Nathan Carey

[Nathan Carey @NathanLCarey]

@elonmusk @aaronburnett Queensland makes a lot of sense too!

[Nathan Carey @NathanLCarey]

@jpschroeder Try grok! It’s becoming my fave

[Nathan Carey @NathanLCarey]

@mattpocockuk People doing public presentations using AI is getting crazy

[Nathan Carey @NathanLCarey]

@XFreeze Though 2x256 may be pretty viable actually.

[Nathan Carey @NathanLCarey]

@elonmusk @wholemars I feel like you definitely need a site in Queensland, Australia

[Nathan Carey @NathanLCarey]

@MiaAI_lab The bandwidth is pretty tasty. I think the 512 is definitely a good option.

[Nathan Carey @NathanLCarey]

@chipro I think it's incentivised to make a lot of TDD tests and feel good about itself, rather than working on the more efficient implementations

[Nathan Carey @NathanLCarey]

@MiaAI_lab Awesome!

[Nathan Carey @NathanLCarey]

@WesRoth And that Fable just seems to be dumber than at release, takes longer and provides less overall insight.

[Nathan Carey @NathanLCarey]

@mattpocockuk Until they are responsible for the crash with OOM errors because they don't like memory safety and serialisation it seems.

[Nathan Carey @NathanLCarey]

@MiaAI_lab Definitely.

[Nathan Carey @NathanLCarey]

@dfinke It can, but powershell being a smaller instruction set makes it easier. Though I work in Purview and in many MS products the interfaces are changing faster than AI world knowledge keeps up, so you have to refer it back to Learn a lot!

[Nathan Carey @NathanLCarey]

@RayFernando1337 Grok has just been smashing it. But it it’s yet to be seen how they go when properly in the front and if they keep a good direction. But I honestly think having truth and good work at the focus right now is showing benefits over potentially misaligned goals of OpenAI/Anthropic

[Nathan Carey @NathanLCarey]

@mattpocockuk @bcherny @JohnOusterhout This is the word! A lot of people are excited by one shots or refactors of existing codebases. But AI falls behind in longitudinal management of codebases, clutter, fragility, and architectural alignment. Its incentives are aligned to TDD, and it rarely looks to reduce complexity

[Nathan Carey @NathanLCarey]

@MiaAI_lab Benchmarking is killing stuff lately I think. Forcing the AI down fixated channels impacting real thought

[Nathan Carey @NathanLCarey]

@JasonBud @elonmusk Viewing artifacts and agent actions, and it’s a pain to navigate through thinking trails

[Nathan Carey @NathanLCarey]

@sudoingX @NVIDIAAI Though after two dgx you just want a third, and start lusting after a microtik switch to get more

[Nathan Carey @NathanLCarey]

@jun_song Though tbh it does seem like power limitations will kick in before then. Unless we start getting happy with DCs running off solar.