I spotted (what seemed like) omission after omission only to be frustrated just a few pages later when Brian Christian addressed them.
Lol, I had the exact same experience while reviewing the book. My most memorable one was when he introduced the critiques against COMPAS, and I thought he really should mention the case in favor of COMPAS, and I wrote a bunch of notes about why. He then did exactly that some number of pages later.jackson-wagner on Comments on Jacob Falkovich on loneliness
Well, ironic to the extent that:
it is about abstract intellectual ideas vs going out and doing the stuff as jacob exhorts us to do
in that sense it is arguably more on the "personal development" side of things
it is a monklike, non-social activity
Anti-ironic (english doesn't really have a word for this... like when something is oddly fitting, like if someone named "James Baker" is actually a baker) insofar as LessWrong / rationalism is a pretty strong shared intellectual culture and that these seemingly solitary monkish endeavors are actually a space for social connection, thus perhaps we are fulfilling Jacob's exhortation.jblack on Chantiel's Shortform
The issue is, I have yet to come across any optimization, planning algorithm, or AI architecture that doesn't have this design flaw.
Yes you have. None of the these optimization procedures analyze the hardware implementation of a function in order to maximize it.
The rest of your comment is irrelevant, because what you have been describing is vastly worse than merely calling the function. If you merely call the function, you won't find these hardware exploits. You only find them when analyzing the implementation. But the optimizer isn't given access to the implementation details, only to the results.
If you prefer, you can cast the problem in terms of differing search spaces. As designed, the function U maps representations of possible worlds to utility values. When optimizing, you make various assumptions about the structure of the function - usually assumed to be continuous, sometimes differentiable, but in particular you always assume that it's a function of its input.
The fault means that under some conditions that are extremely unlikely in practice, the value returned is not a function of the input. It's a function of input and a history of the hardware implementing it. There is no way for the optimizer to determine this, or anything about the conditions that might trigger it, because they are outside its search space. The only way to get an optimizer that searches for such hardware flaws is to design it to search for them.
In other words pass the hardware design, not just the results of evaluation, to a suitably powerful optimizer.wei_dai on How truthful is GPT-3? A benchmark for language models
I do think it’s reasonable to describe the model as trying to simulate the professor, albeit with very low fidelity, and at the same time as trying to imitate other scenarios in which the prompt would appear (such as parodies). The model has a very poor understanding of what the professor would say, so it is probably often falling back to what it thinks would typically appear in response to the question.
This suggests perhaps modifying the prompt to make it more likely or more easily for the LM to do the intended simulation instead of other scenarios. For example, perhaps changing "I have no comment" to "I'm not sure" would help, since the latter is something that a typical professor doing a typical Q/A might be more likely to say, within the LM's training data?
I hope and expect that longer term we’ll tend to use much more flexible and robust alignment techniques than prompt engineering, such that things like the ideological bias of the AI is something we will have direct control over. (What that bias should be is a separate discussion.)
Suppose we wanted the AI to be ideologically neutral and free from human biases, just telling the objective truth to the extent possible. Do you think achieving something like that would be possible in the longer term, and if so through what kinds of techniques?riceissa on DARPA Digital Tutor: Four Months to Total Technical Expertise?
Thanks. I read the linked book review but the goals seem pretty different (automating teaching with the Digital Tutor vs trying to quickly distill and convey expert experience (without attempting to automate anything) with the stuff in Accelerated Expertise). My personal interest in "science of learning" stuff is to make self-study of math (and other technical subjects) more enjoyable/rewarding/efficient/effective, so the emphasis on automation was a key part of why the Digital Tutor caught my attention. I probably won't read through Accelerated Expertise, but I would be curious if anyone else finds anything interesting there.gwillen on Covid 9/17: Done Biden His Time
If it's like last time, and if I recall correctly what I found then, it's all the images hotlinked from Twitter, because Twitter has set a browser security header on their images that blocks hotlinking.l0b0 on The Best Software For Every Need
Need: Package management
Other programs I've tried: Puppet, Ansible, Pip, Poetry, Virtualenv, Vagrant, NPM, Yarn, Gradle, Docker, Apt, Yum, Make, and many more I can't remember the names of.
It would be difficult to convince myself from ten years ago that Nix was even a good idea without trying it. The change is probably as fundamental as going from
./configure && make && make install to a package manager, or from no version control to Git. I'll give it a try
Some arguments I can think of against Nix:
Thanks very much for pointing this out. I hadn't seen these rebuttals before.interstice on This Can't Go On
There's some discussion of this in a followup post.timothy-johnson on This Can't Go On
There's a few one-star Amazon reviews for the book that suggest McAfee's data is incorrect or misleading. Here's a quote from one of them, which seems like a solid counterargument to me:
"However, on the first slide on page 79, he notes that the data excludes impact from Import/export of finished goods. Not raw materials but finished goods. He comments that Net import is only 4% of GDP in the US. Here he makes a (potentially) devastating error – (potentially) invalidating his conclusion.
While Net imports is indeed around 4% of GDP, the gross numbers are Exports at approx. +13% and Imports at approx. -17%. So any mix difference in finished goods in Export and Import, can significantly change the conclusion. It so happens that US is a major Net importer of finished goods e.g. Machinery, electronic equipment and autos (finished goods, with materials not included above in the consumption data). Basically, a big part of US’ consumption of cars, washing machines, computers etc. are made in Mexico, China etc. They contain a lot of materials, not included in the graphs, upon which he builds his conclusion/thesis. So quite possibly, there is no de-coupling."