jonfw

> If all companies are using the same "frontier" LLMs, and if they are competitive

The answer is that all companies will not be using the same frontier LLMs competitively

This will be yet another technology that will benefit from economies of scale. Big businesses and those in PE portfolios will lead the charge, adopt AI, increase efficiency, and gain market share at the expense of mom and pop shops.

This will be good for our 401ks because we're all primarily invested in big business

Investors can be very generous if they see opportunities for market capture of the next big thing

TL;DR: AI generated blogpost on using personal AI productivity tools to develop personal AI productivity tools

Mobile support?

It is not obvious to me that determinism is a requirement for an abstraction

wheels are so much simpler that it seems much easier and more cost efficient to solve the "transporting a wheeled robot up the stairs problem" than it does to go fully bipedal.

Just like math is learned by solving equations, software engineering is learned by writing code

Due to technological advances, solving equations stopped being a marketable skill, but understanding mathematics is as important as ever.

Software engineering will follow a similar route as math- the marketable skill will no longer be to write code, but writing code will be necessary to understand the big picture and build the marketable skills.

You are not stuck between public API pricing for frontier models via Claude and self hosted.

Remember that there are other LLM providers, open models, and previous gen models, that are way cheaper that frontier Claude and still way better than what can realistically run locally

One token of GPT seems to go farther than an opus token in my experience, those books are a little cooked

If you're interested in keyboard navigation of websites, consider a browser or extension with link hinting support! It worked really well in my experience a few years ago, although I've since became much more of a mouse guy and stopped using it.

Qutebrowser was my favorite browser for keyboard navigation but firefox, chrome, etc. have extensions for this as well.

We are discussing a gas tax, and there is a strong correlation between gas consumption and weight, which implies more taxes for trucks and SUVs

I think that this has the benefit of both raising a non trivial amount of money as well as reducing housing demand from the wealthy in a supply constrained market

I used to work with a guy who would always say "if you're looking for trouble, you are going to find it"

When I hear that "we found X bugs using some new tool", where the standard for bugs is low and doesn't neccessarily require user impact in realistic scenarios, I think to myself- duh! You went looking for bugs, of course you found them.

For a sufficiently complicated product, in my experience, you don't have to look far.

Humans can certainly be self improving, both on an individual basis and in aggregate.

In humans, it seems that improvement in a new domain seems to follow a logarithmic scale.

Why wouldn’t this be the same for an AI?

Easy to say that! Some problems are legitimately hard to solve though. Github is likely seeing usage patterns that have never been seen before and I bet some of these failure modes are novel

If you are at the limits of your architecture you may need to re-write things, and if you are rewriting things you can not arbitrarily speed that up by throwing dollars at it.

> Could be. But 99% of the repos are static garbage with no PR nor actions.

But the 1% of repos that do have PRs and actions are likely going to be seeing enormous increases in volumes

I have been a part of two very large companies with self hosted gits and I've seen enough to be confident that this is an incredibly hard thing to manage

> If I can use an open source highly effective LLM locally, and have it do all of the things ChatGPT can do

And if my grandmother had wheels, she'd be a bicycle!

We are miles off from open models having parity with what the AI labs are putting out.

And that's putting aside the compute requirements to run the top open models and the poor economics of running these models locally.

With a good CLI, an agent may be able to do something outside of the scope of it's skill fairly easily, by running help commands or similar. With even a well written API it is not as easy.

I suppose that curl + API docs could replace a CLI but that's really token inefficient

I dislike that the conversation about wealth is increasingly tied to comparisons with the wealth of others.

Just because I am not likely to join the class of billionaires, does not mean I can not become wealthy! I prefer to measure wealth in terms of quality of life rather than by comparison to others.

I do not have to join the capital class to produce a great life for my family and live in stability and happiness

I totally agree that republicans are irresponsible with the deficit. But americans don't seem motivated by the deficit, they seem motivated by inflation.

Inflation and the deficit don't have a 1:1 relationship. For the same dollar of debt, you'll see more inflation from social service spending than you will from tax cuts.

Generally speaking, people thought that government spending led to massive inflation, and the republicans have stronger rhetoric around cutting government spending.

Whatever your political affiliation and thoughts on the war, I hope we can all agree that it would an awful thing to base our foreign policy on the US election cycle.

I'm sure claude does great at that, but it would be objectively better, for a large variety of reasons, if claude didn't have to keep syntax examples in it's context.

If you are not terrified of your production terraform, you are doing it wrong!

Rather than saying that microsoft bashing has ramped up, I'd say that it is getting closer to it's standard levels.

Microsoft experience a sort of reputational resurgence in the tech world these past few years with some commitment to open source contribution and a really nice pivot towards linux and cloud.

Their pivot to AI is much less popular!

Yes, you are on the money. A cloud service provider needs to maintain reliability first and foremost, which means they won't have a runtime dependency on their billing system.

This means that billing happens asynchronously. You may use queues, you may do batching, etc. But you won't have a realtime view of the costs

> Decades ago companies would train new hires out out college, but that trend ended in the 90s.

Decades ago engineering salaries were a fraction of what they are today, developing countries did not have computing and educational infrastructure, and we had worse solutions to the logistics challenges from off-shoring.

It is increasingly difficult to justify the US salaries and I'm not sure that the talent pipeline in the US is so superior to make up for it.

As folks optimize for getting these high paying jobs it is increasingly difficult to find someone who has legitimate problem solving skills vs someone who has invested a lot of effort into looking hireable.

This is very much a tangent, and was asked in bad faith, but I’ll answer anyways!

One of the interesting things about working on distributed systems, is that you can reproduce problems without having to reproduce or mock a long stack trace

So I certainly don’t see the case you’re talking about where it takes hours to reproduce or understand a problem without a debugger. Of course there are still many times when a debugger should be consulted! There is always a right tool for a given job.

Keep cattle, not pets! The advice that used to apply for managing large numbers of machines also applies to managing coding agents.

If you rely on monitoring the behaviors of an individual coding agent to produce the output you want, you won't scale

“I have a PR from <feature-branch> into main. Please break it into chunks and dispatch a background agent to review each chunk for <review-criteria>, and then go through the chunks one at a time with me, pausing between each for my feedback”