Posts
Comments
Interesting, thanks!
Thanks. I think I get it now. (at least one of) my confusion was something between confusing a "transformer run" and "number of FLOPS".
And I get the thing about cost, that's what I meant but I articulated it poorly.
Heh, I actually think it's answered here.
Got it, thanks!
But to process the 1001st input token, you also need to load all the 1000 tokens in memory, forming the cache (it does happen in one step though). And for each new output token, you surely don't dump all the existing KV cache after each generation, only to load it again to append an extra KV vectors for the last generated token. So isn't the extra work for output tokens just that the KV cache is accessed, generated, expanded, one token at a time, and that's where the "more work" come from?
Is there any reason why this would imply the ratio of pricing of output:input tokens being commonly something like 3:1?
Thanks for the answer, I appreciate it!
Intuitively, it seems that output tokens should be more expensive. The autoregressive model has to run once for each output token, and as these runs progress, output tokens gradually become a part of the input (so the last token is generated with context being all input and almost all output).
I agree with the intuition, but I think that's where I am confused. Thanks to the KV cache we do not run the new input sequence (previous sequence + last generated token) through the encoders (as we do for the input sequence during prefill). It's all cached (from prefill + from the last token generation for that sequence+token). So... I don't know - it doesn't feel like the output tokens are more expensive in this case (you run "once", the same way as you run "once" for every input token)?
I think they do amortize their costs among all uses. A number of runs (number of output tokens) multiplied by a (varying) cost of the each run is unlikely to be close to linear.
Do you mind saying more about this? I am not sure what you mean. I.e. some pay more and some pay less (e.g. heavy hitters pay less while small prompters pay comparatively more per token?)
Even though some commenters mentioned some issues with the article, I really want to appreciate the attempt and being upfront with the estimates. It's very relevant for the thing I am now trying to figure out. As I have almost no intuitions about this except about some raw FLOPS, it pointed to important flaws my analysis would have. There are not many public sources that would explain that [are not a book or don't require me reading one-to-many to understand it]
Yes, but to defend (hehe) OP, he seems to be fully aware of that and addresses that explicitly in the linked article (which is also excellent, like this one):
In part because of those aforementioned stats on the frequency of guilty pleas, public defenders have garnered a reputation for being trial-averse, for pressuring clients to cop a plea just to keep the machine humming along. I think this reputation is ill-deserved. It’s completely counter to my own experience, at least, as few things are talked about with as much awed respect among one’s public-defender peers as the number of trials you have accumulated. It’s the functional equivalent of an attorney’s XP level.
Thanks for the feedback and the encouragement, I will incorporate these.
Btw. for questions 2-4 there is an intentional redundancy.
(slightly tangential) I think people are doing a terrible bucket error with competency, and that is that people overestimate how are others competent across all dimensions. I.e. it's often enough for a person to show great competency in providing vision, and we assume that person also needs to be great in leadership or management, and people are shocked it's not the case. Other examples:
- scientist being great in research, therefore they need to be great teachers or college management
- doctors are great in diagnosing, therefore they need to be great surgeons
- programmer being great in programming, therefore they need to be great in leading the rest of team or mentoring ...
Hey. I decided on a private school with more of a "democratic approach". I unfortunately wasn't able to find suitable tutors etc.
I am also trying to process what ChatGPT-like platforms will do with the landscape. E.g. my partner is using coding almost exclusively with ChatGPT and it's outstanding. Kids gonna follow IMHO.
Thanks Duncan, I really appreciate you posting this, even though you are unsure about how exactly it all fits together. I am still glad to read it in this version, likely because you are quite clear about it, and not "leaving it as an exercise for the reader" to figure out where things do fit together and where they don't (or worse, trying to make it more profound).
All of these might be stating obvious to some of you, but I am trying to clarify my thoughts and maybe some people will find it useful or correct me. At least part of this relates to (by me endorsed) aphorism (?) of "everyone is a mess", or less controversially, "everyone is struggling". Something something hedonic treadmill / adaptation - people will somehow struggle the same regardless of the bubble size, then adapt to what they were used to before by overcoming the main challenges (and learning how to deal with them) and then also "reprioritize" costs. I would definitely self-report that happening during my life. I do think this realization is important in a way I relate to others, including e.g. my daughter - the things she struggles with might seem trivial to me, but are not trivial to her - on contrary, they probably feel to her about the same magnitude as my "bigger" problems look to me. Same for everyone, everywhere.
Almost like I had some capacity of how much I can deal with stuff, and I always fill in this capacity with the things around me (my bubble?). Something like doing busy work if you don't choose what to let into your to-do list. Or something closer to "I can always do 10 things, and their size doesn't matter" (bear with me, I know this is not true and I do indeed sometimes work on a single project because it's eating all of my capacity). If I let "bigger things" into it, I will be dealing with bigger things, while also just leaving some stuff behind me or "go wrong" in a way I wouldn't allow in the N-1 bubble (something like not dealing with every single fuckup in my work, starting to take Uber instead of public transport etc).
It doesn't hold entirely, I have clearly seen people who just couldn't deal with e.g. a promotion (because it was too much for them at the given time), or, similarly, people who said they want to do more / be more ambitious but can't for some objective reasons (like a physical disease).
Dunno.
Oh, I really enjoyed reading this, this is so LW-rationality-curiosity-boggle-at-things post. Thanks!
Thanks, that's helpful.
Also, kudos for Lily to know active listening and being awesome.
I am curious about how you introduced money to your kids? Do you have some "framework" for that? I did a small research and didn't end up with any really novel ideas (I am happy to share my findings and conclusions, but it's a fairly small page in roam).
Basically, what I want to do with my daughter:
- give her an opportunity to earn money by doing age-appropriate jobs (helping me to clean the kitchen), but not for things that she's expected to do for "reasonable reasons", such as cleaning her toys in common areas, rather lower allowance (but some)
- make it clear that she can make more by doing more elaborate stuff (for which she might have to learn a few things first - happy to help), encourage her to come up with new ideas by herself (entrepreneurship!)
- introduce saving mechanism, e.g. giving her interest for money she's gonna put into my pocket
- she can make more by betting and bonds as per your posts
- she can do what she wants with her money (as long as it's safe, but can't really think about anything I would consider "forbidden"), let her do mistakes
I checked the marginal revolution repost and saw a few things that could go wrong...
Any other ideas?
This is an excellent post. I have been doing bets with my 4y old daughter already as well (and I am following your projects for a quite some time already)!
Yeah, that's useful. Agree on the assessment, I want to give it a shot with one of those Bridgelux Vesta Thrive thing, it sounds like a good hobby project I would like to try. If that happens, I would do a post about it here.
By the way, I asked about this setup on reddit. They also recommend some custom COBs, which seems to be the most powerful solution, but isn't as practical as strips.
These look very promising, ship to Europe too. Extra high-CRI, very powerful (up 2600 lumens/m) and even dimmable? Wow. A bit unfortunate they are 5x times as much expensive than other high-CRI high-power led strip.
I am glad there are more posts on this. Are there any reasons why not considering LED strips at all? When installed properly with the "milk" diffuser and as indirect lightning, it's IMO quite nice and effective. They seem to be powerful enough (20W are about 2k lumens/m), can be also found in high-CRI variants, less expensive, various CCT, dimmable, etc. I am considering using them in a new house. Basically, multiple parallel led (with different CCT, like 3000, 4500, 6400) strips diffused against a wall/ceiling, controlled via smart relays and incorporated in home automation (so physical switches turn e.g. just the one with 3000K after ~9pm).
I need to validate that in some studio, but my hope is that e.g. 50 000 lumens from 6x4 meters of high-CRI led strips of 3 different CCT diffused over a wall would provide sufficient daylight feeling + not taking any space, generate many shadows, looks nice and are OK to look at. If I could make it dimmable, it would be great, but there seem to be tradeoffs (dimmable ones are not that good on CRI apparently).
Btw. there are apps for phones with light detectors which seem to follow basic physics (like showing 4x times smaller number when being 2x times further from the source)
I just recently made my new ugly but sufficient lumenator (a combination of 6400K and 4500K CRI95+ 1500 lumen bulbs, still adding additional ones) and also managed to integrate a WiFi relay with Smart Bulbs to my home automation via Home Assistant with a fallback to manual switch. I also tried Home Assistant's "flux" component which sets the colour and brightness based on the sun position but replaced that idea with being less-sun-dependent. I want to have a lot of light until e.g. 8 pm regardless of the light outside and only then start dimming to red. Simple variant is plain automation with templating, hard but dynamic.
I am so glad this question is here, as it's very relevant to my post a few weeks back about Effective Children Education.
By the way, I recommend following Duncan Sabien (referenced in the post below) on Facebook, he has good posts about children edu, e.g. his speech for sixth-graders (referenced by someone else here - but she picked the good parts).
As mentioned below, Julia Galef also sometimes mentions something related, but I haven't found much
Hi! This is an excellent answer, thanks.
[...] I believe your questions relate to all three ways to different extents (although the title of the post leans towards the HOW types of questions), but I found it useful to differentiate between these issues in order to make sure my time, money and efforts are well spent.
Needless to say that I updated significantly in the past month since I posted this question and the "Why" and "What" has definitely enlarged. I agree with you that it's a useful framework to have. I am also thankful for the practical bits.
I would like to close my comment by emphasizing that children are born rational beings, but they just start taking over from their parents when they provide them with irrational explanations for things that happen or might happen to them (one of my favorites, here in Romania, is that "If you don't stop crying, a big bad wolf will come and get you"), or they build defenses of their own that are irrational, to protect themselves from being hurt. I believe that our job as parents is, therefore, to be rational and predictive in relation to them, as well as gently making them aware of the instances when they use irrationality to run away from their feelings.
I probably understand what you mean here and agree at least with my steelman version of your argument. But I would also like to emphasize that I want the "other" people which are important for me to understand "rationality" that I can't imagine how one could "born" with. Examples are all the great knowledge of giants on which shoulders we stand on (like, say, probability), "simply" results of scientific progress (like how minds work, or some biases...), or practical guide of "you are a human, try to work with that fact in the best way you can". At least I and virtually everyone I met didn't born with these, nor they figure all that out by themselves.
The contraception didn't work and it was too late for abortion (should we choose it as an option)
It's great to see some support like this. Not just to help with motivation but also to see what the interest is.
I think there is a massive opportunity in creating a k-12 home schooling version of Lambda School but targeted at general knowledge. Why not start by work together on it?
I am interested as it's probably clear from my question, but I don't think I would be a good fit to actually put it together (or that it would be cost-effective). I would be happy to put and run some structure which could do this. I should emphasize though that I am not trying to create a generic program for everyone as I think that's vastly harder and would IMHO need much more.
Hi. Thanks a lot for a really nice write-up.
It seems that the regulations in the Czech Republic are actually legally "workable", i.e. it's possible to teach kids close to self-directed without having to do a lot of "compulsory curriculum" (i.e. my estimate is <5%). It also seems there is a "subculture" of families doing this and I managed to get to some people who know how to deal with this.
My conclusion is that there is no simple answer.
I don't aim for a simple answer and I do not expect there is some. But as I said, the current system seems so broken that just answering "try not to harm" is a good substitution question. Also, it doesn't really seem that kids would learn [and retain] that much [useful knowledge] compared to "watching youtube videos".
I have concluded that school reproduces more than knowledge. It reproduces culture.
It does so indeed, but that culture doesn't seem to be very worthwhile to me. I am rather tempted to risk this one, as the main thing it signals doesn't seem to be that problematic to pick up later. I.e. homeschooled kids doesn't seem to have issues when they want to switch to formal education, and at "worst" they pay a year or two of whatever-made-them-happy-or-stronger-in-other-ways for that (again, e.g. watching videos :-) ). I have read/heard this claim quite a lot, and I am less and less convinced that this aspect of schooling is positive.
I also think that some other benefits can be supplied in other ways and even more effectively. For example a good social network - it's vastly superior/efficient to learn a few networking skills and just infiltrate e.g. some organizations, events or groups with a "good network". It's unlikely that it's your random class that is "good social network", and e.g. I have literally 0 friends from my primary and secondary school (I have a lot from university though - but anyone can do the uni I did if they want).
The other aspect is that you and I might have good ideas about which curriculum would be best
That kind of links to my previous answer. I am coming to a conclusion that the skills I would like to convey are rather very generic, blurry and meta. I want to teach her how to understand herself, her motivation, discipline, emotions, reasoning, goals and drive, how to understand other minds and how to model the world. I want her to understand how she can learn by herself whatever she wants to learn. It really seems to me that if we don't want to learn something, then we should rather not learn it as we are not going to remember that anyway. And given the right tools and meta knowledge, we can learn almost everything in much shorter times than the "most common" path. Therefore, I don't really care that much about the "object-level" curriculum. I think (but that's still subject to a research) that the concepts important to me can be taught via mostly arbitrary picks in the real world (like nature or engineering) or just going outside and explicitly talk about that person kicking that soda machine. And I can't imagine how these skills would not be useful or universal to time, at least on ~15 years from now scale, as they seem to be useful since eternity. I can imagine my job going obsolete - but I would use exactly these skills to find and learn something else which would be the best in that time for me.
Thanks for the tip and links. Unfortunately, it doesn't show much in Prague (but still gives me a hint about what to look for even if some school isn't registered in the linked project).
The OP seems intent on designing/engineering the perfect education, when the answer from this perspective will require a lot of letting go.
Hm... I don't think I would have issues of having to do so. I am trying to understand how to think about this, and this simply didn't occur to me before. In fact, it seems that me and my girlfriend are currently rather at the side of trying to figure out how to do this in self-directed-way, but it's still in early stages.
Thanks Vil. I agree with Ericf comment that you seem to try to take it more generically than I intended (i.e. I realize that I have resources 99% of the local population doesn't). That said, I fully agree with you on these points.
it takes about 1 hour to teach at home what they teach at school in 1 day
These are good datapoints, thanks.
And yeah, I would hope that with internet and some good courses which would give the kids some "library" of what I could learn + mixed with the self-driven learning wouldn't need a full attention of a tutor.
Thanks for clarifying my questions.
The key point (I'm synthesizing this from How Children Learn and How Children Fail, by John Holt) [...]
It's very much similar to what @Raj mentioned above, am I right? Seems that Holt advocates for self-driven learning, e.g. from a goodreads review:
Holt believes that children learn best when they learn at their own pace and pursue their own interests--learning should never be forced or uniform, but spontaneous and dynamic. Children don't need to be "taught" -- they simply need to be given opportunites to LEARN
Thanks for the tips though, those two books (How Children Learn | Fail) are definitely going on my reading list.
Hey Raj. Thanks a lot for an insightful post, it's definitely that sort of things I was looking after, regardless if I immediately agree with them or not.
1-self learning: How I read it so far is that instead of selecting "the way" first and optimizing it later, instead it might be a good idea to focus on learning how to learn by yourself first, recognizing what's the most effective in any given case, be it via internet or an actual human resource such as a tutor.
By the way, my solely main motivation for her to know English was the access to much better materials so she can learn by herself.
I would also check out Sudbury or democratic schools, which are schools with high behavioral space and very low in coercion.
A quick search shows that there are some very reasonably priced democratic schools around. So great to see there are some options like this (at least, I will be able to get some references or even visit them myself)
there are a lot of priors I'm guessing we don't share that might make some of this not make sense. Please bring them up and I will try to answer them though most answers could probably be found in some way on supermemo.guru.
I think your post does make sense to me on this level. I think I have to first go through the linked material, which will surely take some time to process (thanks!).
Thanks @ericf!
How neurotypical is your child?
She's regular kid, so neurotypical. Goes to an English speaking kindergarten (so she speaks fluently two languages + we are starting with Spanish) where she does above average according to the teachers with behavior and socializing, although she prefer playing with teachers and older kids. Thanks to the fact that one of us didn't have to work, we could spend a lot of time with her in the first 3 years and now she can read simple words and sentences (in English - how I hate its irregularities damn!) and do simple arithmetic already. That likely puts her in higher percentiles, but nothing super exceptional nor she shows signs of extremely high intelligence or capabilities. Frankly, she also doesn't have anyone to inherit these from...
Do you have someone who can spend 10-40 hours a week educating them (depending on the first answer), or the cash to hire someone for that amount of time?
Fortunately for me, salaries for teachers are one of the lowest from the OECD. Depending on the efficiency, me paying a private tutor ~30h/week would be technically feasible and I would still have some money left, but it would be at the edge of what I could spend. I believe this cost could be shared through multiple kids if that was the way to go. I myself have masters in CompSci and my girlfriend masters in pedagogy (not the child development though, although I mean it means little here anyway), so homeschooling done by us is also an option but for me there is definitely much higher opportunity cost (and I wouldn't enjoy it full time, same as my girlfriend).
1-on-1 instruction for the (child-dependent) sufficient amount of time each week can be applied by anyone with 1SD above average intelligence and a little bit of "how to teach" education. This is the simplest way to ensure someone learns all the key skills (Arithmatic, Algebra, Reading to learn, Writing to communicate, basic facts about history & science)
Do you have any resources or even anecdotal evidence for this? It sounds true to me, but I would say it's super obvious. Are you referring to e.g. Blooms 2SD?
Once they are old enough to attend High School classes, if you have access to a large (400+ per grade level) school they can get some good specialized classes that would be harder to do with general tutors. Junior college classes are also an option, depending on maturity level.
Sorry, I am not entirely familiar with these as we are not US-based. Based on googling, is what you mean:
- large schools having specialized classes: is it that once they are big enough, they can have some special programs like in-depth dive into, say, physics or something? Or do you have some specific classes in mind?
- Junior colleges - we don't really have that where I live, we only have universities, but I think it still applies - she could attend them as they are free to attend for anyone actually (at least lectures).
To be fair though, I am very skeptical I could find many good classes in these, but it's surely an area I want to explore (although it's not a priority now, given her age).
In a way that you never "go back" and edit the "immutable" previous writeups, right?
Sorry, a quick question: linear means something like:
- Tuesday: I bough bananas and strawberries
- Wednesday: Bananas are good
while non-linear means
- Bananas:
- I bought them (Tuesday)
- They are good (Wednesday)
- Strawberries
- I bought them (Tuesday)
?
Yeah, seems that we use success system by default then. Thanks again!
Thanks, that's pretty interesting, it's good to get some inspiration from it and then replace "inappropriate" words by something from her vocabulary (like princesses, dogs, cats instead of murderer, undead, zombie, ...) :-D . We'll get there, eventually.
Thanks, that's stupid simple, love it. It seems that the little one likes cooperative storytelling a lot, but she doesn't understand the dices and the concept of opposing checks very well. I still do some hoping she picks up, eventually...
Thanks! That's useful, didn't know about it.
Hello. Is it possible for the author to review this and possibly update it? It has been already 4 years. I wonder, if something changed.
I answered to that thread.
And I think I wrote you on Facebook. You should have my message in "others".
I was thinking about today - Tuesday. But it seems a bit in hurry for other to notice. Do you have some date which would suit you? For example next week Wednesday 11?
I am so sorry about not appearing on the meeting - I've got stuck in a train from east for several hours. I should have at least post it here when I knew that I can't make it. I am still really looking forward to meet you guys.
What about meeting on November 3 (Tuesday)?
On this meetup there was a guy from Ostrava. We exchanged numbers and emails and I promised that I will keep in touch. Unfortunately, my mobile phone crashed and I had to reinstall it, loosing my message history. I couldn't find him, since I do not remember the name. The only I remember is that he is doing his doctorate on VSB - FBI. Unfortunately, I couldn't find any name what would remind me him.
So if you are reading this, contact me!
Can I know to who and where the money for the book goes?
It worth to note one more thing - I'm not really skilled Bayesian and rationalist, but I do my best and I'm currently studying. So far I've finished HPMOR, An Abridged Introduction to Less Wrong and now I'm working on core sequences. There I've just finished Map and Territory.
For anyone interested, I've made an ebook variants for myself (epub, mobi, PDF, odt). It is far from awesome, but at least readable on e-book reader. https://www.dropbox.com/sh/6agp4otiukejb0g/AACO-5V1J8i0USBWUFL9nw74a
Hello.
I was searching more about my interests and I've found a opportunity has a nice Bachelor's topic in maths/informatics/neuroscience. I was offered two topics:
- Analyse properties of correlation matrix graphs (e.g small-world property)
- Conditional mutual information - how to detect synergy and which values it can takes when there is restriction on cardinality of given variables.
Both are connected with neuroscience (e.g the correlation matrix is created by brain activity, variables are activities of different parts of brain etc.)
Does anyone have any informations or advices to this?
Hi.
After some more research and digesting these answers (and some other sources), maybe this is just to heavy for me. But it is really interesting reading and thank you for that.
Thank you for answer.
Could you redirect me to somewhere, where I could find what problems/directions are you talking about? Since I'm not so shining mathematician, maybe I could contribute in these areas, which I found similar interesting.
Thank you. I'm just going to go through the papers publishers. Great idea!
The "mainstream-friendly" stuffs are maybe the middle-path for which I'm looking for, since response from Risto_Saarelma is pretty explanatory about my possibilities.
And it is possible to do similar kind of Bachelor's thesis and I believe it would be possible. That is not a problem. But, to be honest, I'd like to do some work which I find fulfilling even at tiniest amount. I'm doing literature review in my free-time.
Thanks for the answer.
I don't know how I could miss MIRI's course recommendation list. It looks great. Will definitely take a closer look at it.
Second part is a bit disappointment for me, since I'm not that kind of student. I'm in the stronger group of mathematicians in my university, but in that group I'm in or below average (they are one of the best in my country).
Maybe I put too much weight too maths part of AGI, which are obviously aren't for me. And I'm not sure about taking PhD in it right now also. Do I understand correctly that right now there are no less complicated problems or problems for regular people in AGI? Nothing, where I could develop my skills which could be useful even for other ways than taking a PhD and do heavy research with AGI world leaders?
Thanks
Hello,
I'd like to get some opinions about my future goals.
I'm 21 and I'm a second-year student of engineering in Prague, Czech Republic, focusing mainly on math and then physics.
My background is not stunning - I was born in 93, visiting sporting primary school and then general high school. Until I was in second year of high school, I behaved as an idiot with below-average results in almost everything, paradoxically except extraordinary "general study presupposes" (whatever it means). My not so bad IQ - according to IQ test I took when I was 15 - is about 130 points. When I was 17, I realized that there is something about the world that needs to be done with. I started to study, mainly math and physics. I was horrible at it - I had very big disadvantage because I missed basics and wasn't able to recognize it. Anyway, I tried (but, unfortunately, not as much as I had to) and reached so-so level and I got on the technical university. Here I tried really hard and I achieved relatively good results and got into the best maths-focused student group. I'm below-average in this group (about 30 students) and my results are satisfactory. I'm quite popular thanks to collaborating on some non-study events for my schoolmates. I also created a presentation for high school students about engineering and I distributed it among faculty workers and students, who are connected to propagation.
About 10 years I obtained ECDL and it started my curiosity about informatics. But nothing special - I was autodidact in HTML and "computer administration" for regular usage. I was also very interested in economy, as my father is working in this area. I actively did cross-country skiing and play on piano and trombone.
I have high charisma, authority and ability to organize people and some bigger events, which I was usually asked to prepare (the graduate prom, matriculation etc.). I have good reasoning skills and ability to negotiate even under heavy pressure and stress. People usually enjoy time with me and appreciates me for my honesty, empathy and "cold-think" reasoning solutions, which in most time shows there were the best possible. I'm in healthy relationship for two years. My family is good background for my activities and support me. They also support me financially. My expense per month is not more than 300 USD including accommodation with in an apartment (university students, two of them from my university and domain), food and social activities.
Currently, apart from my school activities, I'm also attending some kind of philosophy group every week, where we usually discuss some topic about epistemology, relationships, culture, religions etc., we read some philosophic works (Platon), deal with art (classical music or paintings) or we write some kind of voluntary essays. I'm really interested in discussions about these topics and I try to develop my reasoning skills as often I can. For example, now I contacted a priest from local temple with whom I want to discuss some religion based questions. I autodidact psychology (last book I've read was Kahneman: Think fast and slow), rationality (started to read LW sequences), and programming. I enjoy using open source software on my Archlinux laptop and now I dived into Python as a scripting language. I also develop some web for my mother using Django and I also signed for a statistical research task about datamining in Python (pandas, numpy, scikit-learn...) or R. In school I have courses of C++ also. I'm not the most talented or generally best mathematician or programmer, but I have quite good learning (and also teaching) skills.
I've chosen my "path" - I'd like to do what's right and true and seek for the truth whenever it is possible. I feel that I'm not getting everything (e.g. from my school) I need for changing the world to a better place. I could do more. I can't decide where to focus and how to divide my attention and possibilities. Should I do aggressive autodidact of sequences? Should I focus on maths and algorithms or biases? Should I try to develop my social skills?
And the second question is simple:" Are there any Czechs who are interested in meetups in PRAGUE?"
Thank you