Insect farming is growing in popularity as an alternative to traditional livestock and feed production. A scholar evaluates what that means in terms of trillions of insect lives.
Yup. Spend enough time with GPT-3 and its relatives like the FairSeq/XGLM family and you'll be able to spot ML writing fingerprints everywhere in English-language news media. At this point, I assume any article under one page is likely machine written, and I'm seeing more that are longer that show signs of being botted. If you look around at some of the smaller hosting services that offer commercial API access to different transformers, a lot of them specifically advertise their use for article and ad copywriting, and some note they have version that are specifically fine-tuned for those tasks to improve performance. Fucking waste of such a fascinating technology.
Do you have a list of obvious things, like weird logic issues gpt gets into or grammar mistakes? I definitely agree though. I don't doubt either that a lot of third world spam sites were the test bed.
Did anyone ever tell the jews that the beaners who ate bugs did so because they were either too stupid to learn to hunt properly or had no edible animals nearby?
Bugs were eaten out of necessity, not choice.
Even today, they're eaten as a novelty and doused in salt and spice.
The supply crisis might finally be what puts this bullshit to bed. In most cases, insects will be pretty finnicky to care for in this capacity. In mass production I doubt the viability of breeding something like crickets. Even considering some bugs that make more sense than this they still require quite a lot of space to be able to produce enough food for 1 day for 1 person.
nah this is nonsense. when you compare growing insects to raising real animals, they:
>take up much less space
>grow much faster
>are much less picky eaters
>are not subject to animal rights regulations
you can lock them in some basement all day, feed them with random garbage, and it'll be just fine. depending on what you feed them, raising insects might actually be even more cost efficient than growing crops.
anyway, personally i'd rather turn cannibal and eat the bug eating people than eat the bugs themselves. just my personal preference.
nah this is nonsense. when you compare growing insects to raising real animals, they:
>take up much less space
>grow much faster
>are much less picky eaters
>are not subject to animal rights regulations
you can lock them in some basement all day, feed them with random garbage, and it'll be just fine. depending on what you feed them, raising insects might actually be even more cost efficient than growing crops.
Compare the cost of crickets in large quantities right now to meat and then compare mass. Then try and guess how that will be scaled up. A lot of insects, even if they aren't picky eaters, do die quite easily to stresses that would be relatively minor to mammals. Crickets especially, which is why they are such a bad choice for whomever is pushing this message.
Did anyone ever tell the jews that the beaners who ate bugs did so because they were either too stupid to learn to hunt properly or had no edible animals nearby?
Bugs were eaten out of necessity, not choice.
Even today, they're eaten as a novelty and doused in salt and spice.
Yeah, like boiled silkworm larvae being a snack in China and Japan. When you have a major industry that has those as a side effect (silk making) and they are not poisonous it's no wonder somebody snacked on them. They didn't boil silkworm larvae as food, they boiled the cocoons to get the silk, and ended up with them as a "bonus". It's like saying we slaughter cows to make bully sticks for dogs. That's a side business if anything.
And most of the rest were due to starvation making people try ANYTHING.
Compare the cost of crickets in large quantities right now to meat and then compare mass. Then try and guess how that will be scaled up. A lot of insects, even if they aren't picky eaters, do die quite easily to stresses that would be relatively minor to mammals. Crickets especially, which is why they are such a bad choice for whomever is pushing this message.
you can buy live crickets for like 15 cents each already, and for now it's still a tiny niche industry. if you apply real economics of scale, insects are so much more efficient to produce than real meat, it's not even a serious competition.
you can buy live crickets for like 15 cents each already, and for now it's still a tiny niche industry. if you apply real economics of scale, insects are so much more efficient to produce than real meat, it's not even a serious competition.
Compare the cost of crickets in large quantities right now to meat and then compare mass. Then try and guess how that will be scaled up. A lot of insects, even if they aren't picky eaters, do die quite easily to stresses that would be relatively minor to mammals. Crickets especially, which is why they are such a bad choice for whomever is pushing this message.
Step 1: Replace meat with bugs.
Step 2: Rolling brownouts cause mass dieoffs of crickets.
Step 3: Massive malnutrition for plebs caused by a severe lack of protein.
Do you have a list of obvious things, like weird logic issues gpt gets into or grammar mistakes? I definitely agree though. I don't doubt either that a lot of third world spam sites were the test bed.
Usually you don't see as much of the severe logic flaws these days, those stop being as prevalent once you hit about the 6 billion parameter mark (which is now very accessible), and I'm assuming most places that aren't SEO farms seem to be willing to do a human editor pass to clean that stuff up. The biggest tell is basically the writing style, especially for the non-FairSeq stuff. Not real helpful, I know, but you know it when you see it once you've worked with the models enough. My notes below are mostly about the default flavor of output -- once you know how to work with these, you can go far beyond the default output quality and style. But it takes time and effort.
As a rough generalization, GPT3/GPT-Neo tend to be fairly laconic and dry, if you aren't specifically guiding them into a more specific writing style by priming them with example material, which these shit content mills definitely are not (no way they're putting in that kind of effort, or paying someone who knows how to make these models sing). I've heard Eleuther's new GPT-NeoX 20B has a similar "voice" by default, but I can't say from experience on that one, don't have the power to run it local yet and don't have an account on GooseAI or NovelAI for API access. FairSeq by default tends to produce more stylistically-elaborate language, at the same model size, but it caps out at 13B parameters so I can't tell you how it compares to full-fat GPT-3 DaVinci (175B paramenters), unfortunately. 13B is already pretty slick, though, especially if wrangled with care. That one has the amusing quirk of tending to default to making up news article bits when it gets confused, in my experience. (There was a LOT of news article scrapes in its training data.)
Otherwise the overall structure of their unguided output can be kind of jumpy -- you'll get a few sentences that stay on one thought, then the switch to the next concept tends to be more abrupt than if a human wrote it -- they don't necessarily understand how important transitional phrases are for a smooth writing flow. This can also snowball into the really weird "ADHD kid telling you about their day" feel you might be thinking of, especially if you don't intervene to either correct its output as it goes, or co-write with it by interjecting chunks of human-written text and making the machine take turns writing with you.
If they fuck up real bad (memory corrupted, or you hit them with text they really don't know anything about), or are very small models, you can get fun shit like typos, mangled grammar, or junk characters, but I don't see output that low quality outside of botted YouTube and Disqus comments usually. That stuff is more an issue with outdated models like the GPT-2 family, and their tiny parameter counts.
If you want to play around with a number of newer, open-source models, some of which are much larger than most of what I see on Huggingface's "Write With Transformer" (I checked "Talk To Transformer"/Inferkit and I'm not sure what model they're allowing free access to these days -- it might be the BERT version of Megatron 10B, if they're not cheaping out), you can check out KoboldAI. It adds a very user-friendly frontend, and you can either run models locally, if you have a machine with enough grunt (the 6 billion parameter models will fit on a single RTX3090, and FairSeq 13B can be stuffed onto 32-36 gigs of VRAM, if you have two GPUs. NeoX needs a pair of 3090s unfortunately), or you can spin up a Google Collab instance. They have a Collab notebook already set up to make it easy, and all of them but NeoX can be run from a free account.
EDIT -- Almost forgot. If you haven't checked out Gwern, it's got some decent articles talking about these large language models you might enjoy.
EDIT #2 -- Remembered a public sample I could share easily. There's a stock prompt that asks the transformer to produce a news article about the discovery of unicorns in a remote valley. It's a common fun test for large language models that's been around for a while. Here's a test run with GPT-3 DaVinci, it'll give you an idea of what I meant above in my sperging about their default, unguided writing style for articles. There's more GPT-3 one/few shot article-type examples HERE.
Prompt
In a shocking finding, scientist discovered a herd of unicorns living in a remote, previously unexplored valley, in the Andes Mountains. Even more surprising to the researchers was the fact that the unicorns spoke perfect English.
Output
They also were found to have perfectly coiffed hair, and wore what appeared to be Dior makeup.
“We were shocked to discover the unicorns,” said anthropologist Daniel St. Maurice. “They were like nothing we had ever seen before. We had heard legends of the unicorns, but never thought they actually existed.”
When the scientists first arrived in the valley, the unicorns were surprised and startled by the presence of humans, but were also excited. The unicorns welcomed the researchers and explained that they had been waiting for them for a very long time.
“The unicorns said that they had been waiting for us for a very long time,” said Dr. St. Maurice. “They said they had always known that humans would eventually discover them, but that they had also always known that humans would be too stupid to realize the unicorns had been waiting for them.”
The unicorns told the scientists that they had left the valley to explore the world, but that after years of traveling the world, they found humans to be incredibly stupid and decided to return to the valley to live in peace.
“We were shocked,” said Dr. St. Maurice. “We had no idea that the unicorns had been waiting for us. We also had no idea that unicorns were capable of speech. It was a truly magical moment.”
The unicorns said that they had long ago become tired of humans, but that they were excited to learn about the progress that had been made in the last 100 years.
“They were also disappointed to learn that in the past 100 years, we had not discovered the cure for cancer,” said Dr. St. Maurice. “They also seemed shocked that we were still fighting in wars.”
After a day of interacting with the unicorns, the scientists were eager to return to civilization and tell everyone about their amazing discovery. However, the unicorns had other plans.
“We wanted to return to civilization to tell everyone about the unicorns, but the unicorns refused to let us leave,” said Dr. St. Maurice. “They said they were very disappointed that we were so stupid and that we had wasted our lives on Earth.”
The unicorns then asked the scientists to help them kill all of the humans on Earth. The scientists agreed, saying that they were <sample end>
id probably try crickets and id probably like them, but since you jewish faggots want me to eat them, fuck yourself id rather starve out of spite. the Antichrist can suck my dick.
If chewing on a fried tarantula leg or biting into a crispy grasshopper makes you cringe, you could be missing out. More than a quarter of the global population regularly munches on edible insects.
They keep having to try and support it, but can't come up with any compelling arguments. It is beyond pathetic at this point. It does feel like someone has just hired every article mill to produce this crap from a mix and match of the same talking points.
The big thing that shows how big of a fail this is, is that they can't even get major celebrities to go full "eat the bug" to try and trick the masses.......
(for context: in the Judge Dredd comic, there was a famous reoccurring storyline about a bunch of wealthy mob who kept meeting to plan ways to commit perfect crimes: one of them was to have everyone become faceless bloblike creatures as a fashion trend so that everyone would become them and no one could tell who was who, which would led their gangs, having been turned into faceless blobs, commit crimes with impunity since the judges would not know who was who).
The elite won't eat the bugs personally since the whole point of the push is to force the rabble to eat the bug so they can keep on eating steak and meat without running out of it. You'll never see big time celebrities eating bugs and declaring they'll only eat bug meat going forward or pushing celebrity chiefs to make fancy bug meat meals to make it "look appealing" and trick people into spending huge amounts of money for it. It's the sort of thing where it's never going to catch on with the rabble because the people pushing it don't even hide the fact that it's another form of suborning the masses into a horrific form of serfdom control by taking away their steak, chicken, hamburger, pork from them since ZERO elites are rushing to eat the bug to "normalize" it or at least, make it trendy for the rabble.
theconversation.com
www.bloomberg.com
