DecaDeciHuman
No bio...
User ID: 3518
DecaDeciHuman
6d ago
If I have built 90% of a bridge, and realize that the policy that led to me building it was a mistake, the correct decision is often to finish the bridge regardless of the fact that overall building the bridge was a net negative.
DecaDeciHuman
6d ago
would you speak Spanish to a fellow American expat in Mexico City? Or Thai to one in Bangkok?
If I was emigrating to there, and it was a public space? Absolutely. Admittedly, that is one of my pushes away from emigrating to a country with a different language, as I am terrible enough at my native tongue.
Be aware that immigration != tourism.
DecaDeciHuman
6d ago
Attempting to ban a view preferentially results in banning actual discussion of that view.
If a view is common you'll still get one-offs of people who either do not realize the view is banned, or don't care, and who drop an unbacked opinion.
This has the obvious feedback effect.
Fair enough.
Read 'sadness' as shorthand for 'a small but decidedly nonzero tick of update towards having to weight views on this site based on presumed self-censorship, inexpressibility of some views, and resulting skewed representation, when this is one of shockingly few sites left on the public internet that appeared to have not fully gone down that path.'
DecaDeciHuman
7d ago
Also, speak English [or rather, the country's language whatever it may be] in public spaces. Yes, even when speaking to other people who share a different common tongue.
This may be so self-evident to some as to be not worth mentioning.
Cliques of people speaking a language you don't is demoralizing at best.
DecaDeciHuman
7d ago
Contribute something other than how much you don't like Indians
That is not how this post came across to me, and it saddens me that this appears to be the Official mod interpretation of that post.
DecaDeciHuman
7d ago
The last post came across as "I want to move to the states myself, as I see it as good, and wished more people saw it as good".
This post came across as "I want more people (or at the very least don't want less people) to move to the states, and I am going to criticize a place I wish to move to for its policies on who can move to it".
These are not the same thing, and it does not surprise me that some people overall like the former but not the latter.
DecaDeciHuman
7d ago
They can also cause problems upstream:
(Assumes driving on the right / CCW flow through the roundabout.) If you have a roundabout with heavy traffic flow from, say, the South entrance to the West exit, you can get perpetual backups for traffic coming from the East.
DecaDeciHuman
8d ago
Conservatives [...] insist the women who get abortion are blameless
This does not match the attitudes of conservatives I know at all.
If this is how the Conservative party has trended, that would be an interesting growing disconnect.
DecaDeciHuman
8d ago
To reduce debt, you want more income
Insofar as immigration is seen as having a significant baseline cost per immigrant, raising the amount of money received per immigrant can result in higher net income even if revenue falls.
DecaDeciHuman
8d ago
I fail to see how "please put the conclusion at the start" is any easier than "please put a conclusion at the end" in this regard.
If people are going to adhere to a format they can adhere to either.
If people are not going to adhere to a format they are unlikely to adhere to either.
DecaDeciHuman
8d ago
Let me rephrase:
I see three valid approaches:
- Structure things with the conclusion at the start, and have the convention be to skip over the conclusion if reading the full thing, then read the rest, then read the conclusion.
- Structure things with the conclusion at the start, and have the convention be to read the conclusion then read the rest.
- Structure things with the conclusion at the end, and have people who may wish to skip read the last paragraph first to decide if they should read the rest.
Also, putting the tldr at the bottom of a post is bad netiquette
Putting the tldr at the bottom matches perfectly with a convention that people who are unsure should read the end first to check.
DecaDeciHuman
8d ago
I don't necessarily disbelieve you.
Sidenote: this comes across as patronizing.
If this is not your intent, I suggest attempting to rephrase to better communicate in future.
If this is indeed your intent, I may as well stop the conversation now, as there can be no additional learnings.
DecaDeciHuman
8d ago
pandemics can be highly dangerous.
Sure. So can lockdowns.
(How many routine medical procedures & tests were cancelled due to COVID lockdowns & overreactions? How many of those were the same ones that the Medical Consensus(TM) said were hugely important for health? Oh, the Medical Consensus(TM) has since suddenly shifted to saying no, those weren't helpful actually? Oh, the Medical Consensus(TM) has decided to classify all deaths with detectable levels of COVID as solely due to COVID, and now there is no evidence for increased death rates due to diseases that could have been caught via said procedures? How... convenient.)
(Hopefully not overly snippy. I had to go back and tone it down; hopefully I toned it down enough. Still sarcastic; hopefully not snippy.)
DecaDeciHuman
8d ago
Do you have better alternatives? At the end of the day, if you're unhappy with the government, then you need to elect a better government.
Sidenote: if you've given me a 3sat problem and purported solution, expecting me to solve said 3sat problem just in order to point out that your purported solution does not actually satisfy all of the equations is inane.
In a way, the new Republican government reflects the deep unrest with previous medical policy.
Agreed.
DecaDeciHuman
8d ago
I think an initial plea of "this is looking like it could be a pandemic: please restrict social contact", along with the governmental support to allow people to do so*, would have been a very good idea and would have been sensible in such a case. This is not the same as a government-enforced lockdown, especially one as hamfistedly done as COVID was.
In some ways this is less effective than an outright lockdown; this is far less likely to cause widespread backlash.
Of course, it's now in many ways a moot point. That credibility was burnt; this is close enough that it was caught in the backlash and also wouldn't be feasible now.
[* e.g. government-financed no-questions-asked refunds for travel, decrees that workers must be able to take sick days, vacation, or leave without retaliation for the next X days, that sort of thing. I am overall very much against Big Government, but as long as we're already taking the downsides...]
Once it becomes clear that it's nowhere as bad as it was thought, and the primary risk was for old and sick people, then lockdowns should have been lifted. I don't think that this was obvious until several months in, and I was doing my best to stay abreast.
I think the primary place our calculations here differ is mine includes the time lost in lockdown in the downside. If you lock people up for a hundred man-years to save one person, you haven't actually gained much of anything.
This pushes the tipping point earlier (assuming lockdowns in the first place).
My bigger issue here is that I've heard a growing amount of attempted retroactive changes of the narrative, of the same people who were proclaiming on day 90 that LOCKDOWNS MUST CONTINUE who are now backing off and attempting to say they were saying otherwise on day 90.
DecaDeciHuman
8d ago
I've burnt more weeks of my life than I care to admit on building throwaway prototype proof-of-concepts that showed that said concept did not, in fact, work.
DecaDeciHuman
8d ago
What is the advantage of this over reading effortposts starting with the concluding paragraph?
DecaDeciHuman
8d ago
COVID lockdowns were a bust, but even if they'd curbed the disease, I'm unsure if it was worth it if it pissed off tens of millions of people in the States to the point that even basic medical knowledge became untrustworthy.
...which was repeatedly stated at the start of the pandemic by people who were later denounced as COVID deniers and largely memory-holed since then.
(I'd love to give references, but for the obvious problem with this.)
DecaDeciHuman
8d ago
I think a sensible reaction to covid pandemic measures would be to demand governments provide far more evidence to justify a lockdown longer than a month.
My concern at this point is not justifying a lockdown more than a month. I believe there is now sufficient backlash on that particular subject.
My concern at this point is what other thing that might be demanded next time.
If I have a pilot who forgot to put the landing gear down, went through remedial training on the subject, and was reinstated, my concern is not so much the pilot forgetting to put the landing gear down a second time. That particular item has been well-established. My concern is moreso "what other things might the pilot have forgotten that weren't covered in the remedial training?" Same idea.
(This is why remedial training often covers far more than just the specific incident.)
If you really want to, there's always the option of litigation against government officials if you can prove they willfully lied.
Government officials often have immunity.
If they encouraged lockdowns when it became clear it wouldn't help, or overstated the benefits of vaccination (without the benefit of hindsight), then they should be accountable for the harm.
The burden of proof lies the wrong way for this to help in practice, as all of these are nigh-unfalsifiable.
Of course, if your intent is for this to be a bureaucratic tarpit then your job is done here. I sincerely hope that is not your goal.
DecaDeciHuman
8d ago
Have replaced plumbing in slab before. On the one hand, not fun. On the other hand, doable.
If you're running wiring through slab you really want electrical conduit - and also pullcords preinstalled, if possible.That being said, pay very close attention to electrical code.
Also, take far more photos than you think you need prior to pouring the slab, with measuring tapes in the shot. Yes, things do shift somewhat during the slab pour, but it's still a much better starting point than you'd otherwise have.
DecaDeciHuman
9d ago
"trust less" does not imply "always ignore". The input is still taken into consideration; it is weighted less than it would otherwise be.
The solution to "our experts can't be trusted when the topic is political" is not to always ignore experts
Someone stating something results in upweighting to some extent or another all hypotheses that are compatible with that observation. (Assuming you have made sure your hypotheses are not overlapping.)
In this case, someone stating a falsehood on something political results in upweighting the hypothesis that they are incorrect on political matters and correct on apolitical matters, as this is compatible with said observation. It also results in upweighting the hypothesis that they are incorrect on political matters and incorrect on apolitical matters, as this is also compatible with said observation.
[N.B. I have stated nothing here about how much said updates change the weighting.]
DecaDeciHuman
9d ago
It's worse than you think.
Look at figure 8 of the GPT4 'technical report'. Or figure 9 of this paper on mode collapse. It's all across the calibration scale that gets messed up, not just the low extreme.
If you, say, ask a LLM for the result of a fair d4 roll (1-4), with sufficient formatting/etc such that with overwhelming probability it will output just the tokens '1', '2', '3', or '4', a properly calibrated model "should" result in the following output probabilities:
'1': 0.25
'2': 0.25
'3': 0.25
'4': 0.25
And many base models are pretty close. Not perfect, but reasonable.
With RLHF, however, you'll often see something like, say:
'1': 0.02
'2': 0.07
'3': 0.90
'4': 0.01
Why? Short answer:
Consider the case of a slightly weighted coin tossed once per training session, that flips heads 60% of the time. A base model will result in the highest training score if it flips heads 60% of the time.
But a RLHF'd model will result in the highest training score if it flips heads 100% of the time. Because when the user sees "which is a more likely answer: heads or tails" - they will answer "heads". And so the model will be trained to answer "heads".
DecaDeciHuman
9d ago
Indeed, that's a large chunk of the way gain-of-function research tends to work in the context of bioweapon development. Make multiple different things, combine them together, and see what sticks.
See e.g. https://www.science.org/doi/10.1126/science.1229455
If I don't defect in a prisoner's dilemma while the other person defects, is it my self interest that I get a long term in prison?
More options
Context Copy link