Hard disagree. I like to know what it is I’m reading. I got the strange feeling that this text was way more powerful/cogent than what I thought GPT-3 was capable of, and I feel very mislead that one of the crippling defects of GPT-3 (inability to maintain long-term coherency) was in fact being papered over by human intervention.
Not knowing beforehand sure did help me train my bullshit detector, though.
Thanks to Elizabeth for encouraging me to write this, and for her helpful feedback!
I’m sorry, but I have a very hard time believing your 1-15% range for long-term disability. If 15% of the labor force lost (say) half their productivity, we would already be seeing severe economic consequences across the board (all industries, worldwide).
I didn’t follow your calculations, but I do know that some monitoring projects define long covid as “having any symptom on a list (fatigue, more frequent headaches, etc) a few months after infection”. This is the proper definition from a scientific perspective: if we want to understand long covid, we want to capture it at all levels of severity and in many different manifestations. But this obviously produces a huge overestimate in terms of severe cases, as it also captures the upper range of the usual fluctuations of these symptoms.
The responses are cherry-picked, so this is way better than what GPT-3 is capable of. See the discussion in the downvoted subthread.
Your problem is called a clustering problem. First of all, you need to answer how you measure your error (information loss, as you call it). Typical error norms used are l1 (sum of individual errors), l2 (sum of squares of errors, penalizes larger errors more) and l-infinity (maximum error).
Once you select a norm, there always exists a partition that minimizes your error, and to find it there are a bunch of heuristic algorithms, e.g. k-means clustering. Luckily, since your data is one-dimensional and you have very few categories, you can just brute force it (for 4 categories you need to correctly place 3 boundaries, and naively trying all possible positions takes only n^3 runtime)
Hope this helps.
Thanks for posting this, I recognize this is emotionally hard for you. Please don’t interpret the rest of this post as being negative towards you specifically. I’m not trying to put you down, merely sharing the thoughts that came up as I read this.
I think you’re being very naive with your ideas about how this “could easily happen to anyone”. Several other commenters were focusing on how lonely people specifically are vulnerable to this. But I think it’s actually emotionally immature people who are vulnerable, specifically people with a high-openness, “taking ideas seriously” kind of personality, coupled with a lack of groundedness (too few points of contact with the physical world).
This is hard to explain without digressing at least a bit, so I’m going to elaborate, as much for my own benefit as yours.
As I’ve aged (late 30′s now), there’s been some hard to pin down changes in my personality. I feel more solidified than a decade ago. I now perceive past versions of myself almost as being a bit hollow; lots of stuff going on at the surface level, but my thoughts and experiences weren’t yet weaving together into the deep structures (below what’s immediately happening) that give a kind of “earthiness” or “groundedness” to all my thoughts and actions now. The world has been getting less confusing with each new thing I learn, so whatever I encounter, I tend to have related experiences already in my repertoire of ideas I’ve digested and integrated. Thus, acquisition of new information/modes of thinking/etc becomes faster and faster, even as my personality shifts less and less from each encounter with something new. I feel freer, more agenty now. This way of saying it is very focused on the intellect, but something analogous is going on at the emotional level as well.
I’ve started getting this impression of hollowness from many people around me, especially from young people who have had a very narrow life path, even highly intelligent ones. Correlates: living in the same place/culture all their life, doing the same activity all their life e.g. high school into undergrad into phd without anything in between, never having faced death, never having questioned or been exposed to failure modes of our social reality, etc etc.
I know it’s outrageously offensive to say, but at least some instincive part of me has stopped perceiving these beings as actual people. They’re just sort of fluttering around, letting every little thing knock them off-balance, because they lack the heft to keep their own momentum going, no will of their own. Talking to these people I’m more and more having the problem of the inferential distances being too high to get any communication beyond social niceties going. You must think I’m super arrogant, but I’m just trying to communicate this important, hard to grasp idea.
Most people don’t ever become solified in this way (the default mode for humans seems to be to shut off the vulnerable surface layer entirely as they age), but that’s yet another digression...
All of this is a prelude to saying that I’m confident I wouldn’t fall for these AI tricks. That’s not a boast, or put-down, or hubris, just my best estimation based on what I know about myself. I’d consider being vulnerable in this way as a major character flaw. This not only applies to interacting with an AI btw, but also with actual humans that follow similar exploitative patterns of behavior, from prospective lovers, to companies with internal cultures full of bullshit, all the way up to literal cults. (Don’t get me wrong, I have plenty of other character flaws, I’m not claiming sainthood here)
As other people have already pointed out, you’ve been shifting goalposts a lot discussing this, letting yourself get enchanted by what could be, as opposed to what actually is, and this painfully reminds me of several people I know, who are so open-minded that their brain falls out occasionally, as the saying goes. And I don’t think it’s a coincidence that this happens a lot to rationalist types, it seems to be somehow woven into the culture that solidifying and grounding yourself in the way I’m gesturing at is not something that’s valued.
Relatedly, in the last few years there’s been several precipitating events that have made me distance myself a bit from the capital-R Rationalist movement. In particular the drama around Leverage research and other Rationalist/EA institutions, which seem to boil down to a lack of common sense and a failure to make use of the organizational wisdom that human institutions have developed over millenia. A general lack of concern for robustness, defense-in-depth, designing with the expectation of failure, etc. The recent FTX blow-up wrt EA also has a whiff of this same hubris. Again, I don’t think it’s a coincidence, just a result of the kind of people that are drawn to the rationalist idea-space doing their thing and sharing the same blind spots.
As long as I’m being offensively contrarian anyway, might as well throw in that I’m very skeptical of the median LW narrative about AGI being very near. The emotional temperature on LW wrt these topics has been rising steadily, in a way that’s reminiscent of your own too-generous read of “Charlotte”’s behavior. You can even see a bunch of it in the discussion of this post, people who IMO are in the process of losing their grasp on reality. I guess time will tell if the joke’s on me after all.
I’m not buying the premise. Passing the Turing test requires to fool an alert, smart person who is deliberately probing the limits of the system. ChatGPT isn’t at that level.
A specially tuned persona that is optimized for this task might do better than the “assistant” persona we have available now, but the model is currently incapable of holding a conversation without going on long, unwanted tangents, getting trapped in loops, etc.
In many cases, it is being spent in net-negative ways.
For example, many journals have copyeditors that do the paper layout for you, but frequently introduce mistakes in mathematical notation and other subtle issues like that. Result: the author has to proofread the editor’s work multiple times before giving their approval. A latex compiler does better work for free. A friend of mine literally spent months going back and forth with a for-profit journal over issues like this.
Why does Elsevier persist in this obviously inefficient way of doing things? because it keeps them in the picture. They deliberately cultivate the impression that doing a paper layout is something scary and hard. This prevents academics from gaining agency for themselves (especially in less technical fields where latex is still uncommon). Meanwhile, OA journals just put up a latex template for download, and give your final submission a quick once-over to make sure you’re following their formatting guidelines.
Other needless expenses (epistemic status: speculative). Marketing budgets. Buying up smaller journals. Funding needless bloat in conferences, websites and other academic tools. Lobbying and legal work. Contract negotiation, billing departments. Shipping stacks of journals that sit in university libraries unread for a few years before being thrown away.
Sure, let me do this as an exercise (ep stat: babble mode). Your predicions are pretty sane overall, but I’d say you handwave away problems (like integration over a variety of domains, long-term coherent behavior, and so on) that I see as (potentially) hard barriers to progress.
2022
2022 is basically over and I can’t get a GPT instance to order me a USB stick online.
2023
basically agree, this is where we’re at right now (perhaps with the intensity turned down a notch)
2024
you’re postulating that “It’s easy to make a bureaucracy and fine-tune it and get it to do some pretty impressive stuff, but for most tasks it’s not yet possible to get it to do OK all the time.” I have a fundamental disagreement here. I don’t think these tools will be effective at doing any task autonomously (fooling other humans doesn’t count, neither does forcing humans to only interact with a company through one of these). Currently (2022) chatGPT is arguably useful as a babbling tool, stimulating human creativity and allowing it to make templating easier (this includes things like easy coding tasks). I don’t see anything in your post that justifies the implicit jump in capabilities you’ve snuck in here.
broadly agree with your ideas on propaganda, from the production side (i.e. that lots of companies/governments will be doing lots of this stuff). But I think that general attitudes in the population will shift (cynicism etc) and provide some amount of herd immunity. Note that the influence of the woke movement is already fading, shortly after it went truly mainstream and started having visible influence in average people’s lives. This is not a coincidence.
2025
Doing well at diplomacy is not very related to general reasoning skills. I broadly agree with Zvi’s take and also left some of my thoughts there.
I’m very skeptical that bureaucracies will be the way forward. They work for trivial tasks but reliably get lost in the weeds and start talking to themselves in circles for anything requiring a non-trivial amount of context.
disagree on orders of magnitude improvements in hardware. You’re proposing a 100x decrease in costs compared to 2020, when it’s not even clear our civilization is capable of keeping hardware at current levels generally available, let alone cope with a significant increase in demand. Semiconductor production is much more centralized/fragile than people think, so even though billions of these things are produced per year, the efficient market hypothesis does not apply to this domain.
2026
Here you’re again postulating jumps in capabilities that I don’t see justified. You talk about the “general understanding and knowledge of pretrained transformers”, when understanding is definitely not there, and knowledge keeps getting corrupted by the AI’s tendency to synthesize falsities as confidently as truths. Insofar as the AI can be said to be intelligent at all, it’s all symbol manipulation at a high simulacron level. Integration with real-world tasks keeps mysteriously failing as the AI flounders around in a way that is simultaneously very sophisticated, but oh so very reminiscent of 2022.
disagree about your thoughts on propaganda, which is just an obvious extension of my 2024 thoughts above. I also notice that social changes this large take orders of magnitude longer to percolate through society than what you predict, so I disagree with your predictions even conditioned on your views of the raw effectiveness of these systems.
“chatbots quickly learn about themselves” etc. Here you’re conflating the regurgitation of desirable phrases with actual understanding. I notice that as you write your timeline, your language morphs to make your AIs more and more conscious, but you’re not justifying this in any way other than… something something self-referential, something something trained on their own arxiv papers. I don’t mean to be overly harsh, but here you seem to be sneaking in the very thing that’s under debate!
Ports and shipping companies have complex contracts covering all kinds of services. Container storage for a few days is just part of normal operations, included in the docking fee. Longer-term storage is often included as well.
Even if your “free storage” clause has some built-in limit, it is not trivial to invoke such an exception. Keep in mind that this is an industry with few players where people know each other, and have to work together on a long-term basis. If the port tries charging, then maybe the shipping company replies “well, technically you guys aren’t upholding the guaranteed turn-around times, which we have been very gracious about” etc etc.
And yes, $100 per container-day is a ludicrous number. It’s meant to exert pressure, not to clear the market. It’s easily more than 10x of a reasonable fee (I can’t put an exact figure on this, since it’s usually not charged on a per-container basis, and varies a lot by location).
I’m a bit confused about the situation you’re describing. The shipping company promised to deliver the container at B, so they arrange for the proper routing. Any transfer port along the way has a contract with the shipping company and transfers the containers as directed by them.
Delivery is not guaranteed to happen on the exact predicted schedule though, which is what you might be talking about. For various reasons, containers often end up missing their connection and waiting for the next available ship (of the same service going to the same destination). It also happens occasionally that a ship has to alter its route (e.g. skipping a port due to congestion), which leads to additional unplanned transfers. All of this is ok, because the contract between shipping company and customer is on a “best effort” basis, i.e., it’s ok to deliver with a delay if it’s unavoidable.
If a reefer container has issues while on board a ship, it will be unloaded at the next possible port, because this makes it possible to save the cargo by moving it to another container in the event of total failure. But we want to avoid this (because it involves high costs and breaking the container seal) so we’ll try to repair the container while the cargo remains inside. If repaired successfully it can continue its journey on the next available ship.
Hope that makes everything clear.
Yes, there are corridors between the stacks to allow access to the containers. And you’re proposing to get rid of them and just make one huge, contiguous block of boxes, yes?
Not all those containers are empty (I’d actually expect most of them to be full if we’re seeing historically high import volumes). Even if the boxes were empty, they wouldn’t be fungible. They belong to different shipping companies, there are different types of containers, and even within the same type they have different ratings in terms of being foodgrade, maximum weight, etc. Sometimes they’re also rented, which can affect priorization and where you want to send them.
But the biggest problem I see is that not all the empties at the port can be in the same place, because there are many ships docking at different quays, and the containers to be loaded must be pre-stacked (in a specific order) before the ship arrives, to make the operation fast and efficient.
It’s an interesting idea though, I’m sure some amount of improvement is possible.
This story has been adapted into a (relatively faithful) movie: https://en.wikipedia.org/wiki/North_Face_(film)
I agree that life is like that. However, the game still violates consent, the same way as if I assaulted you on the street because I think it’s good preparation for being assaulted “for real”.
To me, this game falls in the same category as gift giving, surprise parties, pranks, rude/aggressive jokes etc. There needs to be a meta-level agreement that this kind of thing is ok, even though “being a surprise” is an essential part of the thing itself.
It seems to me that you are attempting to write a timeless, prescriptive reference piece. Then a paragraph sneaks in that is heavily time and culture dependent.
I’m honestly not certain about the intended meaning. I think you intent mask wearing to be an example of a small and reasonable cost. As a non-american, I’m vaguely aware what costco is, but don’t know if there’s some connotation or reference to current events that I’m missing. And if I’m confused now, imagine someone reading this in 2030...
Without getting into the object-level discussion, I think such references have no place in the kind of post this is supposed to be, and should be cut or made more neutral.
I read your original post and I understood your point perfectly well. But I have to insist that you’re typical-minding here. How do you know that you were exactly at my stage at some point? You don’t.
You’re trying to project your experiences to a 1-dimensional scale that every human falls on. Just because I dismiss a scenario, same as you did, does not imply that I have anywhere near the same reasons / mental state for asserting this. In essence, you’re presenting me with a fully general counterargument, and I’m not convinced.
24h is not enough time to produce the physiological changes typical for fasting. That requires ~72h. I’m not saying that skipping a meal or two, doing intermittent “fasting” and so on aren’t good things, you just shouldn’t confuse them with an actual fast.
It takes around 12 hours until your stomach is empty (depending on what you ate before obviously). At ~24h, your body’s glycogen reserves deplete, but the fat burning process is just beginning in earnest. This is the hardest point IME. At ~72h, you’re in full starvation mode and symptoms should start to decrease. I’ve never gone longer than 72h though.
W.r.t. blood sugar, you usually have a dip around expected mealtimes (your body adapts to your habits and secretes insulin in anticipation of eating), but the truly hard period is between hours 24 and 72.
Opposition to vaccines comes in various gradings and flavors. I think it’s pretty clear that initial vaccination has positive expected value for almost everyone. But it’s much less clear if a booster shot can claim the same value proposition: it has a much smaller direct benefit to the bostee, so the argument for it is mostly based on solidarity. Is that enough to make it mandatory?
You can also be pro vaccines but against public policy about them. There has been some amount of goalpost shifting, e.g., in the scope and duration of covid passes. In some places, there was an implicit deal on the table: get vaccinated and you will be allowed to visit restaurants and bars. But a few months later this was changed to requiring recent vaccination (4 months instead of 12). Regardless of whether you think that’s a good motivation to get vaccination, the promise has clearly been broken.
I think that there’s a misunderstanding here: the rule did not apply to the Port of Long Beach, which has always stacked to reasonable heights, but to container yards in surrounding areas. On top of this I assume it applies only to small container yards (the kind owned by a trucking company that’s little more than a patch of dirt to park their truck fleet), because you can’t possibly operate a dedicated yard (the kind where major shipping companies keep their stock at, thousands of boxes) while stacking only two high.
This is not a nitpick, since the images of containers stacked 5 high make you imagine that storage capacity in the whole area just increased 250% overnight. I am highly confident this is not the case. The increase might only have been a few percentage points. Can’t tell without more details on the local conditions.
Given this, I’d like to see some actual evidence that this is a panacea, as opposed to just a first step (still important!) in the long tower-of-hanoi logistics nightmare that clearing the bottleneck actually is.
One thing that is clear to me: untying this kind of knot almost certainly requires coordination from a higher authority than the individual companies (shipowners, truckers, exporters/importers), because those groups have goals and incentives that are at odds with each other in a situation like this.
(Source: I worked in the container shipping industry)