sroerick

This is a good discussion of these principles. Largely inspired by Emacs, I built an interpreted lisp which runs in a web server and stores the AST Postgres. It gives me this kind of malleable computing and a Lispy feel with very little overhead.

Instead of calling an API, agents can just use functions in a REPL. They can also write new views and improve the program if they run into an issue. Honestly, pretty amazing stuff.

Being able to design a program specifically for the workflow I want is pretty neat. There's some vibeslop jank, but Emacs is a little bit jank too.

How about XMPP?

I had the same thought - but maybe it's good? I was thinking that if you have a decentralized store than you can connect to it regardless of dev environment. I've used XMPP for orchestration but I can see ergonomic advantages to this.

I have been using synthetic and my understanding is that they plan on hosting Kimi3. Open models mean Inferrence is distributed

I'm getting a LOT of agent activity scraping my code here, so I figured I'd submit this before Anthropic tries to release their own version lol. Not production ready, not really good for anything other than pointing Claude at it and asking questions about the code base.

How do Kimi's subscriptions work? I find their price structure pretty confusing

https://pricklypear.rocks/welcome https://pricklypear.rocks/code

Just to be clear this is not a release, lol, I just made it to show a couple buddies. My one friend was saying he learned a lot just by reading code and pointing Claude at it and asking questions about what he didn't understand. It is provided as-is with no warranty, or installation instructions. If you really get stoked about it I could probably clean up multi user stuff a bit to get you logged on, but I think even just the idea is pretty neat.

I'm doing fully interpreted Lisp with the AST stored in Postgres. I can access the REPL remotely and I honestly dont know if I'm ever going to build an API or CLI again

I have a web framework running with a lisp interpreter built in, and I think it unlocks a LOT for the LLMs.

The Allegorical Agent-Aeon?

I made an interpreted lisp and stored the AST in postgres and exposed the functions with htmx, and honestly it works pretty great.

The second I opened Deepseek, it had my harness scan my entire home dir. Not sure what's worse here.

One more thing - just seeing your affinity for Julia. I built a fully interpreted Lisp in OCaml which you might find interesting. I added my email to my profile here, feel free to reach out.

I was very happy with GLM 5.1 and Kimi 2.7. I did see an improvement in GLM 5.2. I found it to be more realistic about what was and wasn't good architectural choices.

I have a pretty elaborate harness setup, I use my own spec DSL to plan and execute code. So what works for me may not totally apply to others.

I haven't used the frontier models in a few months, so I can't speak at all to this latest batch. When I switched over from frontier I found that what I lost in training data, I more than made up for with Pi harness.

Your process seems more thorough than mine. Mine was literally "I like camels". Switching over to OCaml I found wonderful - I had not worked with FP before and I found it delightful, it kind of cured my programming burnout a bit. I hand built a CRUD app in OCaml/Dream but still feel like a rank beginner, especially when it comes to actually writing code, lol. But I'm getting much better at reading it.

I think the typing and structure of OCaml is a great fit for LLMs, but also the proximity to other things like DSLs, formal proofs, and algebraic structures has tons of untapped potential.

I'm having an absolute blast with it. I like programming again.

I posted this above but honestly OCaml makes vibe coding so nice. I went back to go after a few months with OCaml and it was deeply frustrating

OCaml works incredibly well with LLMs.

I must admit - I still don't understand macros. I get that they're code that's generated at compile time. But I don't understand how that's different than a function which evaluates other functions. I guess the latter would actually be evaluated at runtime? I think I get it conceptually but I'm not sure I have the muscle memory to reach for them. Anybody here have an "ah hah!" Moment with macros?

Pi.dev is great and with only a little customization made even previous gen open weights feel superior.

It also doesn't feel like they're trying to sell me on transhumanism all the time.

It also doesn't get mysteriously downgraded. It's just consistent, even before 5.2.

5.2 is great in a lot of ways - but it's best quality is that it gives some pushback and isn't nearly as synchophantic

The Allegorical Agent Aeon

This looks like an interesting design, but I kind of just want to wait 3 years and see if some new VC system emerges victorious rather than try out all these new things

I'm curious how folks here are using LangGraph

I'm using synthetic.new and Neuralwatt with pi and its good and also cheap

Maybe SWE is just a trade?

If you believe you have that, that's a privilege, but most people in the world don't.

All is vanity.

Its okay to be a tradesman. You don't have to join a cult and save the world. Work is hard and unpleasant. That's okay.

This is exactly it.

AI companies are becoming a catch-all / symbolic lightning rod for "the ruling class".

I live in rural Wyoming and it's a huge deal here, if these projects go through many people are affected.

Some of these data centers are multiple orders of magnitude bigger than a blast furnace.

I think people are pissed that they're subsidizing the R+D cost of their own unemployment.

I genuinely have no idea how some of these companies got so far over their skis on AI. It simply does not make sense to me.

Just call it vibe coding. You can still be an engineer.

Cool! I've been using OCaml and I'm really happy with it