RamblingCTO

Any reason you let the PM do the wireframe? I also use claude for this? I really like the old school flow of user journeys -> wireframes -> designs/components. Going to become a moat to do proper handcraft if everybody just slops around

Socially inhibited nerds than think we need it bc it's the future and "people just don't get it".

Just bc you can do it doesn't mean you should. But some people don't get it.

Claude desktop, chatgpt desktop, pi, claude code, also coop vm + pi setup for more juicy stuff. Also just setup paseo via tailscale to have something like amp. imho all the fancy setup and the skills make the agents more retarded. they access tools and skills way too broadly and it poisons the context.

also killed my hermes web setup (also via talescale) for security reasons. not interested in having something online like that with all the vulns popping up (ssh, ubuntu, tailscale etc.)

> Ironically this tiny utilty was produced to make keeping up with Ai code a little easier.

which is fine. I just think we're losing something important when we reduce friction and effort. and for me--totally personal preference--keeping raw claude output is just that. I don't wanna throw too much shade on you or your project tho, that wasn't my intention. I'm just fed up with the moronic way of how claude writes, designs, etc.

I also just went long on figma yesterday and got a yearly subscription. If you iterate on designs a lot it's the best workflow imho. claude desktop + figma is dope.

I'm doing this via claude desktop actually. A bit weird that claude uses html for that mainly, but it works well.

specific aesthetic means raw claude slop w/o changing anything about it? that's no lazy extrapolation my dude. I'm just saying it lowers the value of the project, bc I can assume that it's probably not gonna be touched any further (without throwing shade on op, I think it's a cool idea. it's just way behind it's potential bc of the claude copy and the claude design in the report). it's a few commits made by claude, I assume there was no real work on the side of op. that lowers my felt value of the project. that's neither lazy nor unfair, it's the last 2 years of vibecoding? if op decides to clean it up, I'd be happy to explore it actually. I kinda do measure projects by human effort. I don't know why you're throwing shade bc you seem to disagree.

And what would effort look like, besides inspecting code and the result to find the message I quoted, huh?

> Every Go package in github.com/peterretief/yggstore, what it builds on, and the functions in each of its files. Tests are left out.

it's a cool idea, but reads super sloppy, sorry. There is literally no effort in this.

Okay, so I'm supposed to replace a well-supported, stable company's product by a vibecoded slopfest of a vindictive, laid off engineer? Who in their right mind would do that? That's the other side of the medal of managers pushing for "we'll just use AI for that and build it ourselves".

It's not the same commodity. Most stuff has already open source alternatives.

I'm sorry, but if I'd see this when checking your GitHub in an application, it'd be a red flag for me.

The basic premise sounds fine, humans like to belong. But you really notice how cynic the takes are and that the models have been trained on the internet mostly.

> Humans are far more decent than they believe they are, and their cynicism about each other does more damage than their actual flaws.

I think that's more the truth.

Not sure I feel good about this. LLMs are so majorly misaligned. Having takes like these asks for trouble.

I always wanted to exponentially grow slop, thank you. The less I understand about what's happening, the better. Totally what I as a user want. Thanks G.

When I hand off work to an agent it always, always, always gets the level of abstraction, important details or just missed context wrong. I'd rather have a lot of reviews than hand off work. Imho it's just not there yet if you care about what is actually generated.

Pretty lame trolling attempt.

That's my point ;)

Because they can't learn out of their own lessons? That's not hypocrisy

'"Forest" is a polite way to refer to what appears to be a monoculture plantation.' That would mean that Europe almost has no Forest according to you and there's nothing worth protecting?

As the other commenter said, not helpful. If you go with cf countries, Germany is one of the worst offenders. But only because it has Frankfurt as a big traffic node and lots of hosting providers.

Super interesting! Good you find it. Sounds like math rock/post rock somehow, or at least adjacent to that.

Projects like this need to start supporting coop et al.

I'm also reading the first dungeon crawler carl series. You might really like the bobiverse, just got another book as well! Before that I read "a short stay in hell", super good! Or discordia by max barry or road kill by dennis e. taylor. All very entertaining.

it's a generalistic classifier w/training needed tho. as far as I'm concerned that is somehwat novel and very practical. you get an ok classifier without working out training data and whatnot. sure, you can do classic ML experiments, but the closest that comes to mind is auto ml. not sure how far we got there, it's been a minute for me on this front.

so I don't agree on the premise that it's "just a classifier". it's not something earth shattering, but practical nonetheless

/e: there's another post from sebastian raschka on this: https://magazine.sebastianraschka.com/p/classifier-history-a...

Super dope. If it would ship as prod ready code supporting mps as well that would be even doper.

But funny that jev is getting its lunch eaten apparently in under two weeks?

part of agentic engineering is classification as well tho. reviewing/gating, what to read etc. don't need a full LLM

Super cool, love the aesthetic. Buy testing the whole blog a bit better would've surfaced the "all posts" link going to an empty page I guess? Also: not reading code for a toy project like this vs in a commercial setting shouldn't be compared at all. Would also really love to see the token spend on this.

It's also a gamble that you don't have to look at the underlying layers because it's "the new abstraction layer". Which is intellectually incredibly lazy.

They talk about astractions as if "requirements -> code" is the same as "Go -> ASM -> bytecode". Which is just utterly stupid. But downvotes tell me a different story haha.

I'm also management btw, and the two startups I am a CTO at all have clear buy-in from other C-levels that we can't just spam AI slop w/o reading it and hope for the best. But I do see that most probably don't see it.

I really don't get this discussion at all. Ignoring human comprehension etc., every code review of agentically created code shows bad tests (lots of overfitting, testing on specific implementation details, not general behaviour/business logic), weird and repetitive abstractions, duplications of code, circumventing architecture etc. etc. It's a complete mess. Unmaintainable, bad craft, bug-prone and corrosive. That's what you want to catch. You're piling on tech debt like hell and you're gonna pay for it down the line. Look at the fucking code and then come back to argue. Jesus. I don't even understand how this is a debate. Look at the code.

> Europe's immigration problems

Nobody said all of Europes problems. Maybe don't argue against a straw man you created yourself?

You should really step outside more and touch some grass

If it has context I appreciate them because it helps me make better informed decisions. Why not? HN isn't an editorials page exclusive for posters (except YC slop lol)