← Feed

Posts

• Post #2822870

  The perfect live album doesn't exis... https://music.apple.com/gb/album/nine-inch-noize/1893171321

• Post #2822869

  It will always be possible to identify LLM writing (independent of any scaling or quality changes in the LLM). LLMs generate text through the mouths of millions, but cannot perceive the volume of text they extrude into the public realm. People do see this, and through the oppositional process of creating personal and community identity, develop new voices. Hence the LLM is oversampled, instantly outdated, and exposed. https://wil.to/posts/llemdashes/

• Post #2822868

  the bluespec contrib AXI module allows specifying subordinate devices with idwidth=0. This isn't valid and I should probably fix it

• Post #2489163

  when I boot my switch 1 into lineage, it briefly decides it's the greatest usb-c top and tries to charge my laptop This goes badly

• Post #2489162

  step -1) exec's use LLMs to brainstorm workplace policy step 0) LLM suggests a policy (measure tasks performed) with a perverse incentive to make up tasks to spend tokens step 1) 110% of Amazon's profit redirected into tokens for tasks impossible to fail https://www.ft.com/content/8ee0d3ef-9548-422d-8ff1-ebd48ad4b2ca?syn-25a6b1a6=1

• Post #2039986

  @david_chisnall I think it's unlikely there will be many quantum computers. The recent factoring cost guesstimates show time-to-solution for a 2x larger machine being far more than 2x shorter, which suggests it's always more cost-effective to scale-out a shared machine.

• Post #2039985

  @david_chisnall I, wearing a hat I'm paid to don, say via modular scaling using photonic interconnects (see https://arxiv.org/abs/2501.14029).

• Post #2039984

  @david_chisnall Under reasonable assumptions regarding the quality of remote entanglement, it does not matter if you build your QC using one phenomenally-large device, or many smaller devices. Thusly, more qubits are available at linear cost* and no additional design time**. For any individual QC manufacturer, when to start scaled production of clusters is a tricky decision***. But it will happen. *,**health warning asterisks, manufacturing is hard. ***silicon spins possibly excepted

• Post #2039978

  FT reports: Trump says US ‘thinking’ of ‘joint venture’ with Iran to charge tolls on Hormuz The don wants his cut!!

• Post #2039977

  @whitequark yes, it is. I didn't even know about the police registration requirement. Many people who consider themselves internationally-minded and tolerant have no idea about any of this.

• Post #2039975

  yess BlueSpec's FSM's getting deserved love! You can do truly complex things and they don't fall over! https://www.youtube.com/watch?v=ztUZWtoZKiM

• Post #2039974

  re GME buying ebay and firing the marketing department: I'm against this kind of thing in general but as a eBay user, PLEASE stop the marketing

• Post #2039973

  TIL the name for the crazy US thing where you have 64 lanes, 32 each direction, and no median: "undivided highway". Hope you are good at counting!

• Post #653527

  the FT reports today that the US Military is using an AI kill chain, built by Anthropic and Palantir. They use it to launder inadequate evidence into kill decisions no-one can be blamed for, to perpetuate a terror program where it's impossible to tell who counts as a "combatant". And they will kill you, your children, your neighbours, and your children's entire school, for reasons you will never know.

• Post #605467

  Why do LLMs produce the sensation of connection with a conscious being in many, yet AI imagery does not?

• Post #605466

  (at anthropic HQ) We've caught another one! https://arxiv.org/search/cs?searchtype=author&query=Borrill%2C+P

• Post #605465

  The paper: Cardiorespiratory fitness is associated with <generally being a good person> The methods: n=40 come onnnnnn guys

• Post #605463

  formal verification of artefacts: famously very approachable, pretty reasonable to set verifying something like a C compiler for a graduate class. Oh, wait. it't not that easy. oh. oh well I'm sure the LLMs will be fine and won't cheat the test

• Post #605462

  Gorton & Denton polls closed. Reform does not need even in principle to defend a Green peg on Kalshi. I'm optimistic!

• Post #605461

  🟩🍉🟩🍉🟩🍉🟩🍉🟩🍉🟩! Go Greens!

• Post #605460

  RE requiem is dire. really really bad. product placement misogyny fatphobia misogyny again and marvel movie "quips". Whist every UE5 effect slider is at max.

• Post #314777

  ... is "Grokking" AI people not plotting loss on a ln(1-accuracy) axis?

• Post #314776

  Macron is serving in those sunnies. That’s a man with the codes

• Post #314775

  @dee @jen @aemylt ?

• Post #314774

  @gigabecquerel MBE target chamber? I've very low confidence in this as that coppery looking maybe-source has a wild crater in it, but maybe that was afterwards?

• Post #314773

  AI writing sounds like the voice of a wannabe demagogue stewed in LinkedIn

• Post #314771

  When you tie a bot to the internet you get a perfect simulacra of a Guy Who Cannot Log Off (because there are no logging off moves available to the Markov chain). And what's the currently existing type of guy who _never_ logs off that the simulacra will emulate? Conspiracy theorists, probably. ClawAnon.