Anchor

40 point diff with #2 is insane. How long before this score is broken again?

this is why I believe Duolingo is designed for brat tamers

Looks like 13% of Americans top-ticked BTC

if you do not find joy in illiquid esoteric financial instruments, consider being an angel investor for their closest competitor

A large part of the electorate is completely offline and defer to whichever candidate they hear on the news more

I hear France is thinking of lending their friends some nukes...

LIMO: Less is More for Reasoning We present a fundamental discovery that challenges our understanding of how complex reasoning emerges in large language models. While conventional wisdom suggests that sophisticated reasoning tasks de...

Which is exactly how much you'd need apparently

arxiv.org/abs/2502.03387

I will resist the duress of my Green Goblin mask (15 years of muscle memory telling me to Buy The Dip)

Truth nuke season is starting early

(PDF) Deep-Learning-based Cryptanalysis through Topic Modeling PDF | Neural cryptography is a technique that uses neural networks for secure data encryption. Cryptoanalysis, on the other hand, deals with analyzing... | Find, read and cite all the research you nee...

Using NNs to break AES-128 encryption.

www.researchgate.net/publication/...

modern rage baiters are 2000s era trolls minus the passion. it's all for money and engagement, but never for the love of The game

does nothing literally ever happen

also a lot of the repetition is boilerplate to make the information palatable to humans (e.g. cliches, tropes, a story where the first 90% is build up and the last 10% is "the lesson", an LLM just needs the lesson)

we have different storage interfaces from LLMs. it's not a good comparison

o7 in chat

there's something beautiful about using chatgpt 4o to classify tweets for a classic NLP research problem. it's like DeWalt investing in a giga-advanced manufacturing facility to go on to say "ok let's use this to make better hammers"

i would kill to see this style of chart stratified by gender and country (i.e. are conservative women as violent with their metaphors?)

their API still allows reading tweets (but it's now a billion dollars)

may there be many balloons

gonna deploys bots to randomly like, mute, and rt my opps to weaken their reward signal

any investors?

This is a surprising result from trump voters

You can just tank an asteroid* hit to the head and walk away with a headache

New 12-sigma cosmological nightmare scenario just dropped

mythical arxiv pull, good stuff

1.4e17g is surprisingly large

The vibes are so bad right now that all the comments are so united in wanting an asteroid to end us that the vibes wrap back around to being good again

It's extra funny when they hate religion but don't realize they're practicing one already

i feel there is some selection bias here as sonnet 3.5 is the tech worker sweetheart (rightfully so) but my aunties have only ever heard of chatgpt

this feels like one of the issues that has an surprising split of support until someone politicizes it and it becomes a 50/50 issue in a week

you know it's a banger benchmark because even they put o1 and 4o behind sonnet 3.5

refusing to take the post down is such a weird hill to die it does make question the validity of his ranking a little

very dark, and very necessary, but, I fear that too many of their beliefs aren't falsifiable enough for it to hurt to be wrong about.

e.g. "planes keep falling out of the sky" is an unexpected consequence of this admin. "It's because of biden's DEI" -> punts it and is basically unfalsifiable