Sufi

building a really REALLY good ai detector rn

[Sufi @Squishysoup]

Why does starting up a VM on nebius always get my heart racing. it's like ohmygosh im on a timer every second wasted is money runnign out oh god im going to die. etc.

[Sufi @Squishysoup]

@langurmoder It's insane from the US too. I wanna buy the welch labs illustrated guides to ai. but i'll have to pay more than the books cost for shipping its so stupid.

[Sufi @Squishysoup]

Qipao.

That is all.

[Sufi @Squishysoup]

@peptranide I'm gonna work on it a bit later, when i'm free. I'm fairly certain it should be fixable. way too busy rn though >_<

[Sufi @Squishysoup]

my gpu died :(

[Sufi @Squishysoup]

@peptranide @0xfulgrim @AniC_dev We shall break free of our chains not through speeches and rallies, but through compute and clusters. Hail, spirit of the machine, essence divine.

[Sufi @Squishysoup]

@0xfulgrim @AniC_dev you're CEO aren't you. pay up, business expense.

[Sufi @Squishysoup]

@JSBingus Thank you for the representation.

[Sufi @Squishysoup]

@_itsNotOverYet_ "Malebrained minifridge" is so good

[Sufi @Squishysoup]

What are these automated interpretability tools...must read moar

[Sufi @Squishysoup]

I'm reorganizing my obsidian repo and i feel like a second brain zettelklasten jarvis larper

[Sufi @Squishysoup]

@langurmoder deepavalimoder

[Sufi @Squishysoup]

@langurmoder The lion does not fear bright lights and loud noises. I do though :(

[Sufi @Squishysoup]

It's kind of crazy that you can go from reading a cool paper in your twitter feed to using its methods on your own model and retraining it to get a fat accuracy improvement in 2-3 days. The speed of iteration w/ LLMs is just absurd.

[Sufi @Squishysoup]

@langurmoder Autosell coming in clutch

[Sufi @Squishysoup]

@0xfulgrim Ayyyy that's our model :D. btw we know 1 comparison isn't much data, we will be releasing proper benchmark results comparing it to the current detection solutions soon (spoiler: they look really good for us)

[aarya @0xfulgrim]

Failed on a first pass on internal benchmarks, clip made using a generative model from 2023. We do it better on our internal model for a fraction of the price while running locally in real time on a MacBook Pro M1.