benson

[benson @benson_perry]

@jakeformayor but like… the messaging from blizzard was “this is actually better and more popular” all along the way, telling players they were wrong for preferring the old way… mark isn’t quite doing that but he obviously can’t speak without the company’s best interests

[benson @benson_perry]

@jakeformayor i think we’ve seen across the games industry that as games change to become popular with a larger audience, they make changes that actually are “worse” for the original fans. it’s not imagined, see classic wow and upcoming classic league of legends

[benson @benson_perry]

@shlummi @max_spero_ apparently, because people get caught lying about their LLM usage via pangram repeatedly. maybe they’ll wise up but for now it’s incredibly useful!

[Max Spero @max_spero_]

People often evaluate Pangram's false positive rate using pre-ChatGPT text. This is where our 1 in 10,000 FPR number comes from. But we often face the question:

"How can you prove that Pangram isn't just memorizing pre-2022 text that's already in its training set?"

To some degree, this is a fair criticism. If you're checking Moby Dick, any pretrained base will have seen this text, and therefore could have memorized that this text is human-written.

I talked to someone who wanted to test this hypothesis by testing Pangram on tokens that didn't exist on the internet. Super interesting. Of course we want to see the results. So I gave them credits, they found a bunch of old books that didn't exist anywhere, and scanned them. Almost three million words in total.

What they found was actually quite interesting! Three AI-assisted segments and one AI-generated segment. But when they looked at their data, they found that it the segments Pangram flagged were actually hallucinated by Mistral OCR. Mistral OCR had hallucinated entire tables onto blank pages, and these were the only segments that Pangram flagged as AI.

No false positives in about three million words / 8,000 segments. Pretty cool!

[benson @benson_perry]

@boneGPT @justalexoki siri is gonna be better soon, they are about to deliver on their promises from 2024 (which sounds terrible but will be good)

[benson @benson_perry]

AI has made me 10x more interested in capturing as much video/audio/photo content of my friends, family, and experiences, because previously i assumed "i'll never sort through that" and now i assume "this will let me relive these moments forever"

[Jack @tracewoodgrains]

One thing I do enjoy about the AI revolution is that it retroactively validates all sorts of data-generating hobbies as not-useless.

“Oh cool, you spent your teenage years posting on Reddit and recording Minecraft videos? Thanks, that will be very useful for our model.” *slurp*

[doomer @uncledoomer]

dont talk to me about data centers and water usage if youre not also calling to imprison the entire canadian parliament for causing 100 years of air pollution in one week with poor forest management

[Cabot Montana @AmanitaFugax]

Pro Waymo, pro data center, pro nuclear, pro solar, pro train, anti zoning, pro housing has to be one of the most correct and unelectable combinations in America

[benson @benson_perry]

@sdamico i agree but after doing some research… detecting gunshots using microphones isn’t something that we’re able to leverage into better outcomes (at least not yet). the racism claims and stuff are retarded tho

[benson @benson_perry]

@LexerLux for me at least it is an issue where the criticisms are almost always founded in some terrible misunderstanding, i had never seen opponents make the “actually detecting gunshots accurately and quickly is not that useful” argument (but i agree!)

[benson @benson_perry]

@LexerLux sorry but when you list the reasons you don’t like something and include bad ones it makes me question 😂

[benson @benson_perry]

@LexerLux you listed a bunch of reasons, only the last of which was "it doesn't improve outcomes" - i was saying that i don't care if civil liberties groups don't like it, or if cops don't like it, that has zero bearing on whether or not we should use it. why do you find those compelling?

[benson @benson_perry]

@robkhenderson usually people are just dumb, sometimes they are dumb and high, but in both cases you're confused why they are having trouble following simple conversation

[benson @benson_perry]

@robkhenderson to clarify because people are misunderstanding. interacting with a bank teller, a mcdonalds cashier, a CVS clerk - this has never been a super enjoyable experience in NYC! i reject the claim that it has to do with people suddenly being high more often

[benson @benson_perry]

@bowtiedshredder @robkhenderson this has always been the case with people working minimum wage jobs in nyc

[benson @benson_perry]

@tracewoodgrains his point that "this is part of the contract" - no it isn't? does joining the military entitle you to permanent financing _just in case_ you end up disabled by something we didn't even know about until later on?

[benson @benson_perry]

how long until @thsottiaux lets us buy weekly resets directly?

[benson @benson_perry]

@zillow is there actually no longer a way to filter to "2+ full baths?" like... why would you do this @zillowgonewild