Andrew Heiss

Map of the Shire with forests and rivers and village names

Map of the Shire with forests and rivers and village names, but now with rasterized elevation

These work way better in 2D! Here's a before and after

Mt Doom from The Hobbit

Another example of the jaggedness—Erebor looks fantastic on the right here, but Esgaroth/Lake Town looks like it's on a huge plateau on the shores of Long Lake (and same with Mirkwood in the back—weird huge cliffs)

3D render of Rivendell

working on a blog post about it right now haha

here's a very rough Rivendell! The DEM data is really jagged, so I'm working on smoothing it out

YouTube video by Electric Cowboys Normal Car Horn vs. Cybertruck Horn #cars #automobile #tesla

just heard a cybertruck honk its horn for the first time IRL, and holy crap it's a perfect auditory counterpart to the visual hideousness of that stupid car www.youtube.com/watch?v=BOKf...

did she clean tho?

GitHub - bburns/Arda: Maps of J. R. R. Tolkien's Middle Earth using DEM (Digital Elevation Model) and place vectors Maps of J. R. R. Tolkien's Middle Earth using DEM (Digital Elevation Model) and place vectors - bburns/Arda

wait someone rescued them here! github.com/bburns/Arda

ME-GIS GIS files created by the ME-DEM team. The ME-DEM project exists to create a high-resolution digital elevation model of Tolkein's Middle Earth. The users monks and SeerBlue created a number of GIS layers to support this goal. Some of that data is now available here, as the user jvangeld improves the data for publishing in a WMS. Forums are at: http://worlds.outercraft.com/forum/index.php#c1 The 10k DEM is available in .bt format from https://keybase.pub/jvangeld/10k.bt If you want to use the data, just ask. We typically approve most personal and educational uses. Contributors are always welcome. If you are looking for the 3D planetary renderer, head over to http://forum.outerra.com/index.php?topic=1491.0

They used to exist! But that link is dead now :(

5 of 5 stars to Eleanor Roosevelt, Vol 1: 1884-1933 by Blanche Wiesen Cook www.goodreads.com/book/show/88...

Firing the BLS Commissioner — the wonk in charge of the statisticians who track economic reality — is an authoritarian four alarm fire.

It will also backfire: You can't bend economic reality, but you can break the trust of markets. And biased data yields worse policy.

Orlando Bloom Spotted At Dinner With Angela Merkel

Orlando Bloom Spotted At Dinner With Angela Merkel theonion.com/orlando...

And she’s gone!

Yep! My wife and I both went there—I did undergrad and an MPA there. My mother-in-law is library faculty there, and I did a two-year visiting gig there before coming to GSU. We have deep BYU roots :)

It'll be weird going down to n-1 kids at home, a number we haven't been at since 2021. The other kids have been like vultures claiming stuff and space 😂

without doing the math to figure out that *we are that niche*. That episode has actually been helpful for the 3-yo processing the change, so thanks PBS

Prince Wednesday, Prince Tuesday, and Daniel Tiger

Also, IT'S SO WEIRD to be at a stage of life where kids are moving away, but also while we still have a 3-year-old. There's a Daniel Tiger episode where Prince Tuesday moves to college and the little kids are sad and I've always thought "wow, that's a wildly niche situation for their audience"

ahhhhhhh our oldest is moving out today—she's off to BYU!

BarriePalmerSpirling_TrustMeBro.pdf

A+ preprint PDF name too arthurspirling.org/documents/Ba...

Basically they do a ton of coding tasks with humans and different LLMs over time to show that LLM-based coding has all the worst characteristics of human-based approaches (i.e. exact replication impossible + high fragility) & have advice for attempting replicability with LLMs

These jobs range from coding the ideology of manifestos to identifying types of protest events to detecting certain political valences in speeches. The idea is to precisely calibrate exactly how replicable one can expect machines to be in practice, and where (what types of tasks) we can expect better (lower variance, more replication) or worse performance. The human workers provide a baseline comparison in terms of replicability. At a high level, the news is bad: while it is true that LMs can be (very) accurate relative to a gold standard, they also show considerable variance over time. And this is to say nothing of cases where they simply will not run at all, and thus fail the most basic requirement (see, e.g., Benureau and Rougier, 2018) of computational replication. Contrary to popular belief, the problems do not go away even if one sets "temperatures" (or equivalent tunings) to zero; indeed, this induces new but unpredictable problems with replication. Unsurprisingly, this variance affects the substantive answers we get downstream-that is, in subsequent analysis in which the labels

3.2 The Problem with Language Model Replication The central problem with replication for Language Models is that as we will show-the process exhibits the weaknesses of deterministic, stochastic and rule-based replication, without the strengths of any of them. To make this point clear, consider Table 1. There we document replication practices as a typology. What defines the typology is first, whether exact replication is possible; second, whether replication is fragile in the sense we discussed above. With a 2x2 table showing that LLms are not exactly replicable and are fragile

Full results for each outcome and run are displayed in Figures 8 to 10 in SI C. We also give descriptions of what we found. For now, we summarize our main observations: 1. For the manifestos, the crowdworkers perform very well (by LM standards) and their variance is generally lower than the LMs. 2. For the protests crowdworkers are less accurate than the LMs, but very consistent in their performance. 3. Crowdworkers struggle in predictable ways: for example, they are least accurate when manifestos should have 'extreme' codings (far left /far right). 4. LMs struggle in unpredictable ways: for example, GPT made errors on more moderate (liberal manifestos, but it is hard to know why. 5. Comparing across LMs, errors and performances appears to be idiosyncratic: for example, Llama has recall on some tasks on a par with GPT but generally much lower variance. 6. Open LMs have the best replication performance, at least in terms of low variance. For instance, on the static tasks, Llama has practically zero variance in its coding performance.

3. Consider open models that allow offline versioning. We found that, uniquely, our open-weights implementations were replicable to a high standard if that standard is low variance. That is, if the goal is something approaching the Deterministic 'code and data' replication vision above, then local, versioned models are the way to go. These may not deliver top of the line performance (e.g. accuracy) but should be checked as a first resort. We acknowledge that an open LM may not be "transparent" in the sense that it is "easy" to understand how it produces predictions even if one has the weights. But it is obviously a boon to replication insofar as being able to verify that the original researcher did indeed see the results they reported. What is more, recent research into LM interpretability points the way toward more model understanding and control but only if weights are accessible (Cunningham et al., 2023).

Finally got to read this new paper by @cbarrie.bsky.social & @lexipalmer.bsky.social & Arthur Spirling on the lack of replicability in LLM-based research and polisci and it's so good and concise and well-reasoned! arthurspirling.org/documents/Ba...

BarriePalmerSpirling_TrustMeBro.pdf

A+ preprint PDF name too arthurspirling.org/documents/Ba...

Basically they do a ton of coding tasks with humans and different LLMs over time to show that LLM-based coding has all the worst characteristics of human-based approaches (i.e. exact replication impossible + high fragility) & have advice for attempting replicability with LLMs

These jobs range from coding the ideology of manifestos to identifying types of protest events to detecting certain political valences in speeches. The idea is to precisely calibrate exactly how replicable one can expect machines to be in practice, and where (what types of tasks) we can expect better (lower variance, more replication) or worse performance. The human workers provide a baseline comparison in terms of replicability. At a high level, the news is bad: while it is true that LMs can be (very) accurate relative to a gold standard, they also show considerable variance over time. And this is to say nothing of cases where they simply will not run at all, and thus fail the most basic requirement (see, e.g., Benureau and Rougier, 2018) of computational replication. Contrary to popular belief, the problems do not go away even if one sets "temperatures" (or equivalent tunings) to zero; indeed, this induces new but unpredictable problems with replication. Unsurprisingly, this variance affects the substantive answers we get downstream-that is, in subsequent analysis in which the labels

3.2 The Problem with Language Model Replication The central problem with replication for Language Models is that as we will show-the process exhibits the weaknesses of deterministic, stochastic and rule-based replication, without the strengths of any of them. To make this point clear, consider Table 1. There we document replication practices as a typology. What defines the typology is first, whether exact replication is possible; second, whether replication is fragile in the sense we discussed above. With a 2x2 table showing that LLms are not exactly replicable and are fragile

Full results for each outcome and run are displayed in Figures 8 to 10 in SI C. We also give descriptions of what we found. For now, we summarize our main observations: 1. For the manifestos, the crowdworkers perform very well (by LM standards) and their variance is generally lower than the LMs. 2. For the protests crowdworkers are less accurate than the LMs, but very consistent in their performance. 3. Crowdworkers struggle in predictable ways: for example, they are least accurate when manifestos should have 'extreme' codings (far left /far right). 4. LMs struggle in unpredictable ways: for example, GPT made errors on more moderate (liberal manifestos, but it is hard to know why. 5. Comparing across LMs, errors and performances appears to be idiosyncratic: for example, Llama has recall on some tasks on a par with GPT but generally much lower variance. 6. Open LMs have the best replication performance, at least in terms of low variance. For instance, on the static tasks, Llama has practically zero variance in its coding performance.

3. Consider open models that allow offline versioning. We found that, uniquely, our open-weights implementations were replicable to a high standard if that standard is low variance. That is, if the goal is something approaching the Deterministic 'code and data' replication vision above, then local, versioned models are the way to go. These may not deliver top of the line performance (e.g. accuracy) but should be checked as a first resort. We acknowledge that an open LM may not be "transparent" in the sense that it is "easy" to understand how it produces predictions even if one has the weights. But it is obviously a boon to replication insofar as being able to verify that the original researcher did indeed see the results they reported. What is more, recent research into LM interpretability points the way toward more model understanding and control but only if weights are accessible (Cunningham et al., 2023).

Finally got to read this new paper by @cbarrie.bsky.social & @lexipalmer.bsky.social & Arthur Spirling on the lack of replicability in LLM-based research and polisci and it's so good and concise and well-reasoned! arthurspirling.org/documents/Ba...

yes (<ducks from cringe police>)

If Gandalf didn’t want Frodo to put on the ring he shouldn’t have called it “quite cool” when he gave it to him

I think it's basically ISO 639-2 or a subset of 639-3

wooof no 🤮

HWÆT! Hæle min sun

List of languages like Dzongkha Eastern frisian Efik Egyptian (ancient) Ekajuk Elamite English English, middle (1100-1500) English, old (ca.450-1100) Erzya Esperanto Estonian Ewe Ewondo

Checking in a kid at urgent care and have to choose a primary language and I’m amazed at the list they’re using. Almost chose English, old or Egyptian (ancient)

We have appended an Editors' Note to a story about Mohammed Zakaria al-Mutawaq, a child in Gaza who was diagnosed with severe malnutrition. After publication, The Times learned that he also had pre-existing health problems. Read more below. Children in Gaza are malnourished and starving, as New York Times reporters and others have documented. We recently ran a story about Gaza’s most vulnerable civilians, including Mohammed Zakaria al-Mutawaq, who is about 18 months old and suffers from severe malnutrition. We have since learned new information, including from the hospital that treated him and his medical records, and have updated our story to add context about his pre-existing health problems. This additional detail gives readers a greater understanding of his situation. Our reporters and photographers continue to report from Gaza, bravely, sensitively, and at personal risk, so that readers can see firsthand the consequences of the war.

OMG

Here is #TidyTuesday Number Two for my Data Visualization Class!