In books about the Culture sci fi universe such things are described a couple of times. E.g. in the novel “Use of Weapons” the “crew” (the ship is fully automated, so more like permanent passengers) of a ship deliberately weaken their immune system to basically get a seasonal cold just for the experience, which otherwise could not happen due to their genetically enhanced immune system.
Richard Horvath
Karma: 285
Is the data reliable? I just did a search on PornHub, xHamster and iXXX for the following keywords:
violence, incest, rape, torture
Results:
Pornhub returns a warning “Your search could be for illegal and abusive sexual material...etc” (it is a longer description), returning no videos for ‘violence’, ‘incest’ and ‘rape’. It does return results for ‘torture’ though.
The cases is somewhat similar for xHamster and iXXX, except they do not return any special message for the first three, but also do not return any results. They do work fine with violence though.
It does seem there is some official or unofficial policing.
Any opinion on this regarding being a somewhat good solution?
https://www.lesswrong.com/posts/Q3huo2PYxcDGJWR6q/how-to-corner-liars-a-miasma-clearing-protocol
In my personal experience, LLMs can speed up me 10X times only in very specific circumstances, that are (and have always been) a minor part of my job, and I suspect this is true for most developers:
If I need some small, self-contained script to do something relatively simple. E.g. a bash script for file manipulation or an excel VBA code. These are easy to describe just with a couple of sentences, easy to verify, I often don’t have deep enough knowledge in the particular language to write the whole thing without looking up syntax. And more importantly: as they are small and self-contained, there is no actual need to think much about the maintainability, such as unit testing, breaking into modules and knowing business and most environment context. (Anything containing regex is a subset of this, where LLMs are game changer)
If I need to work on some stack I am not too familiar with, LLMs can speed up learning process a lot by providing working (or almost working) examples. I just type what I need and I can get a rough solution that shows what class and methods I can use from which library, or even what general logic/approach can work as a solution. In a lot of cases, even if the solution provided does not actually work, just knowing about how I can interact with a library is a huge help.
Without this I would have to spend hours or days either doing some online course/reading documentation/experimenting.
The cavet here is that even in cases when the solution works, the provided code is not something that I can simply add into an actual project. Usually it has to be broken up and parts may go to different modules, refactored to be object oriented and use consistent abstractions that make sense for the particular project.
Based on above, I suspect that most people who report 10X increase in software development either:
Had very low levels of development knowledge, so the bar of 10X is very low. Likely they also do not recognize/not yet experienced that plugging raw LLM code into a project can have a lot of negative downstream consequences that you described. (Though to be fair, if they could not write it before, some code is still better than no code at all, so a worthy trade-off).
Need to write a lot of small, self-contained solutions (I can imagine someone doing a lot of excel automation or being a dedicated person to build the shell scripts for devops/operations purposes).
Need to experiment with a lot of different stacks/libraries (although it is difficult to imagine a job where this is the majority of long-term tasks)
Were affected by the anchoring effect: where they recently used LLMs to solve something quickly, and extrapolated from that particular example, not from long-term experience.
This is only true if you restrict “nobility” to Great Britain and if you only count “nobles” those who are considered such in our current day. This is a confusion of the current British noble title (specifically members of “Peerage of Great Britain”) with “land owning rentier class that existed before the industrial revolution”. For our discussion, we need to look at the second one.
I do not have specific numbers of UK, but quoting for Europe from wikipedia (https://en.wikipedia.org/wiki/Nobility#Europe):
”The countries with the highest proportion of nobles were Polish–Lithuanian Commonwealth (15% of an 18th-century population of 800,000[citation needed]), Castile (probably 10%), Spain (722,000 in 1768 which was 7–8% of the entire population) and other countries with lower percentages, such as Russia in 1760 with 500,000–600,000 nobles (2–3% of the entire population), and pre-revolutionary France where there were no more than 300,000 prior to 1789, which was 1% of the population (although some scholars believe this figure is an overestimate). In 1718 Sweden had between 10,000 and 15,000 nobles, which was 0.5% of the population. In Germany it was 0.01%.[46]In the Kingdom of Hungary nobles made up 5% of the population.[47] All the nobles in 18th-century Europe numbered perhaps 3–4 million out of a total of 170–190 million inhabitants.[48][49] By contrast, in 1707, when England and Scotland united into Great Britain, there were only 168 English peers, and 154 Scottish ones, though their immediate families were recognised as noble.”
Based on above, I think expecting 1% to be landed rentier is a conservative estimate for 18th century for whole Europe. Even if we go with one tenth of this, expecting 0.1% of the population to retain this (which would imply that their population dropped while all other classes increased dramatically), would mean about 68 thousand people in the UK, and over 700 000 in whole Europe.
AND they are expected to live off from rents of land. I doubt that living of land rents is true for the majority of the 1500 current British nobles you referred to.
Not fully. Most of the nobility is gone. Only like 0.01% remains maybe what could be called “rentier”, or even less compared to what had been before.
You lost me at “Bywayeans generally save up enough to move out of their parents’ houses around age 9”. Likely you will lose most people at “ancap”.
But these are only minor things, and I can imagine that it would be possible to pull up some plausible explanation or just to revise one or the other, and keep the core message. I think the main issue with your description Byway are in these:
″...even though Bywayeans are smarter than Earthlings...”
“Bywayeans have a lot of energy...”
″...they love working and innovating...”
These imply that they are already better then us, and would run the world better even in the structure we have on our real Earth.
The message of Dath Ilan is that they and us are the same people, with the same genetic heritage, hence intelligence and vices. But they can still do better, just by having better institutions/methods for cooperation, hence, we can get better too! If you give special power to the inhabitants of your alternate Earth, it may explain why are not stuck in the same mire as we are, and imply we cannot learn from their success.
Thanks. Some of these seem to be good ideas indeed.
Isn’t it an issue though that in a lot of these cases you/your colleagues have large direct impact on the outcome, and knowing the prediction itself can change your/their behaviour? E.g., if you have 90% public prediction that Adam will complete task X and 10% that Bob will complete Y, their knowledge of these can impact how they approach their work. Maybe Bob will .
That being said, it can still be an effective productivity tool, but it no longer directly measures the outcome of the stated prediction. Instead, it will become a reflective statement (“The probability of Bob completing Y provided he knows this prediction is 10%”).
Of course they are not literally the same, but in the context of the particular article we are commenting on, both Aldi, Lidl and Walmart are large bureaucratic organizations (specialized in retail). They all benefit vastly from economics of scale and to increase output they have to create new stores and hire more people to run them (and increase the throughput of their supply chain).
According to wikipedia, Aldi has ~ 273 000, Lidl has ~376 000 employees.
These are not local mom & pop stores, and scale similarly as Walmart, even if not concentrating on broadness of selection.
I see that A Deepness in the Sky is a prequel to A Fire Upon the Deep, but was written later. Does it make any impact which one I read first? Deepness seems to be more interesting to me, but I generally I prefer to read things in the suggested order.
“A team of 3 top research ML engineers with fine-tuning access to GPT-4o (including SFT and RL), $10 million in compute, and 1 year of time could use GPT-4o to surpass typical naive MTurk performance at ARC-AGI on the test set while using less than $100 per problem at runtime (as denominated by GPT-4o API costs).”
But doing so would tune that GPT-4o to be less good at other tasks, wouldn’t it? Isn’t this way of solving just Goodharting the metric and actually pushing the LLM away from being “General Intelligence”?
That being said, this is a great job you have done just over a week. Thank you for your contribution for science.
Children spending 300 hours per year learning math, on their own time and via well-designed engaging video-game-like apps (with eg AI tutors, video lectures, collaborating with parents to dispense rewards for performance instead of punishments for visible non-compliance, and results measured via standardized tests), at the fastest possible rate for them (or even one of 5 different paces where fewer than 10% are mistakenly placed into the wrong category) would probably result in vastly superior results among every demographic than the current paradigm of ~30-person classrooms.
This sounds great.
Googling “math tutor ai” already gives a bunch of competing suggestions. I wonder how well they work though.
Thank you for sharing. For me your original talk is more convincing, and your Death Star strike plan is the one I would be more willing to follow rather than movie one.
I suppose this is an area where one can have a strong conviction on how things ought to be done, assuming other smart people think the same way, but in reality the way they think about it is closer to the base rate, so communication towards them should be as such.
A lot of important technology do not exists yet for the above.
You assume independent movement and coordination of said movement in a hostile area, while we barely arrived at self-driving cars in well governed, 1st world areas (available only very few places afaik due to not being able to demonstrate high enough reliability to be convincing to lawmakers).
Point-to-point laser communication would be a great solution indeed, but that would also be a great solution for a bunch of other military applications. Yet it is not used, as we do not have reliable solution for working with it in case of moving objects (apart from satelites), too much coordination is needed to find the end points.
There is no software system currently that is even close to completing the requirements.
Let me offer a different point of view on the whole question:
A suicide drone is just a missile. Until recently there was just no way to propel and guide an explosive charge accurately, reliably and cheaply other than using a rocket engine.
A recon drone is just a helicopter. Until recently there was just no way to propel and guide a good enough observer and transmit the information reliably and cheaply other than a human carrying object using gas turbines.
What happened is now we have better battery and material technology, and way better (smaller) electronic devices (and optimized global supply chains). The cost advantage against missiles is mostly due to inertia in the military-industrial complex: most missiles in inventory were designed against targets with different size and performance parameters. You need a a million dollar AMRAAM missile to intercept an Su-30 flying 80 kms away, maybe flying at 2 mach or at 40 000 feet or pulling 9g maneuvers while dispensing chaff, flares and using EW. (and the missile carrying aircraft may have already pulled the same speed and maneuvers, and may have took off from the desert with 45 degrees Celsius and arrived to − 30 degrees Celsius flight attitude in the next couple of minutes, yet the missile must stay safe and reliable).
The infantryman did not get replaced. They just got missiles available in large numbers, and their own miniature recon helicopters in every bush. It is harder to hide and there are more precision fires to throw at you after getting spotted.
Thanks for writing the post.
Regarding the challenge: I suppose one of the errors made by you was using bad sources in some case, e.g. using the msn.com article for supporting Russian losses.
Regarding the article: although what is presented there is generally valid, it do not feel my understanding/defense increased in a meaningful way, as
The suggested steps are more-or-less what I already do.
Following them still gets me nerd-sniped due to the “bullshit asymmetry” principle already mentioned in another comment. If in a hostile dispute space, where one (or more) opponents are pushing an agenda and I have limited time (equal to or less than what they have) for counter-argument, this cannot be really pulled off. It is possible that my expectations are just too high though.
I think what I really liked in your dialog with Isusr was it showing how such techniques looked like when someone was using them himself. I haven’t experienced that before, especially not with that level of clarity and purposefulness.
Thanks for pinging me. Haven’t noticed it yet, will read it now.
I think something along the lines of “Defense Against the Dark Arts” with actionable steps on recognizing and defusing them (and how to practice these) would be great. If you feel like you have the energy and time, more articles on offensive usage (practice) and on theoretical background (how to connect your practical experience to existing LW concepts/memes) would be also nice. But I think the first one (defense) would be the most useful for LW readers.
I think this is an important addition to the site. There had been articles before about the “dark” side/arts, but I think this is the first one where the examples are not thought experiments and abstractions, but actual real world experience from an actual user.
It is helpful for understanding politics.
Well, the disturbed protagonists in the Culture series (as in: books, and in the whole of the fictional universe) are usually not from the “Culture” (one particular civilizations within the whole fictional universe), but outsiders hired to act as agents.