8note

the most obvious example was counting the Rs in strawberry.

figuring out what is and isnt in distribution is quite difficult though given that theres so much data passed in, and transformers have some amount of combining it together

theres some really obvious things though, like picking nonsense tokens together, or things outside of the tokenspace entirely

you could also construct a string that is a series of the least likely tokens in a row. you wont get it to output the next token, despite it being quite clear how you did it

what about a house cat?

they clearly outhink the sparrow at sparrowing in order to predict what the sparrow will and wont react to, and how theyll try to escape the cat

that would imply the learning of what the weights mean can be done unsupervised which is really cool.

it probably has however been trained on papers talking about ways they extracted subsets of LLMs as say, the part that knows chess, and would be able to then use similar commands to process the weights

i think a better question and one that probably has a lnown answer is whether there are infinite nontrivial solutions to train a neural network of a certain size to match a function? Trivial here would be about simplifying `n * 2 / 2` from two neurons to just passing n from the input

do they break down to a finite subset of groups?

unless you are doing something where you stay in latent space without having a random number generator pick a word, the next token predictor part still matters. without making the tokes concrete in between, that plan doesnt exist

if it was designed to edit code, id expect the first translation from latent space out to tokens to be the final code, not be assembling the code by random number generator.

its not really designed to edit code until the tests succeed, we try to manipulate the distribution of tokens to predict by prompt engineering with the test results.

once the TV is in my hands, its up to me what i do with it.

if they made a bad business decision thats up to them

they dont have to.

for example, government ownership of automation would increase the government's spending power

huh? how does the US get to a surplus with deficit spending?

the big spend since 2001 was going to war

> I'm not sure how the economy could do this on its own, we're doing this to ourselves.

i dont understand. the economy is the royal we doing things. the economy doesnt experience time if we arent doing thongs to ourselves. this isnt a distinction with a difference

doesnt social security fund itself?

why is fiscal responsibility an issue for social security, beyond governments stealing from it?

how many cuts to education or defense spending save social security?

its certainly benchmarkable.

make a bunch of "build the feature" vs "heres a detailed spec" prompts, and see which one does better.

if you take out the AI boogieman, and instead go with "next token prediction" then "people are doing it wrong" is much more reasonable

it definitely is real. the so big to be outrageous one is being made in utah, by the same guy

it would be apt if it was "fun fact, despite being incredibly destructive, a nuclear bomb doesnt leave much waste heat" but also a non-sequitur.

a really obvious electricity in, waste heat out is aluminum smelting in quebec. at ~12GWh a day for the biggest ones, this would come out at a similar scale, but make clear that the data center industry in alberta produces far less waste heat than the aluminum industry in quebec, and we should start looking to shutter all those smelters for wasting too much power

bringing up nuclear bombs when the heat output of a nuclear bomb isnt what people are familiar with is intentionally misleading. it also elicits a "cool, this thing can survive 20 nuclear blasts a day?" response which is pretty obviously not intended and isnt true

Canada's is a addon to the aluminum production line, ao it might not take a whole ton to expand that production

thats reasonably the state of anthropics code output as well - not much care and polish

idk i think theyre less idealistic and more monopolistic.

if they think their model is the best at making things safe, then their model should be helping train at any chance, to embed the same safety into new models

with better being more convincing, rather than just stronger

its the same for like a chess bot. i dont want a perfect stockfish that occasionally makes a serious blunder when a dice rolls a certain way. i want one that actually falls into traps

i think it can work for procedurally generated situations, and the llm is acting more like a translator between data model and player rather than being a data storage layer.

thats close to tweaking existing dialog, but with a bit more freedom.

you still dont have to give the player arbitrary response

i think hooking in an llm to dwarf fortress for turning the very mechanistic descriptions into prose. same with the adventure mode talk. instead of a constant "well weve got beasts" an llm could read a bunch of the language tree to make a unique introduction for each.

same with the df artworks. get kitfox to design some art styles for dwarves, elves, goblins, and humans, and then instead of just reading

"this is a fine engraving of a dwarf. it concerns the events of 102 when ulrist fastbuttons was elected mayor. ulrist fastbuttons is sobbing"

i think its more of an effect of factorio's hidden information.

youre not really sure how to set up some of the logistics/how much of what you will need where ahead of time

i doubt it.

that could be done directly by taxing IP licensing directly

russia? others who dont care all that much?

the laundering wouldnt be all that hard to take those buttcoins on in exchange for clean ones.

doesnt work as of today

it is enforceable, just the wrong rule.

the rule is ultimately "dont submit stuff where only your llm session knows whats going on"

if you brainstorm with an llm, then rewrite the llm prose but have no understanding, youve still broken the rule.

if you understand what it suggested, why, and so on, then you do know whats up, and thats unenforceable, but who cares

i think its probably asking people to not submit things they dont understand themselves.

if you want to submit an LLM found bug, you could instead donate the tokens directly to the maintainers, rather than making them read your agent outputs

it looks like a lot of the functionality of old reddit for seeing not-ad-decided sort orders is going away, so the whole old reddit will probably be gone by the end of the year

i think its more of a face saving lie that mcmorrow dropped out over the twitter poll.

she ruined her campaign by trying to make it about hasan piker, who wasnt running, and a good portion of the people she needed as supporters kinda like piker on at least a surface check of what he has to say.

i think more what happened is all her donors were already out, and they just pointed at that poll. its not like it was a unique poll.

pointing at the fake poll gets her(and her consultants) some cover for having made bad campaign choices

i dont know if this would really count as fraud in that case either - its not like they lied about the results of the primary or anything, just influenced it by convincing people to vote a certain way or drop out.

i think this is more like putting out fake weather reports than fake corn harvest reports.

looks like after his main party dissolved, he was running as a member of the republican party?

this still seems like the republican party trying to do some election fraud against themselves?

there's an implication that the fake poll was also therefore incorrect, but its in line with what real polls showed before and after

... and its months after she tanked her campaign.

you cant really plummet from single digit to single digit polls.