Shwetabh Singh

Physics at Karlsruhe Institute of Technology. Previously University of Glasgow. LLMs, LM arch and the limits of stochastic inferencing via natural language.

[Shwetabh Singh @futilesearch]

@KalanisCalves That’s true for businesses with acceptable margins, eventually due to scale or nature of businesses the profits can balloon immensely, at that point how much can you expect them to deploy the capital effectively?

[Shwetabh Singh @futilesearch]

@larchiev Not a pixel

[Shwetabh Singh @futilesearch]

@honkeyhotay @Mubarratc @TheAhmadOsman Caffeinate doesn’t work when the lid is physically closed afaik, it’s a hardware trigger

[Shwetabh Singh @futilesearch]

@elormkdaniel This could have been three lines max

[Shwetabh Singh @futilesearch]

@mehulmpt Those concepts too will face abstraction, the work to make these things will be somewhere in the middle, systems that do not break will go fully abstract, newer building frameworks will emerge.

[Shwetabh Singh @futilesearch]

@kenwheeler How are you going to enforce a framework when the foundations haven’t settled yet.

[Shwetabh Singh @futilesearch]

@SureeeeeeeLuv @championswimmer I’m sure someone respectfully disagreed at the cusp of the Industrial Revolution too. Why is thought automation any different? There is no point to AI without humans using it, we will create needs for this tool to fulfill as well.

[Shwetabh Singh @futilesearch]

@WingEmperorBD @Rahll @newstart_2024 It’s the other way round.

[Shwetabh Singh @futilesearch]

@snoopy_dot_jpg screen -S xyz

[Shwetabh Singh @futilesearch]

@Falliblemusings In a fair economy, absolutely.

[Shwetabh Singh @futilesearch]

@st117_dmitry @arpit_bhayani Use it like a scalpel instead of a hammer. I read it somewhere and have stuck to it, works best.

[Shwetabh Singh @futilesearch]

@chastronomic Sounds good, looking forward to it!

[Shwetabh Singh @futilesearch]

@ylecun @ChrisMurphyCT Exactly, on one hand we have the rhetoric that agents are actually nowhere near actual usability, but then we also have these autonomous attacks happening?

[Shwetabh Singh @futilesearch]

@nomanslandtsts @BorisMPower But you can acknowledge that circumstances have an effect on your ability to perform such duties, a human in purely utilitarian terms will not choose to be honest at the cost of losing. It’s a great social expectation, but there is also the minimal cost of making it viable.

[Shwetabh Singh @futilesearch]

@redtachyon JSTorch when

[Shwetabh Singh @futilesearch]

@ElanaPearl One bug I came across was that for some reason torch.ones() will always initialise the tensor on the CPU, whereas otherwise the default is MPS. This would throw up SiGBUS errors and it took me a couple of days to figure out and explicitly move all tensor initialisations to MPS.

[Shwetabh Singh @futilesearch]

@EmergingRoy As an economist I’d imagine the cost of having to move from one place to another would be quite be apparent to you, humans are generally rational beings

[Shwetabh Singh @futilesearch]

@connorkapoor @bradneuberg Can you link me to some feature analysis material?

[Shwetabh Singh @futilesearch]

@B3yondGenr3 @atullchaurasia Engineering with language models leading upto serving an LLM does require linear algebra and calculus, it’s an entire specialisation of the ML branch.