weeksie

The way to tackle that is with RAG and local embeddings so that you can set examples for the code conventions you prefer. ChatGPT isn't going to do it without a lot of manual copy/pasting, and most tools I've seen are not that great. I just use a bunch of hacked together scripts and a local embed db with my code as an assist and it's worked pretty well.

Yeah, it really didn't seem like there was a lot of "there" there. Maybe I'm missing something in the codebase but they're just piping specs to the gpt-4 api?

When I was a kid on the gillnetter I'd read that book all the time and would practice tying some of the more elaborate knots. I wish I still had my dad's old marlin spike. Had it for years, with a wobbly old turk's head I'd tied on the eyelet. Gobbled up by the mists of time (an old girlfriend threw my shit out a dozen years ago and it was a casualty)

The best way to read terrible fiction is to join a critique group and read other amateurs. This is good because you can run across different levels of writers:

* People who have no business writing and have no idea what they're doing. These are not going to give you much, which is why you want to keep to groups that have some gatekeeping.

* Beginners who are not that talented but can develop skill. These people are good because you can see what it looks like to go from nothing to something, even if that something isn't great

* Untalented people who are workhorses and can do the job: these people are probably the most valuable "bad writers"

* Talented people who suck. These people are also amazing because you can see the power of skill as it develops

* Talented people who are very good. (These won't be in your group)

But really all this about "analyze why the book sucks" is a red herring. What is far more instructive is understanding why something that you don't like is appealing. If you can find what works in something you don't like, it's a lot more likely that you'll be able to understand that technique.

Best selling thrillers with bad dialog and paper characters? Great. That means you can focus on the pacing and plot and understand why the book works despite its obvious flaws. That's harder though because it isn't just stroking your own ego by saying "aw I could do that"

Interesting. There are a lot of old faith Russian communities in Alaska. I'm from a small town and we'd often have "locals vs Russians" teenage feuds through the summers. They're good fishermen.

This is also the case with any thread that touches on economics. I'm sure there are a lot more but if they land outside my Murray-Gellman bubble I don't notice as much.

A (very) tiny bit of econ knowledge is dangerous, isn't it.

Markets find prices, yes, but that's not applicable to what we're discussing. Maybe in some Georgist utopia we might have housing prices being priced purely at their utility.

But that's not the NIMBY complaint. The NIMBY complaint is that wherever they live is at the perfect balance where any more development would only make things worse. Game theory and equilibrium pricing have nothing to do with that because in our current regulatory system, land owners can extract rents by blocking new development.

Every time I post in this Dunning-Kruger tar pit I'm reminded of why I usually abstain.

Extraordinary

> you are actively sacrificing what makes it desirable in the first place

Amazing how the carrying capacity of every location just happens to be exactly the number of people who live there at this moment in time.

> I also think in general you should listen to what local people want and not advocate for things against their interests. Just as a rule of thumb. The alternative is basically colonization (pushing an agenda against the interests of the people who live somewhere).

The local people quoted in the article want to be able to live somewhere. Building enough housing for them to live in gives them that, even if they don't understand it because their heads are full of xenophobic nonsense.

The guy who I quoted above is a renter, he will eventually be priced out. The guy who owns the house he was renting will build a new house or sell to someone else and they will rent it at a much higher rate.

Sorry, that is a very narrow point of view that's only popular as of relatively recently.

People who block housing in cities because they don't want the cities to be "overcrowded" are ridiculous because:

1. Cities are by definition dense areas

2. If you don't want to live somewhere dense, move where fewer people want to live

3. Your right to your own property doesn't extend past it

4. You are destroying productivity and allowing landlords to extract excess productivity from the whole population, so the effect is almost definitely not the one that you think it is.

For smaller communities in ecologically fragile areas, there are more nuanced arguments, but even then those people should prefer managed density because:

1. Density keeps sprawl at bay and is better for the environment

2. Residents won't be able to afford to live on their little patch unless they own the land

3. Even then their area will become a museum around them and they certainly won't be able to preserve the idyll they wish to.

In any case, most of the supply problems can be solved without major changes to what people often refer to as "neighborhood character". Everything from ADUs to just allowing a few extra stories does quite a bit to increase the capacity of an urban area.

If you're arguing that Hawaii should be returned to the Natives and they should have some kind of tribal rule, that's certainly something that you can advocate for.

In a world where this part of Hawaii exists within the US system of property rights, you're arguing that we should grant the current landowners (whoever they may be) perpetual rights to the extract land rents. Of course, given how expensive housing becomes when it is restricted in a high-demand location the implication is that the only folks who will be able to live there are the rich.

Well, it's not even a matter of that. If they don't want new development then the price of what's already there will only skyrocket and that will end up excluding the residents.

In other words, by refusing to develop, the residents of Hawaii are displacing their own natives, unless they happen to own land. This alternative system where the only folks allowed to own property are those who are born there is something I hear a lot of anti-housing people advocate and it's disturbingly similar to the hukou[1] system, which we probably don't want to emulate.

1 https://en.wikipedia.org/wiki/Hukou

Municipalities all over the world refusing to build enough housing (and hotel space) is what's pricing people out. These are all pretty simple problems to solve but there's fractal resistance to making enough room.

In New York we built like 100k new units in the 2010s but we grew by 500k people and still, the default position of people who theoretically advocate for cheaper housing is to block new development.

> "I'm more concerned of big land developers coming in and seeing this charred land as an opportunity to rebuild,"

Of course, people like that will organize to block these big bad developers making housing even more expensive and the only people who will be able to afford to live there will be the rich.

Larry shapes the "Art World" both in and out of the US

I'm reminded of the Gay Talese essay that opens with an account of vast numbers of feral cats waking up all over New York. I'm surprised the story didn't mention that the city used to be full of them.

This book is near and dear to my heart. Back in the mists of time (~05?) when I was learning Haskell I reimplemented several of these in order to get my head around things. Lots of fond memories.

Ha! I love it.

Lots of them mixed in. We get a ton of alders and birches, but in the winter it looks like everything is coniferous b/c the leaves are off the others. Where I was it was lots of spruce, but I grew up in the coastal rainforest in Prince William Sound and Alaska is a big, big place.

Growing up in Alaska I had always taken it for granted that when Bob Ross painted outdoor scenes they looked just like the outdoors I was used to. It's funny how that never stood out enough for me to see it as peculiar, given that absolutely no other media showed landscapes that were similar to the one I grew up in.

Yes.

Javascript is this generation's C++. It's a massive language and the only way to stay sane on a project is to agree to use a well demarcated subset of it.

Nothing wrong with being C++. The reason JS is so massive and weird is because it's the language that everybody uses, or has to use at some point. Upsides and downsides.

I dunno. Fucking duh, a government bond that pays you a fixed amount of interest is going to lose value when inflation is higher than the loan rate.

I don't think you're getting what I'm saying. I'm _faster_ without autocomplete

I've used it and it did nothing helpful. I also find autocomplete slows me down. The code it suggested always needed enough reworking I would have been faster writing it out from scratch. It's just not that helpful for me. Maybe if I didn't know the apis that well but I suspect even then it would be as much a liability as a benefit

I almost always turn autocomplete off except in circumstances where the API has bad documentation. I also found that copilot was an aggravation more than a help after using it for a couple weeks.

A fun experiment but LLM generated stories will never escape the uncanny valley, because they're not novel, they're remixes. I know the critique of that position is "but humans just remix the things they know from the world" but that is the same misunderstanding that those who thought truck driving jobs would be automated by 2018 made about that profession. i.e. The less you know about a field, the more likely you are to believe it will be easy to automate.

Art, in this case illustration and narrative, requires a coherent viewpoint. In literature this is often referred to as the authorial voice. Voice is special because of the uniqueness of the human life behind it, which I realize sounds a bit airy-fairy, but is very true once you start digging into what works and what doesn't. When you read something and wonder why it's so engaging when it's not _that_ much different than other very similar pieces, that's often what's at work.

It's also why some pieces really don't work for some people. The "a kid could draw that" critiques of various modernism styles come from this place.

I don't begrudge people experimenting with art through LLM generation, and perhaps it can be a great tool for scaffolding ideas, I grant that. However, I do worry about the deluge of poorly conceived LLM content that will make the good stuff harder to find. We had a glorious age where the gatekeepers had lost most of their power. With the flood of content about to be released, I'm afraid we'll find ourselves beholden to them once again.

I think I agree with this take.

Is it conceivable that a model could leak PII that is present but extremely hard to detect in the data set? For example, spread out in very different documents in the corpus that aren't obviously related, but that the model would synthesize relatively easily?

Copilot has a ton of well publicised examples of verbatim code being used, but I didn't realize that it was as trivial as all that to go plumbing for it directly.

This was mostly political guff about environmentalism and bias, but one thing I didn't know was that apparently larger models make it easier to extract training data.

> Finally, we note that there are risks associated with the fact that LMs with extremely large numbers of parameters model their training data very closely and can be prompted to output specific information from that training data. For example, [28] demonstrate a methodology for extracting personally identifiable information (PII) from an LM and find that larger LMs are more susceptible to this style of attack than smaller ones. Building training data out of publicly available documents doesn’t fully mitigate this risk: just because the PII was already available in the open on the Internet doesn’t mean there isn’t additional harm in collecting it and providing another avenue to its discovery. This type of risk differs from those noted above because it doesn’t hinge on seeming coherence of synthetic text, but the possibility of a sufficiently motivated user gaining access to training data via the LM. In a similar vein, users might query LMs for ‘dangerous knowledge’ (e.g. tax avoidance advice), knowing that what they were getting was synthetic and therefore not credible but nonetheless representing clues to what is in the training data in order to refine their own search queries

Shame they only gave that one graf. I'd like to know more about this. Again, miss me with the political garbage about "dangerous knowledge", the most concerning thing is the PII leakage as far as I can tell.