lavelganzu
lavelganzu on Hacker News
Meanwhile here in central Austin, it's a 0.9 mile walk from my door to the nearest bus stop that I can use to commute, walking along major stroads some of which don't even have sidewalks, much of the year in Texan heat with no scrap of shade. Then it's up to a 30 minute wait for the delayed or canceled bus, then almost exactly a 1-hour ride on the express 801 to go 7 miles to work downtown.
Somehow we combine inaccessibly rare bus stops with speed barely over walking.
The solution, I imagine, requires many changes that are politically infeasible. First, double the number of buses to reduce the wait between them. Second, add neighborhood circulator buses to get people from the neighborhoods to the express buses. Third, either add dedicated bus lanes in congested areas or, in an ideal world, make all congested inner-city roads toll roads, and use the tolls to subsidize buses.
There is a downside of making buses free, similar to the experience of cities which stopped enforcing "turnstile hopping" for trains, which is that it attracts a small number of hostile and malicious riders. An advantage of treating transit as a public good means this downside becomes an empirical question, not a moral one: Which approach leads to more ridership? In some cases, enforcing fares leads to more ridership by increasing safety and decreasing the amount of time spent cleaning up befouled surfaces.
Gwern was skeptical, & noted that an IQ of 75 [1] in this case study is very low. He additionally raises a few points, including that volume loss is not the same as neuron loss. He also predicts several deficits the case studies didn't report that he'd expect to see, including many small deficits in simple tasks adding up to large deficits in complex tasks.
[1] https://gwern.net/hydrocephalus#sn3
It's been around long enough to have a Wikipedia page [1], which can give you the main facts & demographics. In short, it started in 2006 as a group blog for people interested in AI. This was long before LLMs, and it was expected among the readership that understanding the math of decision theory would be important to AI. This spiraled out into general interest in how to be more rational as humans, and LessWrong is largely responsible for rescuing Bayesian statistics out of the academic wilderness. Many jargon phrases that are now common in nerdy circles originated there as well. They invented the field of AI safety, and are unhappy about the poor state of AI safety at this time.
There are in-person meetups (primarily as a social group) in most large cities. At the meetups, there is no expectation that people have read the website, and these days you're more likely to encounter discussion of the Astral Codex Ten blog than of LessWrong itself. The website is run by a non-profit called LightCone Infrastructure that also operates a campus in Berkeley [2] that is the closest thing to a physical hub of the community.
The community is called "rationalists", and they all hate that name but it's too late to change it. The joke definition of a rationalist is by induction: Eliezer Yudkowsky is the base case for a rationalist, and then anyone who disagrees online with a rationalist is a rationalist.
There are two parallel communities. The first is called "sneer club", and they've bonded into a community over hating and mocking rationalists online. It's not a use of time or emotional energy that makes sense to me, but I guess it's harmless. The second is called "post-rationalism", and they've bonded about being interested in the same topics that rationalists are interested in, but without a desire to be rational about those topics. They're the most normie of the bunch, but at the same time they've also been a fertile source of weird small cults.
[1] https://en.wikipedia.org/wiki/LessWrong
[2] https://www.lighthaven.space/
Excellent chart on that page. Hurrah for asking their degree of confidence! The plurality of respondents had low confidence, of course, as scientists should pending some experimental reason to prefer one interpretation over another.
For those who don't click through:
- It's a Nature news feature from July 2025, including responses from 1100 people with papers in quantum physics
- 36% preferred the Copenhagen interpretation, and nearly half of those indicated "not confident"
- 17% epistemic theories, 15% many-worlds, 7% Bohm-de Broglie pilot wave theory
- small percentages for various others including "none"
- additional charts for related questions
They do it by correctly noting that there's no such doubling. Conservation of energy is within-world, not cross-world.
Definitions are for math. For science it's enough to operationalize: e.g. to study the differences between wakefulness and sleep; or sensory systems and their integration into a model of the environment; or the formation and recall of memories; or self-recognition via the mirror task; or planning behaviors and adaptation when the environment forces plans to change; or cognitive strategies, biases, heuristics, and errors; or meta-cognition; and so on at length. There's a vast amount of scientific knowledge developed in these areas. Saying "scientists can't define consciousness" sounds awkwardly like a failure to look into what the scientists have found. Many scientists have proposed definitions of consciousness, but for now, consensus science hasn't found it useful to give a single definition to consciousness, because there's no single thing unifying all those behaviors.
Another option is to fake a planetary magnetic field by placing a large electromagnetic satellite in mars-stationary orbit. Merely reducing the effect of the solar wind stripping away the atmosphere could lead gradually to a much thicker atmosphere, and even surface liquid.
It's a fun idea to explore in fiction, but it certainly didn't happen. The evolution of our species is too recent.
> Can you honestly say you've never had zipper teeth part company with the fabric or never had to apply candle wax or similar to the teeth so they run smoothly?
Do regular people ever wax their zippers? (ChatGPT says it might be done by sailors or leatherworkers on occasion, for whatever that's worth.)
At the level of industries and large groups, the chief answer to your "Why?" is the same sort of reasoning as the old "Nobody ever got fired for buying IBM": Nobody ever got fired for using established performance metrics.
On the individual level, there's another tricky problem, which is that very few individuals could figure out an alternative performance metric that beats the established one, no matter how gamified the established one is.
On the opposite side of moral hazard, early in my career I worked for a large web security company in tech support. We were not permitted to escalate to engineering at all. Often this meant the only solution was to apply our own, unofficial code changes!
Incidentally, Blackrock doesn't buy homes. Other private equity, combined, own less than half a percent of single-family homes in the US.
One small trick that would greatly improve headlines of this sort: a number. Compare how these two possible headlines affect you:
- Three Scientists Call for Ban on Social Media and Smartphones Before Age 13
- Thousands of Scientists Call for Ban on Social Media and Smartphones Before Age 13
In this case, it's 3.
e.g. design a terrible pathogen
Agreed.
There are technical challenges that are so basic that a programmer should be able to regurgitate them even under normal-for-an-interview stress, and these are still useful filters. (e.g. For a front-end web dev role, we asked candidates to change the web page's text color. Half couldn't.) People who suffer from truly extreme stress at interviews are probably best off treating it as a medical condition to be managed.
I'm impressed by the author's gall in naming his idea an "iron law", without bothering to test how true it is. I suppose it'd get talked about a lot less if he had named it simply "Michel's Conjecture of Oligarchy".
I for one want new giant stone pyramids. :)
It "allows" people with vast resources to spam only until the moderator removes the account, and it ensures the moderator is paid to do so. But more critically, it removes the profit incentive to spam, so even if people with vast resources were "allowed", they won't.
Money is an imperfect but real solution. The simple thing is to charge a small sign-up fee. Obviously this dramatically increases the barrier to entry for real humans. But it should cut the spam even more sharply.
Isn't this merely a technological change? "A few decades or a century ago", being location-specific was the only possible option for a social club. Now there are more options; location-specific options still exist, and location-agnostic options exist also.
You can totally pick a convenient cafe or pub and start hanging out there & inviting your existing friends. In time you'll start to recognize the other regulars, and you can make a point of chatting with them regularly (but not overstaying your welcome especially early on!), find out what they're interested in, offer & request small favors, crack jokes, eventually a bit of friendly competition, casual debate about mutual interests while intentionally trusting them enough to let them change your mind a little, etc -- all the things you'd normally do to build a social connection with another man. The upside and downside of a location-based approach is that it's a very weak filter. The other regulars may be people whom it's a real stretch to learn to connect with.
Location-agnostic social activities are typically focused on an activity or interest, e.g. people who want to hike, watch a movie, play a game or sport, do political activism, do community service, etc. So the social group comes with a filter attached that ensures you will have an easier time connecting with them. This is great! There are some downsides, too, but nothing serious.
There's a plausible argument for it, so it's not a crazy thing. You as a human being can also predict likely completions of partial sentences, or likely lines of code given surrounding lines of code, or similar tasks. You do this by having some understanding of what the words mean and what the purpose of the sentence/code is likely to be. Your understanding is encoded in connections between neurons.
So the argument goes: LLMs were trained to predict the next token, and the most general solution to do this successfully is by encoding real understanding of the semantics.
Correction: "Now something that is also bad, but not nearly as harmful as acid rain, is falling from the sky."
Technically the "due process" clause is in the 14th Amendment, ratified 1868. ;)
For in-person play with ordinary playing cards and an arbitrary number of players, keep the rules the same except:
* Start with 1 deck per each 2 players. Shuffle the decks together.
* Deal 10 cards to each player, & set ante at a constant 50 chips per player.
* At the end-of-round reveal, the common suit is whichever suit has the most cards. If there's a tie, flip cards from the remaining deck until the tie is broken and the common suit is determined.
Build enough housing & prices go down. It's worked everywhere it has been tried, and it'll work in LA, too.
Good catch. RMS (root mean square) error is typical in signal processing to avoid this undesirable cancellation.