Hacker Newsnew | past | comments | ask | show | jobs | submit | B4uler5's commentslogin

I bought a macbook pro in 2020, and for the last 6 years, it has been a workhorse of a machine, it is dear to me in ways I'm ashamed to admit.

This year though its been feeling slow. Slow in ways that don't make sense for a 6 year old laptop and as you can probably guess downgrading helped a lot.

Tahou ruined the experience for me, and after trying and not loving the experience provided by the mainline T2 linux kernel, I decided to buy an upgrade.

I pre-ordered the Framework 13 pro with the intent of having it stick around as my main machine for at least the next decade. That was until yesterday when I came across this project.

Its a collaborative targeted effort to bring the experience of using a T2 mac with linux inline with macos (or what macos was before the OS tanked the performance with its fancy glass effect).

So...guess I don't need a new computer after all!


I made the same choice albeit a year ago. Dont bother with Framework their stuff is super clunky and super expensive.

I use t2linux.org, the install of Mint worked fine on one of my macbooks. The other I keep around only for logic pro. Have had no problems with it.

It is not my daily driver as I have moved on to non-apple hardware for good. Anyway t2linux.org works fine I dont know about this OPs project.


Yeah, but why are we accepting that? Do we envision a scenario where code will be so black boxed that software creators will be pinching pennies and cutting corners so they can eek out a profit margin over the cost of development? It seems insane that we’ve willingly taken this route, the power when using LLMs is immense, but the power we’re giving up is all we had.

I saw a youtube comment a few months ago that has stuck in my head, can’t remember the exact wording but it was something along the lines of “I’ve never seen an industry so excited to cuck itself out of a good wage”.


Yeah? You reckon they are on the same level? I wasn't there but I find it hard to believe people were as worried about PHP as they are blindly prompting an LLM for 10 hours a day.

That's just your brain telling you to engage it every now and then lol "Please, don't let me become slush!"

I have never been thinking harder than I do these days: Engaging with mountains of agent-generated text, weighing decisions and trade-offs, correcting the agent's half-baked ideas and misunderstandings, reviewing the bizarre code that they generate... If I'm not just sick (and I'm pretty sure that I am), then what I have is probably closer burnout than atrophy.

Tbh, its both. Working with multiple agents is exhausting and definitely burns you out but you also do lean less on your ability to push through boring or difficult problems. I’m finding that with lighter AI involvement I am more invested and less burnt out.

This is my problem with it, I can’t tell if people actually think the LLMs writing all their code is the future or they just want it to be the future because they’d prefer not to think and still get a dopamine hit from it.

A compiler is deterministic, even if you don’t understand the assembly at least it will do exactly the same thing every time. I absolutely loath this comparison.

I’ve been using Pi since April. I switched from using copilot CLI which definitely isn’t the most frontier harness going but was more “feature rich” at the time than the stripped down feature set you get with Pi out the box.

The bit I was instantly drawn to was how much access you are given to change every part of it. Its similar to the text buffer in emacs where every cell in the buffer is fully accessible to any edits you need to make. My configurations are fairly minimal, but for instance I match the accept cli calls that could be destructive and edits to files in the same way copilot is set up.

Currently my pi config turns it into a tutor. It no longer makes edits, just advises and is designed to teach. I make all the code edits by hand. Use deepseek flash with it because its stupid cheap. It works for me, but honestly you can set up pi in any way you want which is why I personally think its the best harness by virtue of anything being possible as long as you can dream it up!


Could you share your setup? is it skill-based? I like the idea of using the model as a tutor/pair programmer so that I stay more grounded rather than delegating too much of the coding to it.

Its mostly just in the SYSTEM.md file. Its quite strict honestly, the rules have been built up to better curb my laziness, and also respond with statements that are better designed to teach than provide an answer.

Repo: https://github.com/MauriceElliott/PiCal

System.md: https://github.com/MauriceElliott/PiCal/blob/main/agent/SYST...

It has a break out phrase for when I have a deadline but its pretty useful in the normal mode so I’ve only had to use it a hand full of times.


I started using prime-agent [1] after reading the papers and it has been quite good so far. Better development outcomes vs OpenCode with GPT Sol 6.1.

[1] https://github.com/PrimeIntellect-ai/prime-agent


If you really want a “plan mode” you can ask Pi to add it. This is the strength of Pi. It isn’t opinionated far past excellent execution.

Anything extra you can add yourself. You just ask whatever model you’re using to query the documentation included in the pi application and then its away.


Will keep an eye on this project, love me some Ocaml! Do you have an example of it deployed and in use?

Don’t like the “fresh off the AI block” feeling of the website but could be a bit more confident if I could see it used in the actual wild!


I just wanna throw myself in here and get some advice. I’ve started using lisp recently mostly for emacs configuration but also doing a bit of recreation game programming in Fennel. If I wanted to learn a lisp for general programming/application stuff like writing a shell, a compiler, something more user facing, what would be good to check out?

It doesn't matter. Just start with a Lisp, any Lisp. Grok the idea of being REPL-driven, pick some structural editing idioms. That's all you need. Then, you pick a runtime for your needs. Depending on the runtime semantics you can choose a Lisp dialect for that runtime. Pick up some Clojure, it's enormously practical for any kind of data manipulation and analysis - small or big data. I don't even use jq anymore - studying http response in Clojure REPL is so much more convenient - you sort, filter and group data anyway you like.

I prefer the schemes.... Chicken scheme for standalone programs, s7 for embedding into larger systems. They're simple and easy to mold to whatever shape you want in my experience

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: