HN user
LLMs as tech support is truly the singular new reason causing Linux usage to jump atm. All the other reasons are just slowly accumulative.
LLMs as tech support is the feather breaking the back of the Windows stranglehold.
You need to listen to what Bryan Cantrill said about his experience at Oracle, before you go there.
What you think of Oracle is even truer than what you think it is: "Shit mediocrity, inflict misery, lie our asses off, screw our customers, and make a whole shitload of money"
https://www.youtube.com/watch?v=-zRN7XLCRhc&t=2047s
You need to think of Larry Ellison the way you think of a lawnmower. You don't anthropomorphize your lawnmower, the lawnmower just mows the lawn, you stick your hand in there and it'll chop it off, the end. You don't think 'oh, the lawnmower hates me' -- lawnmower doesn't give a shit about you, lawnmower can't hate you. Don't anthropomorphize the lawnmower. Don't fall into that trap about Oracle.
I got access to the original closed demonoid and I could find anything I was looking for there. But I was never able to download anything as they required you to upload before you could download. I had nothing of value to share, so I was just stuck. How did people solve this?
I've seen that, but that's clearly a skewed dataset, right? People using openrouter will be those seeking other models. They are not the same kind of people who mostly just use gpt or claude.
"Normal users" haven't even heard of Bun and are still on Node.
Just use md files for working memory. No need for large context. LLMs get dumber as you put more in context, as it stretches their attention. Keeping context small is better for quality.
So far I have three different folders in my project for three different types of md documents that I keep creating and revising together with the agent: reports, plans, and reviews
Every large feature has at least one of each file created for it in that order. I start by analysing stuff and making reports that might also contain design thinking and decisions. Once I'm satisfied we have properly analysed and thought though whatever it is we're working on then I might make a plan or just jump into implementation depending on how complex the issue is. And then I might have zero or multiple reviews of different kinds by different agents depending on how much verification I think the issue warrants.
This is my super simple worse-is-better development process. I make up what is needed on the spot.
MD5crypt is a password hashing algorithm that uses the MD5 digest hashing algorithm. MD5crypt basically calls MD5 a bunch of times in order to make it computationally expensive.
Anecdotes. Where's the data?
This is similar to the evolution of Wikipedia, except the format of WP allowed it to transform into a feudal dictatorship of nerds who feel like they are deciding what's true, and they can get off on that.
Literally describing how academia has worked since time immemorial.
How good are the questions?
Yes ai + programmer are the resource.
Ai won't work on it's own until we get agi.
"In economics, the Jevons paradox is said to occur when technological improvements that increase the efficiency of a resource's use lead to a rise, rather than a fall, in total consumption of that resource. Greater efficiency reduces the amount of the resource needed per application, lowering its effective cost; if demand is sufficiently price elastic, this induces demand, frequently resulting in a net increase of total resource consumption."
It's like you're so close to getting it. Xi is serious about "corruption" alright. It's just that in an authoritarian state "corruption" is really just used as a euphemism for illoyality.
Yes and that's also exactly how Xi deals with illoyal people in China - just accuse them of "corruption". It's the authoritarian playbook.
Has Garry produced any software that is actually useful to anybody? Cause that kinda needs to be be lowest we can set the bar, right?
But they are!
They are saying what is generally safe and what is not. Telling people the dry statistics won't help most people. Remember half of all people are below average in IQ. You need to paint with broad strokes if you want to reach most people.
Please show me the data. If embalmers all over the world are seeing this, then there surely should be some data, right?
No one is "killing" anyone. That is just a gross mischaracterization. Sometimes (like this) we just choose the least bad solution, that's it.
Not 100% sure but I think it was this: Never inject into a vein. Always inject into a muscle.
Making better software than my competitors and keeping my business afloat is my goal. I use whatever better tool gives me the edge.
I focus on doing better work for the most part. But there is also stuff I do now, because the SOTA models can handle it, that I would otherwise just wouldn't do.
I'm spending 100$ on Claude and 100$ on Codex and 20$ on Cursor monthly. I use all to keep myself model and provider independent.
I'm not focusing on quantity of output but rather on quality of output.
Artists that need the best quality output use the best pens and papers. Call me a coding artist then haha. But seriously I don't have the time for anything less than SOTA.
If I am working on something simple and want the speed boost then I'll drop the thinking to low or minimal and still get the SOTA model output quality.
But for what I work on I mostly need high or xhigh SOTA model quality output. I don't have the time to deal with anything less.
And quality if you know what you're doing.
I'm making a micro SaaS product. Code quality and code production speed are actually both super important. I don't have the time for non-SOTA model output.
I use whatever model is SOTA. I switch between them in order to avoid lock in.
Yeah I have both the Claude and Codex 100 dollar subscriptions and I try to use both. I also keep the 20 dollar Cursor subscription as there I can play around with everything. I also refuse to use any harness specific features. Claude is particularly annoying with this in that it's the only one that doesn't respect open config standards like .agents/skills
Exactly. I simply don't have the time to deal with non-SOTA model output.
So I keep hearing.