dgb23

Habit + planning + preparing food in bulk can save a lot of time and money. If you're already in the kitchen preparing a Sunday evening meal, you can scale that up a bit and have a few extra lunches ready without much additional effort.

One of the best techniques for using LLM coding is to let it write programs that frontload and automate work in a repeatable and testable manner rather than letting it only do free form turns.

Letting it generate machine code directly would be the opposite of that.

Didn't set it up myself so my info will be meagre, but I think it's litellm and a big enough nvidia card. We use it for development but also for internal applications (where smaller models might escalate a task to it when needed).

Obviously not feasible for (just) personal use, the model is too large for consumer hardware.

Phasing out my usage of cloud based LLMs/harnesses.

Focus is pi + self hosted GLM Flash.

Pi is lightweight and easily extensible and creates minimal extra noise. the output I get is very close to what I want to see.

my workflow is focused on clear turns between planning, testing and implementation. Any useful/recurring tasks, like getting/filtering logs, checking statuses etc get regularly folded into a project cli and agent skills that are local to the project.

I spend time interactively iterating on the plan/design and acceptance tests, then let it do its thing. Usually I have multiple interactive sessions at once that I'm steering this way. 1-2 assigned projects, 1-3 side projects (additional automation, experiments/POCs, workflow/process efficiency etc.). I generally don't do sub-agents. When I need throughput I automate first, frontloading the "deterministic" parts.

The gap in capabilities is mostly quantitative and not qualitative.

That's one (Marxist) definition of it, but the movement is much broader and more varied than that.

When competition has no rules it resorts to people banging each other over their heads with clubs.

Disheartening. macOS seems to get less and less support in a way. For example some of the Blizzard remakes don't run on macOS but the originals do.

"Buy more RAM" usually isn't the thing I want to hear when my linter is slow.

There is a qualitative difference between unsafe Rust and Zig as far as I know.

I don't think you have specified that "," is used to pick up stuff in the bar below the log.

Climate change is too soft of a term. Maybe that's why it doesn't interest people who like to declare war on things.

The targeted term must be something that is clearly human made, something that sounds undeniably bad and something that is easily understood by everyone at first glance:

_War on Pollution_

Nature is good, pollution is bad. People who pollute are _obviously bad_ and they do bad things. Pollution is wasteful and ugly. Yuck!

Also it's more general than climate change. Ocean plastic is also bad. Chemical, electronic and light pollution etc.

The people who think of chemtrails and 5g waves. They really hate pollution so much, they see it everywhere. Give them a war that they can join in.

These corporations exist and do work. Worker owned companies have their own challenges and their own advantages.

For example they tend to be more stable during crisis, because workers tend to vote for lowering salaries/benefits temporarily rather than doing layoffs. So they retain talent better. But they also tend to have difficulty to grow quickly, for obvious reasons.

Besides full on coops, there are also plenty of examples that are hybrids (partially worker owned).

> they would get their face eaten by other more efficient and ruthless corporations

You're possibly of assuming that a company needs to have an adversarial relationship to their workers in order to be competitive. I don't think that's generally true. This approach has advantages in specific situations, but disadvantages in others.

I think you make a point that is worthy of discussion, but the first sentence is unnecessarily hostile. The comment you responded to already made a caveat that they might be too cynical.

I‘m surpised at the presented dichotomy between JSON formatting and what the Apple SDK provides to parse output into structs.

Based on what I understand about how the former works, I would assume that the latter has the same properties and failure modes.

Local could also mean self hosted.

The obvious optimization for the case presented would be to generate all the summaries on a server instead of in the client. Then the totally used compute would scale with the number of articles instead of number of users.

Aside, but I struggled a long time with regular sleep. I have been a night owl since I was a kid. I experience late hours as magical, don’t know how to describe it. So I always slept too little, then not at all, then drifting and sleeping in.

But I somehow managed to have a regular schedule and now I start to sleep at 00:00-01:00 very often, sometimes even earlier.

No idea how I managed to do that. I guess I just did improve many small things, like getting rid of bad habits, being more content, appreciating sleep more, prioritizing things differently.

I wish everyone good, healthy sleep.

The insight for me is that the assumptions of system need to be stated, not just the intent.

I tried to make a joke about the tensions of security and accountability.

Not even vibe coders, but autonomous agents/bots.

I‘ve noticed that some projects have „Claude“ as one of their top three contributors.

I glanced at zhe thread you linked. And as I understand they are in the process of migrating, which will take more than a year still.

If that’s the case, then it’s not necessarily a problem with Azure itself.

But by not having a checklist you avoid that your blind spots get exposed.

It seems to me that this sort of work is a usecase that’s actually very fitting use case for LLM agents and the like. Because they can be trained and tuned to find commonly known vulnerability patterns.

Here, something that looks like the thing is a strong signal, as long as the probability is high enough to be useful.

Remember Netflix‘ chaos monkey?

I tried to make sense of middle eastern politics once. My conclusion has been „It’s complicated.“

Apparently movement plus focus on the distance has a calming, clearing effect on the mind.

Do you have a similar experience when walking or running (deliberately)?

Rumi (the Sufi mystic) apparently walked and turned in circles in order to contemplate. The tradition merges music and movement with philosophy and religious mysticism.

Walking, dancing or manual labor (for example gardening or cleaning) can all be done in a meditative way.

But these are likely different types of meditation that have different effects. Even just a calm, sitting meditations might be vastly different from another, depending on the meditation object.

Of course there are people who lean into specific types over the others as you describe, but I think many of these activities share a common core and experience.

LLMs have a lot of issues with facts, because they are probabilistic and you typically only get one answer per query instead of multiple covering a larger space.

However, they are still useful in these cases if you know the above and use their output as a starting point to think and ask questions.

Or they pack up to leave.

I think you must be right to _some_ degree. The article illustrates that this org doesn’t know why they are doing certain things.

But there‘s something psychologically powerful happening with the interaction of AI. I think we overestimate our ability to be rational and underestimate how essily influenced we are.

> The point of a code review is not simply for good code to make it into a codebase, but to build institutional knowledge as people debate and iterate and compromise, slow as it may be.

I feel like this is a very profound insight.

Of course processes like this can become about the immediate utility. Reviewing is then checking work so, it can be merged and used.

But the process is more about us than the code. And we lose the deeper part when we only care about the superficial one.