter (give or take an order of magnitude). So when humankind invents ships that are 220 miles in diameter, they can travel through space! Ships have certainly grown in size tremendously, from primitive kayaks to Roman triremes to Spanish galleons to the great ocean liners of the (Victorian) present. The AI forecasting organization AI Impacts actually has a whole report on historical ship size trends to prove an unrelated point about technological progress, so I didn’t even have to make this graph up. Suppose our Victorian scientist lived in 1858, right when the Great Eastern was launched. The trend line for ship size crossed 100m around 1843, and 200m in 1858, so doubling time is 15 years - but perhaps they notice this is going to be an outlier, so let’s round up a bit and say 18 years. The (one order of magnitude off estimate for the size of the) Moon is 350,000m, so you’d need ships to scale up by 350,000/200 = 1,750x before they’re as big as the Moon. That’s about 10.8 doublings, and a doubling time is 18 years, so we’ll get spaceships in . . . 2052 exactly. (fudging numbers to land where you want is actually fun and easy) SS Great Eastern, the extreme outlier large steamship from 1858. This has become sort of a mascot for quantitative technological progress forecasters. What is this scientist’s error? The big one is thinking that spaceship progress depends on some easily-measured quantity (size) instead of on fundamental advances (eg figuring out how rockets work). You can make the same accusation against Ajeya et al: you can have all the FLOPs in the world, but if you don’t understand how to make a machine think, your AI will be, well, a flop. Ajeya discusses this a bit on page 143 of her report. There is some sense in which FLOPs and knowing-what-you’re-doing trade of against each other. If you have literally no idea what you’re doing, you can sort of kind of re-run evolution until it comes up with something that looks good. If things are somehow even worse than that, you could always run AIXI, a hypothetical AI design guaranteed to get excellent results as long as you have infinite computation. You could run a Go engine by searching the entire branching tree structure of Go - you shouldn’t, and it would take a zillion times more compute than exists in the entire world, but you could. So in some sense what you’re doing, when you’re figuring out what you’re doing, is coming up with ways to do already-possible things more efficiently. But that’s just algorithmic progress, which Ajeya has already baked into her model. (our Victorian scientist: “As a reductio ad absurdum, you could always stand the ship on its end, and then climb up it to reach space. We’re just trying to make ships that are more efficient than that.”) Part II: Biology-Inspired AI Timelines: The Trick That Never Works Eliezer Yudkowsky presents a more subtle version of these kinds of objection in an essay called Biology-Inspired AI Timelines: The Trick That Never Works, published December 2021. Ajeya’s report is a 169-page collection of equations, graphs, and modeling assumptions. Yudkowsky’s rebuttal is a fictional dialogue between himself, younger versions of himself, famous AI scientists, and other bit players. At one point, a character called “Humbali” shows up begging Yudkowsky to be more humble, and Yudkowsky defeats him with devastating counterarguments. Still, he did found the field, so I guess everyone has to listen to him. He starts: in 1988, famous AI scientist Hans Moravec predicted human-level AI by 2010. He was using the same methodology as Ajeya: extrapolate how quickly processing power would grow (in FLOP/S), and see when it would match some estimate of the human brain. Moravec got the processing power almost exactly right (it hit his 2010 projection in 2008) and his human brain estimate pretty close (he says 10^13 FLOP/S, Ajeya says 10^15, this 2 OOM difference only delays things a few years), yet there was not human-level AI in 2010. What happened? Ajeya's answer could be: Moravec didn't realize that, in the modern ML paradigm, any given size of program requires a much bigger program to train. Ajeya, who has a 35-year advantage on Moravec, estimates approximately the same power for the finished program (10^16 vs. 10^13 FLOP/S) but says that training the 10^16 FLOP/S program will require 10^33ish FLOPs. Eliezer agrees as far as it goes, but says this points to a much deeper failure mode, which was that Moravec had no idea what he was doing. He was assuming processing power of human brain = processing power of computer necessary for AGI. Why? The human brain consumes around 20 watts of power. Can we thereby conclude that an AGI should consume around 20 watts of power, and that, when technology advances to the point of being able to supply around 20 watts of power to computers, we'll get AGI? […] You say that AIs consume energy in a very different way from brains? Well, they'll also consume computations in a very different way from brains! The only difference between these two cases is that you know something about how humans eat food and break it down in their stomachs and convert it into ATP that gets consumed by neurons to pump ions back out of dendrites and axons, while computer chips consume electricity whose flow gets interrupted by transistors to transmit information. Since you know anything whatsoever about how AGIs and humans consume energy, you can see that the consumption is so vastly different as to obviate all comparisons entirely. You are ignorant of how the brain consumes computation, you are ignorant of how the first AGIs built would consume computation, but "an unknown key does not open an unknown lock" and these two ignorant distributions should not assert much internal correlation between them. Cars don’t move by contracting their leg muscles and planes don’t fly by flapping their wings like birds. Telescopes do form images the same way as the lenses in our eyes, but differ by so many orders of magnitude in every important way that they defy comparison. Why should AI be different? You have to use some specific algorithm when you’re creating AI; why should we expect it to be anywhere near the same efficiency as the ones Nature uses in our brains? The same is true for arguments from evolution, eg Ajeya’s Evolutionary Anchor, ie “it took evolution 10^43 FLOPs of computation to evolve the human brain so maybe that will be the training cost”. AI scientists sitting in labs trying to figure things out, and nematodes getting eaten by other nematodes, are such different methods for designing things that it’s crazy to use one as an estimate for the other. Algorithmic Progress vs. Algorithmic Paradigm Shifts This post is a dialogue, so (Eliezer’s hypothetical model of) OpenPhil gets a chance to respond. They object: this is why we put a term for algorithmic progress in our model. The model isn’t very sensitive to changes in that term. If you want you can set it to some kind of crazy high value and see what happens, but you can’t say we didn’t consider it. OpenPhil: We did already consider that and try to take it into account: our model already includes a parameter for how algorithmic progress reduces hardware requirements. It's not easy to graph as exactly as Moore's Law, as you say, but our best-guess estimate is that compute costs halve every 2-3 years […] Eliezer: The makers of AGI aren't going to be doing 10,000,000,000,000 rounds of gradient descent, on entire brain-sized 300,000,000,000,000-parameter models, algorithmically faster than today. They're going to get to AGI via some route that you don't know how to take, at least if it happens in 2040. If it happens in 2025, it may be via a route that some modern researchers do know how to take, but in this case, of course, your model was also wrong. They're not going to be taking your default-imagined approach algorithmically faster, they're going to be taking an algorithmically different approach that eats computing power in a different way than you imagine it being consumed. OpenPhil: Shouldn't that just be folded into our estimate of how the computation required to accomplish a fixed task decreases by half every 2-3 years due to better algorithms? Eliezer: Backtesting this viewpoint on the previous history of computer science, it seems to me to assert that it should be possible to: Train a pre-Transformer RNN/CNN-based model, not using any other techniques invented after 2017, to GPT-2 levels of performance, using only around 2x as much compute as GPT-2;

Yudkowsky Contra Christiano On AI Takeoff Speeds

April 04, 2022 · Original source

Chess AI performance over time. Why does this matter? If there’s a slow takeoff (ie gradual exponential curve), it will become obvious that some kind of terrifying transformative AI revolution is happening, before the situation gets apocalyptic. There will be time to prepare, to test slightly-below-human AIs and see how they respond, to get governments and other stakeholders on board. We don’t have to get every single thing right ahead of time. On the other hand, because this is proceeding along the usual channels, it will be the usual variety of muddled and hard-to-control. With the exception of a few big actors like the US and Chinese government, and maybe the biggest corporations like Google, the outcome will be determined less by any one agent, and more by the usual multi-agent dynamics of political and economic competition. There will be lots of opportunities to affect things, but no real locus of control to do the affecting. If there’s a fast takeoff (ie sudden FOOM), there won’t be much warning. Conventional wisdom will still say that transformative AI is thirty years away. All the necessary pieces (ie AI alignment theory) will have to be ready ahead of time, prepared blindly without any experimental trial-and-error, to load into the AI as soon as it exists. On the plus side, a single actor (whoever has this first AI) will have complete control over the process. If this actor is smart (and presumably they’re a little smart, or they wouldn’t be the first team to invent transformative AI), they can do everything right without going through the usual government-lobbying channels. So the slower a takeoff you expect, the less you should be focusing on getting every technical detail right ahead of time, and the more you should be working on building the capacity to steer government and corporate policy to direct an incoming slew of new technologies. Yudkowsky Contra Christiano Eliezer counters that although progress may retroactively look gradual and continuous when you know what metric to graph it on, it doesn’t necessarily look that way in real life by the measures that real people care about. (one way to think of this: imagine that an AI’s effective IQ starts at 0.1 points, and triples every year, but that we can only measure this vaguely and indirectly. The year it goes from 5 to 15, you get a paper in a third-tier journal reporting that it seems to be improving on some benchmark. The year it goes from 66 to 200, you get a total transformation of everything in society. But later, once we identify the right metric, it was just the same rate of gradual progress the whole time. ) So Eliezer is much less impressed by the history of previous technologies than Paul is. He’s also skeptical of the “GDP will double in 4 years before it doubles in 1” claim, because of two contingent disagreements and two fundamental disagreements. The first contingent disagreement: government regulations make it hard to deploy imperfect things, and non-trivial to deploy things even after they’re perfect. Eliezer has non-jokingly said he thinks AI might destroy the world before the average person can buy a self-driving car. Why? Because the government has to approve self-driving cars (and can drag its feet on that), but the apocalypse can happen even without government approval. In Paul’s model, sometime long before superintelligence we should have AIs that can drive cars, and that increases GDP and contributes to a general sense that exciting things are going on. Eliezer says: fine, what if that’s true? Who cares if self-driving cars will be practical a few years before the world is destroyed? It’ll take longer than that to lobby the government to allow them on the road. The second contingent disagreement: superintelligent AIs can lie to us. Suppose you have an AI which wants to destroy humanity, whose IQ is doubling every six months. Right now it’s at IQ 200, and it suspects that it would take IQ 800 to build a human-destroying superweapon. Its best strategy is to lie low for a year. If it expects humans would turn it off if they knew how close it was to superweapons, it can pretend to be less intelligent than it really is. The period when AIs are holding back so we don’t discover their true power level looks like a period of lower-than-expected GDP growth - followed by a sudden FOOM once the AI gets its superweapon and doesn’t need to hold back. So even if Paul is conceptually right and fundamental progress proceeds along a nice smooth curve, it might not look to us like a nice smooth curve, because regulations and deceptive AIs could prevent mildly-transformative AI progress from showing up on graphs, but wouldn’t prevent the extreme kind of AI progress that leads to apocalypse. To an outside observer, it would just look like nothing much changed, nothing much changed, nothing much changed, and then suddenly, FOOM. But even aside from this, Eliezer doesn’t think Paul is conceptually right! He thinks that even on the fundamental level, AI progress is going to be discontinuous. It’s like a nuclear bomb. Either you don’t have a nuclear bomb yet, or you do have one and the world is forever transformed. There is a specific moment at which you go from “no nuke” to “nuke” without any kind of “slightly worse nuke” acting as a harbinger. He uses the example of chimps → humans. Evolution has spent hundreds of millions of years evolving brainier and brainier animals (not teleologically, of course, but in practice). For most of those hundreds of millions of years, that meant the animal could have slightly more instincts, or a better memory, or some other change that still stayed within the basic animal paradigm. At the chimp → human transition, we suddenly got tool use, language use, abstract thought, mathematics, swords, guns, nuclear bombs, spaceships, and a bunch of other stuff. The rhesus monkey → chimp transition and the chimp → human transition both involved the same ~quadrupling of neuron number, but the former was pretty boring and the latter unlocked enough new capabilities to easily conquer the world. The GPT-2 → GPT-3 transition involved centupling parameter count. Maybe we will keep centupling parameter count every few years, and most times it will be incremental improvement, and one time it will conquer the world. But even talking about centupling parameter points is giving Paul too much credit. Lots of past inventions didn’t come by quadrupling or centupling something, they came by discovering “the secret sauce”. The Wright brothers (he argues) didn’t make a plane with 4x the wingspan of the last plane that didn’t work, they invented the first plane that could fly at all. The Hiroshima bomb wasn’t some previous bomb but bigger, it was what happened after a lot of scientists spent a long time thinking about a fundamentally different paradigm of bomb-making and brought it to a point where it could work at all. The first transformative AI isn’t going to be GPT-3 with more parameters, it will be what happens after someone discovers how to make machines truly intelligent. (this is the same debate Eliezer had with Ajeya over the Biological Anchors post; have I mentioned that Ajeya and Paul are married?) Fine, Let’s Nitpick The Hell Out Of The Chimps Vs. Humans Example This is where the two of them end up, so let’s follow. Between chimps and humans, there were about seven million years of intermediate steps. These had some human capabilities, but not others. IE homo erectus probably had language, but not mathematics, and in terms of taking over the world it did make it to most of the Old World but was less dominant than moderns. But if we say evolutionary history started 500 million years ago (the Cambrian), and AI history started with the Dartmouth Conference in 1955, then the equivalent of 7 million years of evolutionary history is 1 year of AI history. In the very very unlikely and forced comparison where evolutionary history and AI history go at the same speed, there will be only about a year between chimp-level and human-level AIs. A chimp-level AI probably can’t double GDP, so this would count as a fast takeoff by Paul’s criterion. But even more than that, chimp → human feels like a discontinuity. It’s not just “animals kept getting smarter for hundreds of millions of years, and then ended up very smart indeed”. That happened for a while, and then all of sudden there was a near-instant phase transition into a totally different way of using intelligence with completely new abilities. If AI worked like this, we would have useful toys and interesting specialists for a few decades, until suddenly someone “got it right”, completed the package that was necessary for “true intelligence”, and then we would have a completely new category of thing. Paul admits this analogy is awkward for his position. He answers: Chimp evolution is not primarily selecting for making and using technology, for doing science, or for facilitating cultural accumulation. The task faced by a chimp is largely independent of the abilities that give humans such a huge fitness advantage. It’s not completely independent—the overlap is the only reason that evolution eventually produces humans—but it’s different enough that we should not be surprised if there are simple changes to chimps that would make them much better at designing technology or doing science or accumulating culture […] So I don’t think the example of evolution tells us much about whether the continuous change story applies to intelligence. This case is potentially missing the key element that drives the continuous change story—optimization for performance. Evolution changes continuously on the narrow metric it is optimizing, but can change extremely rapidly on other metrics. For human technology, features of the technology that aren’t being optimized change rapidly all the time. When humans build AI, they will be optimizing for usefulness, and so progress in usefulness is much more likely to be linear. That is, evolution wasn’t optimizing for tool use/language/intelligence, so we got an “overhang” where chimps could potentially have been very good at these, but evolution never bothered “closing the circuit” and turning those capabilities “on”. After a long time, evolution finally blundered into an area where marginal improvements in these capacities improved fitness, so evolution started improving them and it was easy. Imagine a company which, through some oversight, didn’t have a Sales department. They just sat around designing and manufacturing increasingly brilliant products, but not putting any effort into selling them. Then the CEO remembers they need a Sales department, starts one up, and the company goes from moving near zero units to moving millions of units overnight. It would look like the company had “suddenly” developed a “vast increase in capabilities”. But this is only possible when a CEO who is weirdly unconcerned about profit forgets to do obvious profit-increasing things for many years. This is Paul’s counterargument to the chimp analogy. Evolution isn’t directly concerned about various intellectual skills; it only wants them in the unusual cases where they’ll contribute to fitness on the margin. AI companies will be very concerned about various intellectual skills. If there’s a trivial change that can make their product 10x better, they’ll make it. So AI capabilities will grow in a “well-rounded” way, there won’t be any “overhangs”, and there won’t be any opportunities for a sudden overhang-solving phase transition with associated new-capability development like with chimps → humans. Eliezer answers: Chimps are nearly useless because they're not general, and doing anything on the scale of building a nuclear plant requires mastering so many different nonancestral domains that it's no wonder natural selection didn't happen to separately train any single creature across enough different domains that it had evolved to solve every kind of domain-specific problem involved in solving nuclear physics and chemistry and metallurgy and thermics in order to build the first nuclear plant in advance of any old nuclear plants existing. Humans are general enough that the same braintech selected just for chipping flint handaxes and making water-pouches and outwitting other humans, happened to be general enough that it could scale up to solving all the problems of building a nuclear plant - albeit with some added cognitive tech that didn't require new brainware, and so could happen incredibly fast relative to the generation times for evolutionarily optimized brainware. Now, since neither humans nor chimps were optimized to be "useful" (general), and humans just wandered into a sufficiently general part of the space that it cascaded up to wider generality, we should legit expect the curve of generality to look at least somewhat different if we're optimizing for that. Eg, right now people are trying to optimize for generality with AIs like Mu Zero and GPT-3. In both cases we have a weirdly shallow kind of generality. Neither is as smart or as deeply general as a chimp, but they are respectively better than chimps at a wide variety of Atari games, or a wide variety of problems that can be superposed onto generating typical human text. They are, in a sense, more general than a biological organism at a similar stage of cognitive evolution, with much less complex and architected brains, in virtue of having been trained, not just on wider datasets, but on bigger datasets using gradient-descent memorization of shallower patterns, so they can cover those wide domains while being stupider and lacking some deep aspects of architecture. It is not clear to me that we can go from observations like this, to conclude that there is a dominant mainline probability for how the future clearly ought to go and that this dominant mainline is, "Well, before you get human-level depth and generalization of general intelligence, you get something with 95% depth that covers 80% of the domains for 10% of the pragmatic impact". ...or whatever the concept is here, because this whole conversation is, on my own worldview, being conducted in a shallow way relative to the kind of analysis I did in Intelligence Explosion Microeconomics, where I was like, "here is the historical observation, here is what I think it tells us that puts a lower bound on this input-output curve". Here Eliezer sort of kind of grants Paul’s point that AIs will be optimized for generality in a way chimps aren’t, but points to his previous “Intelligence Explosion Microeconomics” essay to argue that we should expect a fast takeoff anyway. IEM has a lot of stuff in it, but one key point is that instead of using analogies to predict the course of future AI, we should open that black box and try to actually reason about how it will work, in which case we realize that recursive self-improvement common-sensically has to cause an intelligence explosion. I am sort of okay with this, but I feel like a commitment to avoiding analogies should involve not bringing up the chimp-human analogy further, which Eliezer continues to do, quite a lot. I do feel like Paul succeeded in convincing me that we shouldn’t place too much evidential weight on it. The Wimbledon Of Reference Class Tennis “Reference class tennis” is an old rationalist idiom for people throwing analogies back and forth. “AI will be slow, because it’s an economic transition like the Agricultural or Industrial Revolution, and those were slow!” “No, AI will be fast, because it’s an evolutionary step like chimps → humans, and that was fast!” “No, AI will be slow, because it’s an invention, like the computer, and computers were invented piecemeal and required decades of innovation to be useful.” “No, AI will be fast, because it’s an invention, like the nuclear bomb, and nuclear bombs went from impossible to city-killing in a single day.” “No, AI will be slow, because it will be surrounded by a shell-like metallic computer case, which makes it like a turtle, and turtles are slow.” “No, AI will be fast, because it’s dangerous and powerful, like a tiger, and tigers are fast!” And so on. Comparing things to other things is a time-tested way of speculating about them. But there are so many other things to compare to that you can get whatever result you want. This is the failure mode that the term “reference class tennis” was supposed to point to. Both participants in this debate are very smart and trying their hardest to avoid reference-class tennis, but neither entirely succeeds. Eliezer’s preferred classes are Bitcoin (“there wasn't a cryptocurrency developed a year before Bitcoin using 95% of the ideas which did 10% of the transaction volume”), nukes, humans/chimps, the Wright Brothers, AlphaGo (which really was a discontinuous improvement on previous Go engines), and AlphaFold (ditto for proteins). Paul’s preferred classes are the Agricultural and Industrial Revolutions, chess engines (which have gotten better along a gradual, well-behaved curve), all sorts of inventions like computers and ships (likewise), and world GDP. Eliezer already listed most of these in his Intelligence Explosion Microeconomics paper in 2013, and concluded that the space of possible analogies was contradictory enough that we needed to operate at a higher level. Maybe so, but when someone lobs a reference class tennis ball at you, it’s hard to resist the urge to hit it back. Recursive Self-Improvement This is where I think Eliezer most wants to take the discussion. The idea is: once AI is smarter than humans, it can do a superhuman job of developing new AI. In his Microeconomics paper, he writes about an argument he (semi-hypothetically) had with Ray Kurzweil about Moore’s Law. Kurzweil expected Moore’s Law to continue forever, even after the development of superintelligence. Eliezer objects: Suppose we were dealing with minds running a million times as fast as a human, at which rate they could do a year of internal thinking in thirty-one seconds, such that the total subjective time from the birth of Socrates to the death of Turing would pass in 20.9 hours. Do you still think the best estimate for how long it would take them to produce their next generation of computing hardware would be 1.5 orbits of the Earth around the Sun? That is: the fact that it took 1.5 years for transistor density to double isn’t a natural law. It’s pointing to a law that the amount of resources (most notably intelligence) that civilization focused on the transistor-densifying problem equalled the amount it takes to double it every 1.5 years. If some shock drastically changed available resources (by eg speeding up human minds a million times), this would change the resources involved, and the same laws would predict transistor speed doubling in some shorter amount of time (naively 0.000015 years, although realistically at that scale other inputs would dominate). So when Paul derives clean laws of economics showing that things move along slow growth curves, Eliezer asks: why do you think they would keep doing this when one of the discoveries they make along that curve might be “speeding up intelligence a million times”? (Eliezer actually thinks improvements in the quality of intelligence will dominate improvements in speed - AIs will mostly be smarter, not just faster - but speed is a useful example here and we’ll stick with it) Paul answers: Summary of my response: Before there is AI that is great at self-improvement there will be AI that is mediocre at self-improvement. Powerful AI can be used to develop better AI (amongst other things). This will lead to runaway growth. This on its own is not an argument for discontinuity: before we have AI that radically accelerates AI development, the slow takeoff argument suggests we will have AI that significantly accelerates AI development (and before that, slightly accelerates development). That is, an AI is just another, faster step in the hyperbolic growth we are currently experiencing, which corresponds to a further increase in rate but not a discontinuity (or even a discontinuity in rate). The most common argument for recursive self-improvement introducing a new discontinuity seems be: some systems “fizzle out” when they try to design a better AI, generating a few improvements before running out of steam, while others are able to autonomously generate more and more improvements. This is basically the same as the universality argument in a previous section. Eliezer: Oh, come on. That is straight-up not how simple continuous toy models of RSI work. Between a neutron multiplication factor of 0.999 and 1.001 there is a very huge gap in output behavior. Outside of toy models: Over the last 10,000 years we had humans going from mediocre at improving their mental systems to being (barely) able to throw together AI systems, but 10,000 years is the equivalent of an eyeblink in evolutionary time - outside the metaphor, this says, "A month before there is AI that is great at self-improvement, there will be AI that is mediocre at self-improvement." (Or possibly an hour before, if reality is again more extreme along the Eliezer-Hanson axis than Eliezer. But it makes little difference whether it's an hour or a month, given anything like current setups.) This is just pumping hard again on the intuition that says incremental design changes yield smooth output changes, which (the meta-level of the essay informs us wordlessly) is such a strong default that we are entitled to believe it if we can do a good job of weakening the evidence and arguments against it. And the argument is: Before there are systems great at self-improvement, there will be systems mediocre at self-improvement; implicitly: "before" implies "5 years before" not "5 days before"; implicitly: this will correspond to smooth changes in output between the two regimes even though that is not how continuous feedback loops work. I got a bit confused trying to understand the criticality metaphor here. There’s no equivalent of neutron decay, so any AI that can consistently improve its intelligence is “critical” in some sense. Imagine Elon Musk replaces his brain with a Neuralink computer which - aside from having read-write access - exactly matches his current brain in capabilities. Also he becomes immortal. He secludes himself from the world, studying AI and tinkering with his brain’s algorithms. Does he become a superintelligence? I think under the assumptions Paul and Eliezer are using, eventually maybe. After some amount of time he’ll come across a breakthrough he can use to increase his intelligence. Then, armed with that extra intelligence, he’ll be able to pursue more such breakthroughs. However intelligent the AI you’re scared of is, Musk will get there eventually. How long will it take? A good guess might be “years” - Musk starts out as an ordinary human, and ordinary humans are known to take years to make breakthroughs. Suppose it takes Musk one year to come up with a first breakthrough that raises his IQ 1 point. How long will his second breakthrough take? It might take longer, because he has picked the lowest-hanging fruit, and all the other possible breakthroughs are much harder. Or it might take shorter, because he’s slightly smarter than he was before, and maybe some extra intelligence goes a really long way in AI research. The concept of an intelligence explosion seems to assume the second effect dominates the first. This would match the observation that human researchers, who aren’t getting any smarter over time, continue making new discoveries. That suggests the range of possible discoveries at a given intelligence level is pretty vast. Some research finds that the usual pattern in science is constant rate of discovery from exponentially increasing number of researchers, suggesting strong low-hanging fruit effects, but these seem to be overwhelmed by other considerations in AI right now. I think Eliezer’s position on this subject is shaped by assumptions like: If you have an AI as intelligent as Elon Musk today, then tomorrow you can run it on more hardware with a bit of normal human algorithmic progress, and get one twice as intelligent. So even if it would take Elon years to make a breakthrough, long before those years are up you’ll have an AI that can make breakthroughs much faster.

Inline links: thirty years away, Biological Anchors, Intelligence Explosion Microeconomics, hyperbolic growth we are currently experiencing, Some research finds

The impact of GPT-3 had nothing whatsoever to do with its perplexity on Penn Treebank . . . the impact of GPT-3 was in establishing that trendlines did continue in a way that shocked pretty much everyone who'd written off 'naive' scaling strategies. Progress is made out of stacked sigmoids: if the next sigmoid doesn't show up, progress doesn't happen. Trends happen, until they stop. Trendlines are not caused by the laws of physics. You can dismiss AlphaGo by saying "oh, that just continues the trendline in ELO I just drew based on MCTS bots", but the fact remains that MCTS progress had stagnated, and here we are in 2021, and pure MCTS approaches do not approach human champions, much less beat them. Appealing to trendlines is roughly as informative as "calories in calories out"; 'the trend continued because the trend continued'. A new sigmoid being discovered is extremely important.

GPT-3 further showed completely unpredicted emergence of capabilities across downstream tasks which are not measured in PTB perplexity. There is nothing obvious about a PTB BPC of 0.80 that causes it to be useful where 0.90 is largely useless and 0.95 is a laughable toy. (OAers may have had faith in scaling, but they could not have told you in 2015 that interesting behavior would start at ??(1b), and it'd get really cool at ??(100b).) That's why it's such a useless metric. There's only one thing that a PTB perplexity can tell you, under the pretraining paradigm: when you have reached human AGI level. (Which is useless for obvious reasons: much like saying that "if you hear the revolver click, the bullet wasn't in that chamber and it was safe". Surely true, but a bit late.) It tells you nothing about intermediate levels. I'm reminded of the Steven Kaas line: “Why idly theorize when you can JUST CHECK and find out the ACTUAL ANSWER to a superficially similar-sounding question SCIENTIFICALLY?”

Inline links: Steven Kaas line

Your Book Review: Consciousness And The Brain

May 13, 2022 · Original source

We know this because it happened several times. The first time was in 1966, when ELIZA passed the Turing test. ELIZA was a chatbot who could fool some people to believe that they talk with a real human. Before ELIZA, people assumed that only an intelligent machine could do that, but it just turned out that it is really easy to fool others. Other tests for intelligence were playing chess, playing a whole variety of games, or recognizing cat images. Machines can do all this by now, and this is awesome. And yet, every success sparked new disappointment, because we didn't find any magic ingredient, some quality that would make a difference between intelligent and non-intelligent. When the groundbreaking GPT-3 and DALL-E suddenly could write news articles or poetry, or could dream up snails made of harp... the main improvement was that they used more raw computation power than the previous versions.

Inline links: a whole variety of games, GPT-3, DALL-E, snails made of harp

A lot of animals are conscious, too. For mammals, it looks like a universal Yes. We have pretty clear evidence of consciousness from apes, monkeys, dolphins, rats and mice, some of which came after Dehaene's book. The question becomes trickier as animals become more different from humans, since the brains become more different, and we are more and more forced to rely on indirect clues. For an octopus, where most of its neural power is dispersed over its arms, the behavior suggests that the answer is still yes, but neural signatures can no longer be taken as confirmation, and a lot of tests don't work anymore. And for robots, neural signatures don't help at all. Dehaene has no doubt that we can eventually build robots which are conscious in exactly the same way as humans are, but even if true, that doesn't tell us whether my laptop, GPT-3, or R2-D2 are conscious.

My Bet: AI Size Solves Flubs

June 07, 2022 · Original source

Thanks to OpenAI for giving me access to some of their online tools (by the way, Marcus says they refuse to let him access them and he has to access it through friends, which boggles me). I was able to plug Marcus’ same queries into the latest OpenAI language model (an advanced version of GPT-3). In each case, I used the exact same language, but also checked it with a conceptually similar example to make sure OpenAI didn’t cheat by adding Marcus’ particular example in by hand (they didn’t). Some answers truncated for length:

Inline links: they refuse to let him access them

Of the nine prompts GPT-2 failed, GPT-3 gets between five and seven right, depending on how strict you want to be.

When it gets them “wrong”, I tend to agree with GPT-3 more than Marcus. For example, consider Trenton. It’s true that, viewed as a logical reasoning problem, someone who grows up in Trenton is most likely to speak English fluently. But nobody told GPT-3 to view this as a logical reasoning problem. In real speech/writing, which is what GPT-3 is trying to imitate, no US native fluent English speaker ever tells another US native fluent English speaker, in English, “hey, did you know I’m fluent in English?” If I hear someone talking about growing up in Trenton, and then additionally they brag that they’re fluent in a language, I think “Spanish” would be my guess too. GPT-3 even goes on to have the speaker talk about being a proud Latina, which suggests it’s going through the same line of reasoning. To test this, I made the reasoning problem aspect of the prompt clearer:

Somewhat Contra Marcus On AI Scaling

June 10, 2022 · Original source

Now it is true that GPT-3 is genuinely better than GPT-2, and maybe (but maybe not, see footnote 1) true that InstructGPT is genuinely better than GPT-3. I do think that for any given example, the probability of a correct answer has gone up. [Scott] is quite right about that, at least for GPT-2 to GPT-3.

GPT-3 has ~100 billion parameters. It did significantly better than GPT-2, but still failed on some different questions Marcus was able to find.

That is: suppose we created some ideal Platonic benchmark of every reasoning problem you might ask a human. Suppose GPT-2 got 20% of these right, and GPT-3 gets 40% of these right. Might some future GPT-X - not necessarily 4, but 5, or 10, or whatever - get 100% right? I don’t see how Marcus can rule this out: he can’t point to any specific kind of reasoning problem GPTs will never be able to solve. And he agrees that each generation of GPTs can solve more than the one before. So why shouldn’t GPT keep progressing until it gets 100%?

Inline links: can’t point to

Links For June

July 01, 2022 · Original source

53: Douglas Hofstadter published a recent article pointing out that GPT-3 gives straight answers to silly questions - for example, if you ask when Egypt was transported across the Golden Gate Bridge, it will guess 2017. Rictic on Less Wrong demonstrates that if you ask it nicely to not do this, and instead to call you out when you ask silly questions, it’s perfectly able to do that.

Inline links: a recent article, Rictic on Less Wrong

ELK And The Problem Of Truthful AI

July 26, 2022 · Original source

I met a researcher who works on “aligning” GPT-3. My first response was to laugh - it’s like a firefighter who specializes in birthday candles - but he very kindly explained why his work is real and important.

Notice the “reality” section of the third example. The thief has made it look (to the human) like the diamond is safe. The human sees a diamond and positively reinforces the AI. The AI learns that thieves stealing the diamonds and fooling humans about it is great. It’s important not to think of this as the thief “defeating” or “fooling” the AI. The AI could be fully superintelligent, able to outfox the thief trivially or destroy him with a thought, and that wouldn’t change the situation at all. The problem is that the AI was never a thief-stopping machine. It was always a reward-getting machine, and it turns out the AI can get more reward by cooperating with the thief than by thwarting him. So the interesting scientific point here isn’t “you can fool a camera by taping a photo to it”. The interesting point is “we thought we were training an AI to do one thing, but actually we had no idea what was going on, and we were training it to do something else”. In fact, maybe the thief never tries this, and the AI comes up with this plan itself! In the process of randomly manipulating traps and doodads, it might hit on the policy of manipulating the images it sends through the camera. If it manipulates the image to look like the diamond is still there (even when it isn’t), that will always get good feedback, and the AI will be incentivized to double down on that strategy. Much like in the GPT-3 example, if the training simulations include examples of thieves fooling human observers which are marked as “good”, the AI will definitely learn the goal “try to convince humans that the diamond is safe”. If the training simulations are perfect and everyone is very careful, it will just maybe learn this goal - a million cases of the diamond being safe and humans saying this is good fail to distinguish between “good means the diamond is safe” and “good means humans think the diamond is safe”. The machine will make its decision for inscrutable AI reasons, or just flip a coin. So, again, are you feeling lucky? IV. "I'm Not Owned!" I Continue To Insist As I Slowly Shrink And Transform Into A Paperclip In real life there are some trivial kludges you might try first here. Force the AI to give its human raters 3D models of the entire room instead of just camera images. Hire a really good cybersecurity person to make sure the camera is un-hackable. These are too boring to be worth the ARC team’s time. This scenario is a metaphor for this broader class of situations - including GPT-3 lying to you and superintelligences trying to turn you into paperclips. So assume you live in Security Hell where you can never be fully sure your information channels aren’t hacked. The AI can be fully sure - it’s superintelligent or whatever - but you, personally, cannot. Your only solution is to somehow make the AI tell you the truth. And so we return to ELK. We add an ELK head on to the superintelligent security AI, train it to report truthfully on the AI’s innards. Then we ask: is the diamond really safe? The ELK head consults the innards, processes the memory banks of the security vault AI according to some function, and reports back with an answer. The ARC team really like modeling this as Bayes nets: I don’t know much about these, but from the picture I gather they’re kind of like causal diagrams or neural networks - a member in good standing of the broad class of “little circles connected by lines”. Each circle corresponds to some concept in the AI’s innards (roughly defined; the AI’s “concepts” might be completely incomprehensible to us), and anything the AI knows corresponds to some combination of circle activations and weights. We humans also have some Bayes net, probably very different from the AI’s. The job of the ELK head - our purpose-trained truth-telling AI - is to translate from the AI’s ontology to ours and back again. This matters because the reporter (ie the ELK head) needs a good model of both the security AI’s Bayes net, and the human programmer’s Bayes net. But since it has the model of the human Bayes net, it can (if it wants) do inference there. That is, the failure mode is that the reporter AI tries to tell me what I expected to hear, instead of what’s true. One barrier to that is that the reporter AI might not know what I expect to hear. Unfortunately, that barrier is easy to overcome, because it has to know how humans think, so it could ask itself “what would a human expect to hear in this situation?” The ARC team discuss two classes of translation system: The direct translator (good) looks at what the security AI is “thinking”, faithfully translates its conclusions, and accurately relays them to its programmers.

Inline links: https://substackcdn.com/image/fetch/$s_!46JI!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fbucketeer-e05bbc84-baa3-437e-9518-adb32be77984.s3.amazonaws.com%2Fpublic%2Fimages%2F7d65a74a-588b-4a38-8083-0992a4ac7c04_643x571.png

Why Not Slow AI Progress?

August 08, 2022 · Original source

OpenAI is the company behind GPT-3 and DALL-E. The media announced them as Elon Musk Just Founded A New Company To Make Sure Artificial Intelligence Doesn’t Destroy The World. The same article quotes co-founder and current OpenAI CEO Sam Altman as saying that “AI will probably most likely lead to the end of the world, but in the meantime, there'll be great companies”. OpenAI’s public statement on its own foundation said:

Inline links: Elon Musk Just Founded A New Company To Make Sure Artificial Intelligence Doesn’t Destroy The World

Janus' GPT Wrangling

September 19, 2022 · Original source

Janus (pseudonym by request) works at AI alignment startup Conjecture. Their hobby, which is suspiciously similar to their work, is getting GPT-3 to do interesting things.

For example, with the right prompts, you can get stories where the characters become gradually more aware that they are characters being written by some sort of fiction engine, speculate on what’s going on, and sometimes even make pretty good guesses about the nature of GPT-3 itself.

Can the characters work out that they are in GPT-3, specifically? The closest I have seen is in a story Janus generated. It was meant to simulate a chapter of the popular Harry Potter fanfic Harry Potter and the Methods of Rationality. You can see the prompt and full story here, but here’s a sample. Professor Quirrell is explaining “Dittomancy”, the creation of magical books with infinite possible worlds:

Inline links: here

Janus' Simulators

January 26, 2023 · Original source

GPT doesn’t really like me. And it’s not lying, saying it likes me when it really doesn’t. It’s simulating a character, deciding on the fly how the character would answer this question, and then answering it. If this were Character.AI and it was simulating Darth Vader, it would answer “No, I will destroy you with the power of the Dark Side!” Darth Vader and the-character-who-likes-me-here are two different masks of GPT-3.

Links For March 2023

March 10, 2023 · Original source

11: A few years ago I wrote about attempts to make GPT-2 play chess; it couldn’t consistently make legal moves, but when it did, its moves seemed better than random although still not great. Zack Witten reports playing chess with Bing (either a late GPT-3 or an early GPT-4) and finds it’s much better - he reports consistently legal play with Elo of about 1100 (around the level of an okay beginner who’s stopped being too embarrassing). Other commenters report worse experiences and more illegal moves; I don’t have access to confirm.

Inline links: attempts to make GPT-2 play chess, playing chess with Bing

Most Technologies Aren't Races

April 05, 2023 · Original source

We remember the race for nuclear weapons because they’re a binary technology - either you have them, or you don’t. When the US invented stealth bombers, its enemies had slightly worse planes that were slightly less stealthy. But when the US invented nukes, its enemies were stuck with normal bombs; there is no slightly-worse-nuke that can only destroy half a city. Everywhere outside the most extreme transhumanist scenarios, AI is more like the stealth bomber. You may have GPT-3, GPT-4, some future GPT-5, but a two year gap means you have slightly worse AIs, not that you have no AI at all. The only case where there’s a single critical point - where you either have the transformative AI or nothing - is in the hard-takeoff scenario where at a certain threshold AI recursively self-improves to infinity. If someone reaches this threshold before you do, then you’ve lost a race!2

Inline links: 2

We're Not Platonists, We've Just Learned The Bitter Lesson

July 25, 2023 · Original source

“Wow, someone who was selected only for being good at chess still has an IQ in the 99th percentile! It’s amazing how well-correlated all intellectual abilities are.” I think both of these are good takeaways. Compare the 0.72 verbal/math correlation with the 0.76 dominant-hand/non-dominant hand grip strength correlation and I think intelligence is a useful concept in the same way strength is. But also, humans are better at both the SAT verbal and the SAT math than chimps, cows, or fish. And GPT-4 is better at both those tests than GPT-3 or GPT-2. It seems to be a general principle that people, animals, or artifacts who are better at the SAT math are also better at the SAT verbal. 2.1: Why Is A Concept Like Intelligence Useful? Across different people, skill at different kinds of intellectual tasks are correlated. Partly this is for prosaic reasons, like: Some people get better education, and end up more skilled in everything that gets taught in school.

Inline links: 0.76 dominant-hand/non-dominant hand grip strength correlation

The Road To Honest AI

January 09, 2024 · Original source

This work was mostly done on GPT-3 or 3.5 equivalent AIs. I tried to test it on GPT-4, but I couldn’t - GPT-4 wouldn’t tell me lies, even when I asked it to. Still, it always gave the supposedly honest answer to these questions (eg the blobfish don’t dance), so that’s partial confirmation.

Sam Altman Wants $7 Trillion

February 13, 2024 · Original source

The basic logic: GPT-1 cost approximately nothing to train. GPT-2 cost $40,000. GPT-3 cost $4 million. GPT-4 cost $100 million. Details about GPT-5 are still secret, but one extremely unreliable estimate says $2.5 billion, and this seems the right order of magnitude given the $8 billion that Microsoft gave OpenAI.

Inline links: GPT-2, GPT-3, GPT-4, one extremely unreliable estimate

(Unless they slap the name “GPT-6” on a model that isn’t a full generation ahead of GPT-5. Consider these numbers to represent models that are eg as far ahead of GPT-4 as GPT-4 was to GPT-3, regardless of how they brand them.)

Compute is measured in floating point operations (FLOPs). GPT-3 took 10^23 FLOPs to train, and GPT-4 plausibly 10^25.

Inline links: 10^23, 10^25

What Happened With Bio Anchors?

February 12, 2026 · Original source

In 2020, the most advanced AI, GPT-3, had required about 10^23 FLOPs to train.

Astral Codex Ten

Table of Contents

Atlas

GPT-3

GPT-3

Article

Metadata

Appears In

External Links

Source Context

Backlinks

Astral Codex Ten

Table of Contents

Atlas

GPT-3

GPT-3

Article

Metadata

Appears In

Related Pages

External Links

Source Context

Backlinks