yeahexcept

geek.

[yeahexcept @yeahexcept]

@Suhail @trq212 Same. I want to read through the plan before committing to code, especially in long horizon tasks or when I’m exploring the solution space and want to feel out the shape of the code before writing it

Otoh you can just prompt it to write a plan to implement such and such

[yeahexcept @yeahexcept]

@literallydenis now burn them and ask it to reassemble based on the ashes

[yeahexcept @yeahexcept]

@DadaJudith @patrickc People focus on Germany because it *could* do better, but as with the rest of Europe it has cementified bureaucracy that bottlenecks initiative and has little prospects of reform. How do you reform with the same bureaucratic class that led it down this path over decades?

[yeahexcept @yeahexcept]

@R13TBR03K @AmericaPartyX The same could and has been said about every technological leap and yet we managed to thrive, often because of the science and tech. This hasn’t always been forward obvious and people’s legitimate default is to worry (also not without cause).

[yeahexcept @yeahexcept]

@abhiShandy @levelsio Bullshit.

[Eric Topol @EricTopol]

Waymo is the exemplar of superhuman AI performance to date. "It’s time to stop treating this like a tech moonshot and start treating it like a public health intervention."
https://t.co/byLC1HzazM

[yeahexcept @yeahexcept]

@robertpiosik @mrthomastaylor @thdxr That’s great. I think developing stuff from scratch with llms will be at its most effective when you limit the range of options available to the model: this architecture, this library, etc. And ship the common stuff ready made, using the model for glue and biz logic.

[yeahexcept @yeahexcept]

@robertpiosik @mrthomastaylor @thdxr Ideally, yeah, but right now coding agents tend to be too narrowly focused on the task at hand and end up messing up other pathways, requiring hand holding and directing. Still a massive help, but you need to micromanage them to get adequate results.

[yeahexcept @yeahexcept]

@mrthomastaylor @jamonholmgren @thdxr Off the top of my head:

Proper data modelling is non-negotiable. You need to understand the concepts you’re working with. Everything is downstream of that.

ultimately, you may be able to survive those 6mo, only to be granted another 6mo but with your hands (self) tied.

[yeahexcept @yeahexcept]

@jamonholmgren 100%. “we’ll fix it later” perpetually stated. It’s a tricky thing as you also need to keep perfectionism in check and the tendency is to give in, but navigating this adequately is what makes or breaks an otherwise validated product.

[yeahexcept @yeahexcept]

@mrthomastaylor @thdxr Not really. You shouldn’t go over the top architecting for circumstances you haven’t reached, but also shoudn’t aim for slop because it tends to stick around, compound and slow you down later as you manage tech debt instead of doing actual work.

[yeahexcept @yeahexcept]

@EBL2017 I predict the team that scores the most goals will win.

[yeahexcept @yeahexcept]

@DavidBCollum Because most of the human world is designed to be interacted with by humans, thus training data availability. I'm inclined to think it's a step towards specialised designs but for now, general purpose humanoid robots seem like the best bet for baseline utility.

[yeahexcept @yeahexcept]

@yacineMTB you must've stumbled across some php. it's normal.

[yeahexcept @yeahexcept]

@nickcammarata I don't see the big fuss with models using emdashes — (sic) yes it's a crude heuristic for detecting genai slop, but they are used correctly and enhance the readability of the text.

[yeahexcept @yeahexcept]

@TheAhmadOsman @d33v33d0 the likelihood we’re being trolled is above 0

[yeahexcept @yeahexcept]

@donvito Not to say they are the same but “best” labels tend not to stick for long.

[yeahexcept @yeahexcept]

@donvito Coding agents being so new and improving means any pinning one down as the best is temporary and likely to come down to getting stuck on something with agent A, starting from scratch with agent B (with a fresh context but improved mental clarity) and getting a better result.

[yeahexcept @yeahexcept]

@adarbutel @chribjel @vercel_dev @rauchg I instinctively call bullshit.

[yeahexcept @yeahexcept]

@ryanflorence broadly a cleaner version of this

[yeahexcept @yeahexcept]

@philipjohnston @Starcloud_Inc1 @theallinpod Found it, sounds great

[Philip Johnston @PhilipJohnston]

Starcloud is building data centers in space. Thanks for highlighting our work, World Economic Forum @WEF

[yeahexcept @yeahexcept]

@philipjohnston @Starcloud_Inc1 @theallinpod Can you point me to any resources explaining how this would work? What energy sources (guessing solar as the obvious one), energy storage, etc and why it would not be cheaper building it in the desert? Also, how do you get massive bandwidth at low latency in space?