AnIndefiniteArticle

Avec les yeux ouverts, on peut voir le monde.

Word to the media: don’t bother us with this shit until they have a working idea on how to monitor and regulate the shit so we can all have flying cars and not just one dude’s stupid prototype. The hard thing isn’t making a car that can fly. That shit is relatively easy. The hard part is figuring out how air traffic would work if everyone was up there. It’s already a PITA and there aren’t a helluva lot of planes compared to cars.

If they’re all automated and coordinating, NASA has put a lot of effort into researching how to coordinate a swarm of airborne agents with diverse goals.. I’d imagine that these prototypes have gotten further development and a reality-collision as drone systems have been deployed and evolving in the Ukrainian theatre. The tech for this might actually finally be ready.

Try asking one to write a sentence that ends with the letter “r”, or a poem that rhymes.

They know words as black boxen with weights attached for how likely they are to appear in certain contexts. Prediction happens by comparing the chain of these boxes leading up to the current cursor and using weights and statistics to fill in the next box.

They don’t understand that those words are made of letters unless they have been programmed to break each word down into its component letters/syllables. None of them have been programmed to do this because that increases the already astronomical compute and training costs.

About a decade ago I played with an LLM whose markov chain did predictions based on what letter came next instead of what word came next (pretty easy modification of the base code). It was surprisingly comparably good at putting sentences and grammar together when working at the letter-scale. It also was horribly less efficient to train (which is saying something in comparison to word-level prediction LLMs) because it needs to consider many more units (letters vs words) leading up to the current one to maintain the same coherence. If the markov chain was looking at the past 10 words, a word-level prediction has 10 boxes to factor into its calculations and trainings. If those words have an average of 5 letters, then letter-level prediction needs to consider at least 50 boxes to maintain the same awareness of context within a sentence/paragraph. This is a five-fold increase in memory footprint, and an even greater increase in compute time (since most operations are at least of linear order and sometimes more).

That efficiency hit would allow for LLMs to understand sub-word concepts like alphabetization, rhyming, root words, etc. The expense and energy requirements aren’t worth this modest expansion of understanding.

Adding a General Purpose Transformer just adds some plasticity to those weights and statistics beyond the markov chain example I use above.

Sharing data with the plebs is the true crime.

This article’s preamble cites this news story as the motivation for writing the rebuttal presented.

Literally the news story that this author cites as motivation for writing this article in the preamble to the article.

They gain by preventing copycat killers. If other poors think that he got away with it, they might think they can, too.

Good catch! Thanks for checking the source.

I love organic maps and openstreetmaps. The biggest thing missing is satellite view. I like to wander around and explore an area on maps before visiting. OSM has more interesting/relevant details and better visual color coding than the vector street map on google. Google has a satellite map, which is non-negotiable for me especially if I need to quickly orient myself while driving in a new place. I use three layers loaded into qgis for planning trips: OSM, google maps satellite, and a topographic map from USGS. I sometimes use organic maps on my phone if I don’t have access to a computer with qgis. I rely on Google while on location because organic maps lacks a satellite feed.

Yes.

These ecosystems are well studied.

This paper from the oil industry opens with the sentence “Methane-oxidizing bacteria (MOB) have long been recognized as an important bioindicator for oil and gas exploration.” We literally look for these ecosystems to know where to extract their food.

This isn’t controversial in the slightest. We are destroying unique ecosystems with every barrel we extract.

Ok, but what about the ecosystems dependent on that chemical energy staying underground?

It’s a thermostat.

I’m coming from a field where supporting software written in the 70s is the norm.

Your argument is horribly short-sighted and wasteful.

Only 16 years old is extremely recent software that ought to be easily maintained in any sane world.

~4 NASAs (FY2023)