pnathan

I've been working on https://arbiter.housecarl.cloud/ .

It's a deterministic judgement engine for disagreeing inputs. The input interface is shaped for LLMs now, but there's no particular reason it has to stay limited that way & it has a normal API.

I pulled in a lot of research on conflict logics to build it out.

The next big push technically will be an ingest pipeline and a more sophisticated logic engine.

> Suppose you have taste. Suppose you paid the full price and you can feel the sag in the sentence and the wrongness in the function. > > Congratulations! You now ship at exactly the same speed as the person who cannot.

Yes, previously the person who didn't understand shipped faster. It was a long standing problem. The people who crapped, got the bonus, and ran were much faster than the patient engineer of quality.

By the way, what were the essays that formulated Claude Style? if the author was indeed formed by them, I'd like to read them - or at least enough of them to understand.

Reflecting on the matter with a colleague -

Some people just want to _cook the steak_.

They don't want to attend or understand to cuts, thermometers, and searing.

(1) personally very annoying. I have been using fable to try to collect cutting edge math in one area and work on a hopefully new result with lean verification.

(2) I am really tired of the AI community trying to threaten everyone with grey goo and finding out the hype doesn't land comfy with others. It's a freaking text generator, not god in a pocket.

Shoot at high f numbers - as high as the lens will take without inducing diffraction.

I have found TTArtisans 50mm f0.95 to be quite nice.

This is good. I recently had to replace a generally working phone because the battery was dying and there was no cost effective & reliable means of replacing.

A proper gasket and screws needs to be the standard solution here.

Data is missing on this chart.

It's my experience that opus 4, and then, particularly, 4.5, in Claude code, are head and shoulders above the competition.

I wrote an agentic coder years ago and it yielded trash. (Tried to make it do then what kiro does today).

The models are better. Now, caveat - I don't use anything but opus for coding - Sonnet doesn't do the trick. My experience with Codex and Gemini is that their top models are as good as Sonnet for coding...

I am experimenting at a very early stage with using Verus in Rust to generate proveably correct Rust. I let the AI bang on the proof and trust the proof assistant to confirm it.

There is another route with Lean where Rust generates the Lean and there is proof done there but I haven't chased that down fully.

I think formal verification is a big win in the LLM era.

I'd have to see more to have a final thought.

As presented, Gollum is badly off, I reckon - missing the books textual description. The flowers are out of line.

The dragon scene is wonderful and captures the situation.

The dwarves are a bit dopy looking but I think could cohere with the early introduction in the Hobbit.

Great article.

I do have a deep fondness for SF billboards being building-stuff oriented. I don't care for consumerism.

The vapidity of the products created is remarkable, however.

I'm looking forward to when I can run a tolerably useful model locally. Next time I buy a desktop one of its core purposes will be to run models for 24/7 work.

a lot of AI assisted development goes into project management and system design.

I have been tolerably successful. However, I have almost 30 years of coding experience, and have the judgement on how big a component should be - when I push that myself _or_ with AI, things go hairy.

ymmv.

Claude code's code review is _sufficient_ imo.

still need HITL, but the human is shifted right and can do other things rather than grinding through fiddly details.

I don't remember all the places I got software... :)

> 30 years ago, it was safe to go to vendor XY page and download his latest version and it was more or less waterproof.

You _are_ joking, right? I distinctly remember all sorts of dubious freewarez sites with slightly modified installers. 1997-2000 era. And anti-virus was a thing in MS-DOS even.

I am working on a little project in my offhours, and asked a non-hacker (but competent programmer) friend to take a run at exploiting it. Great success: my project was successfully exploited.

The industrialization of exploit generation is here IMO.

I am _much_ more interested in i. building cool software for other things and ii. understanding underlying underlying models than building "better claude code".

Did the operators _want_ to poke the well connected & well funded bear with a historical anger problem?

This is it. Otherwise I'd know more about $that_technology than I would otherwise.

I - senior - can patch an application in an unknown language and framework with the AI. I know enough to tell it to stop the wildly stupid ideas.

But I don't learn. That's not what I'm trying to do- I'm trying to fix the bug. Hmm.

I'm pretty sure AI is going to lead us to a deskilling crash.

Food for thought.

Commercial law nuts and bolts is very algorithmic often.

Criminal law is often fundamentally subjective, incorporating questions of intent and remorse.

As a layoff justification and a hurryup tool, it is pretty loathesome. People use their jobs for their housing, food, etc.

AI the hype beast product and the club for workers is a plague that I frankly hate.

AI the manual algorithm to generate code and analyze images is quite an interesting underlying tech.

The question of leadership is much larger, more general, and more timeless than the last 15 years. I invite those curious about it to look into the American Army.

> Leadership is the process of influencing people by providing purpose, direction, and motivation to accomplish the mission and improve the organization.

taken from -

-- https://www.eiu.edu/armyrotc/docs/adp6_22.pdf

Jules is nifty. Weirdly heavy on the browser CPU.

There was an attitude at a University about 20 years ago when I was an undergrad, around, hmm, stochastic learning algorithms. And the attitude was, "we don't care why or how it works - we want to make the outcome happen".

I found it intellectually reprehensible then, and now.

The number of junior roles I have seen my companies open in the past 8 years - and I've been at a few different shops - is less than he fingers I have. And this is _before_ the AI-copolypse.

Management has generally become persuaded that juniors are not worth hiring. My current shop is a bit more thoughtful on this which is good. But. The desire for senior+ is out of line. Particularly when companies want to pay junior rates. =}

---

Dear author,

I don't know any period in the past 20 years where entry level jobs were properly allocated outside of FAANG. I have always advised talking to Microsoft, as my perspective is that they have the best entry level pipeline.

"The market can stay irrational longer than you can afford to stay in it" is an old stock traders proverb. And I believe it applies the AI fad. I do not believe that the dreams of actually replacing humans will work out generally. But it will be a painful experience for the employees and the prospective employees.

Yeah - I know in US law some terms are simply unenforceable and void. Much of the FOSS movement is designed around US contract law. There are issues with some US licenses being enforceable under other legal regimes - I was chatting a decade or so ago with a Russian who understood the...GPL(? I don't remember exactly) to be invalid in Russia and so it had to be bundled in some fashion to be usable.

Or to put another way, a license (a contract) is a tuple (terms, jurisdiction), and the juridical evaluation process will take both into account.

I would be curious if that is how Japanese courts would view it. They may not consider that a valid way. Or they might. But different jurisdictions vary.