Retric

That slowly changing landscape killed SO. Every year a higher percentage of answers were wrong because they got outdated, and removing duplicates made it worse.

I don’t think anyone in the org wanted to face the fact that the underlying assumptions in their model was flawed. The older an upvote the less reliable it becomes.

Consistently catching errors is vastly harder when you’re not the one doing the legwork. Send the wrong file, do a bad copy and paste, or skip one verification step, etc is all it takes for most people to majorly mess up.

For most professionals it’s just safer to avoid having output in your workflow. Keep using AI, but stop copying and pasting AI output.

The issue is the opposite of hallucinations where the LLM ignores relevant information.

Sure you can combine this with a regular search but there’s opportunity cost whenever you say LLM + something else rather than just the other thing.

There’s models created by people with heavy right wing affiliation, but it’s harder than you think to bias models consistently because political beliefs are so broad reaching.

Calling out bias however is a standard practice for groups whose beliefs don’t match reality. You see that on political extremes, but also revisionist history, many religious groups, conspiracy theories, etc.

That’s only true over a relatively short period.

Stuff you can do for a week doesn’t automatically extend to maintaining that over the next 50 years.

> the numbers I gave you

You didn’t give any numbers, just stated things very confidently without actually reviewing any of the relevant literature.

> the false assumption that moving from each tier is a low probability event

Doesn’t need to be a low probability events, multiply probabilities and the result becomes less likely every single time. And that’s of course ignoring the underlying reasons why the probability at each stage isn’t low in the first place. A smart kid at 6 is unlikely to turn into a dumb kid at 16 etc.

> massively increases

The data doesn’t support that assertion. Elite collages do favor elite high schools, and elite high schools do favor elite middle schools etc but each link in the chain dilutes the impact of the steps further up the chain. Kids have many opportunities to jump up the chain before college and many to fail to keep up with the elite track. Which is why statistically elite preschools have such a minimal impact.

Of course that’s adjusting for the role of the parents which impacts attendance at elite schools but is obviously not impacted by their kids attendance.

Elite early education minimal little long term benefit, people tend to revert to the mean over time. The education system has plenty of slack for intelligent individuals to excel, so play the long game.

Health and wellbeing pay the best dividends as you can maximize the nurture aspect of intelligence early and build resilience against unexpected hardships.

I’d expect there’s many situations where sending a helicopter would be considered too dangerous for the crew, but the lack of medical care on a drone seems like a big limitation.

Drones are not at equal risk in all missions and neither are wounded soldiers in equal need for evacuation. A long list of Americans have died in helicopter accidents while receiving medical transport but a much longer list had their lives saved by those same basic methods.

Look at the “legal %” field on that chart. That’s utter trash even counting per move vs games without errors. Further they outperform vs bots who stick to main lines more than normal players, actual performance vs humans means both lower ELO and more errors.

Note: * = Anchor player (fixed rating, used for calibration)

LLM’s are terrible at following the rules of Chess when playing a game. Determinism isn’t helpful because valid moves are so dependent upon the total state of the board.

Programming state gets way more complicated, but code is generally written to help people understand it greatly simplifying the process. Feed in “cursed” code leveraging obscure language features in creative ways and LLM’s and most people fail.

HOA’s kill any character an area has. The fees more than offset the economic benefits of joining one. The damming economics largely comes down to local governments skipping road maintenance without giving you a discount on your taxes.

Net effect you lose money living in a worse location.

At home charging is actually the #1 reason I don’t want to buy an ICE. IMO it’s by far EV’s biggest benefit.

The nice thing about maps is they show every country, so you can quickly see how few lack such links. If like tables add up what percentage of countries aren’t tied in with at least one of their neighbors. It’s a short list, even NK is tied to China.

You may discount England but it’s got enough DC links to move a large fraction of demand.

Japan has two networks for historic reasons, but they can move well over 2% of each grid’s average demand to the other. It’s just significantly less efficient but that’s relatively minor vs the other disadvantages of a split 50hz vs 60hz national grid.

Russia, China, and most of the rest of the world has interconnected their grid with other countries.

This is a map of where the grid is frequency matched across internal borders allowing easy exchange. DC connections don’t need this synchronization, but it reduces losses with AC. https://en.wikipedia.org/wiki/Continental_Europe_Synchronous...

In war this might be a minor issue, but in war electrical infrastructure ends up a target so it doesn’t matter that much.

National borders don’t limit power transmission.

Chile still wants transmission north south because in the winter solar closer to the equator has longer days and in the winter solar near the poles does.

Transmission lines can act like storage when there is PV to the east or west of you. This doesn’t remove the need for storage, but can dramatically reduce it.

The old model of cheap peaking power plants and cheaper base-load power is equivalent to cheap peaking power + cheap but inconsistent green energy. It’s a very complicated optimization problem to work out the optimal mix of locations and quantities to of each, but price signals can help assuming you haven’t completely distorted the market.

You’re trying to apply an arbitrary metric to two radically different process. Suppose you could pay an extra 1,000$ so it fils in 40 seconds vs 4 hours at home (and only at home).

Personally it wouldn’t be worth the money to me because the only thing that matters is the time to plug it in and the time to unplug it.

> it's not really a benefit because charging takes a lot longer than a gas fill up.

Plug in at home takes way less of your personal time than a gas station and costs significantly less than public charging.

If you feel that way, don’t start pissing contests.

Even individual words can be an issue. Humans understand when you want a new story about a magic school there’s issues using Hogwarts as the name.

LLM’s will happily regurgitate multiple paragraphs practically verbatim if you don’t specifically design them not to behave in this way.

Because ultimately when you design a system to best predict the next output based on a given input perfect replication of prior training is the correct response mathematically.

Have some conviction. You don’t need to respond, but going limp isn’t any way to live.

You misunderstood, “Larger patterns still emerge” means yes you notice LLM’s but it’s for different kinds of patterns than you get from people.

> not because of plagiarism concerns

I’ve second hand accounts of meetings focused on exactly the problem of how much verbatim text was being spat out and what to do about it. Some versions of specific models are far worse at this but it’s a huge legal liability not to address it.

Amazon’s logistics are shit.

You’ll get something to show up on schedule but it’s frequently needs to be returned because of a problem on their end. No I don’t accept paying full price for a used product in subpar repair. No I don’t want a convincing fake.

The hassle of even 1 return in 20 isn’t worth faster delivery. Worse even if I just throw away their junk I still need to make a new order.

That’s a weak follow-up, the first question was a solid 9 on the asshole scale. I’d have trouble justifying a pity 3 on your second attempt.

> how insanely much compute AI needs

Needs isn’t quite the right term as AI energy use is way higher than the market actually supports.

A large fraction of AI’s energy use is pure bubble me too redundancy. A long list of companies are doing essentially the same training with essentially the same data. When grok dies all that training goes away independent of any use of AI by consumers.

Similarly, by subsidizing usage with VC funds you get a lot of demand that goes away at market rates.

People use language in their own distinct fashion. To the point where you can often figure out the writer from a long enough sample. Newspapers and internet comments end up a blend of different voices but in visually distinct boxes.

LLM intentionally randomize output to cover up all the plagiarism, but this largely destroys the consistency that cause writers to feel unique. Larger patterns still emerge but it just feels stale.

> I happen to have been following research around beta-carbolines for almost two decades and am curious whether clinical practice follows that research.

So you’re just clueless here, wow what a rude question.

> Did we? Where are those fusion plants installed?

Oxfordshire etc but most of the cost is stuff you’d see at any nuclear power plant.

> Only $11 billion of the $183.3 is true expending

That’s not how money works, there’s zero difference between a tax break and handing out money because money is fungible. The IRS will write you a check on a “tax” break.