abound

I'd also be curious if anyone's working on alternative base board with a slightly beefier MCU inside. The 2.5 MB RAM is likely too small for functional navigation capabilities (the site doesn't appear to advertise any), but it doesn't need too much more RAM to get there.

For OP, when I've run into this in the past, it's usually some CI/CD pipeline missing a `lfs: true` or `git lfs pull` invocation.

To be clear, those are CVEs in the tooling, not in the generated static sites. Not great, but very different from this WordPress CVE

There are two (or more) concepts named after the same person:

- Pareto efficiency/Pareto curves: Basically the convex hull of points along the edge of a graph, indicating the best tradeoff between the axes. This is what the post is talking about.

- Pareto principle: this is the 80/20 rule you're talking about

This is sometimes called the "Toupee Fallacy": https://en.wiktionary.org/wiki/toupee_fallacy

Not OP, but I think the leading and trailing paragraphs were mostly human-written (and nice to read), but the memory leak example cases had a very different flavor of prose and code comments that smelled very Claude-y to me

Oh wow, good to know! Cool project, best of luck with your motor controller firmware

This is an unbelievably impressive project and a killer write up, very satisfying. I wonder if the author looked into using Slint [1] for the UI, given the codegen bloat issues they had with buoyant (which also looks very cool). I haven't used either, but I've done a few projects with Embassy and am always looking for an excuse to try embedded GUI tools

[1] https://slint.rs/

> Here’s the alphabet, in encoding order:

> ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789!#$%&'()+,-;=@[]^_`{}~

Yeah I definitely don't want any of those last ~22 characters in file names if I need to do anything with them from a shell

"Than what it's worth" is basically a shorthand for "if the cost of repairs is more than what it'd cost to buy a less problematic car, it probably makes more sense to just buy a less problematic car with that money"

I drive an 04 Civic I got from a coworker for $2000 seven years ago. One door handle is missing, the trunk leaks a bit when it rains, some of the ceiling fabric is starting to droop, the clear coat is mostly peeled off, first gear requires some finesse, and if you roll down one of the windows too far, it's going to take two people and 45 minutes to get it back up.

But it drives, requires only the usual regular maintenance, and isn't tracking me. I'm going to drive it until it turns to dust beneath me (or more likely, like the article says, when repairs get more costly than what it's worth).

A Rust implementation of this idea is Crux: https://redbadger.github.io/crux/

It might not be "foom!", it might just be like...all the computers and networking infra in the world go dark over the course of a few minutes. Could really look like anything, part of the issue is that we haven't the slightest idea what "misalignment" looks like for a superintelligent system.

I think you've picked the most cynical interpretation for each one. More charitably, you could say:

- Ideate on things you can build for your fish tank

- Iterate quickly on ideas for a tattoo

- Get inspiration for a new metalworking (or 3D printing?) project

- Get a sense for what a new haircut might look like, before you commit

- Iterate on arrangements for a bouquet of flowers

- Do silly things with childhood photos

- Do silly things with pet photos

- Clean up low-res, damaged, or weirdly formatted photos

- Do some creative storytelling with your long-distance friends

- Okay yeah fine they should just make the damn bed, but I can imagine small tweaks being useful for staging a home (easy to abuse though)

Yeah I was interested for the first few paragraphs, then all of a sudden I get hit with two "genuinely"s and a

> That’s the third property, and it’s the one that decides this.

and I gave up at that point.

I think this is more of a nitpick than anything. SourceHut is a code forge, code forges are where you store your code (and tickets). The policy going forward is a ban on using LLMs to write code (and tickets). The use cases listed there (reviewing work, asking questions, etc) are minority use cases and mostly would be unenforceable even if included in the policy.

I recently migrated from Google Voice to jmp.chat, which is a telephony/SMS/etc bridge to XMPP. I use Dino on my laptop, and Cheogram on my phone, and the experience has been quite nice and seamless. Next step is to host my own Prosody server and import my whole Google Voice history.

To put a name on what a lot of folks in this thread are describing (I didn't see the term mentioned in the article): Negative Visualization [1].

Basically imagining living life without a given luxury (hot water, infinite music on tap, functioning eyesight, etc), really internalizing that feeling, and then practicing gratitude for the fact that you do indeed have that thing.

[1] https://en.wikipedia.org/wiki/Negative_visualization

This is the exact opposite of what the parent comments are talking about.

I don't disagree with anything you said here, but I find auto-mode is too permissive on the reading side of things.

I used Claude this morning to convert the financial books for my nonprofit from some export of a proprietary software (Aplos) to plaintext accounting (Beancount), and was surprised to find it had found my personal Beancount ledger elsewhere on my hard drive and modelled its plan on that. That was correct and helpful, but a bit of an overreach in my opinion, enough that I'll be sandboxing agents (with tools like nono, smolmachines, etc) going forward.

My personal benchmark is a directory containing a bunch of research papers on the physics of popping popcorn kernels, and a prompt about creating high fidelity, photo realistic 3D models of all the different kinds of popped kernels. Fable (surprisingly? unsurprisingly?) refused to do it last time I tried, and the results from other models are, well, fine, but there's still plenty of headroom on this particular one.

Arson may be the original source, as may campfires, downed power lines, sparks from powered tools, etc, but the fact that these fires grow much larger than in the past is because of climate change.

If it wasn't humans, it'd be lightning strikes and other natural phenomena.

One reason to switch models mid-session that I didn't see explored in this post: cost. As we get ever-more-powerful models, it's going to make sense to use them for some exploratory/research/design phase, and then smaller models for other bits, within a single session. It'll likely even be managed automagically by another model that is just trained to do routing.

As people (and especially companies) adopt these tools for more and more usecases, there's just too much money being wasted on using an unnecessarily large model for a given job.

Not yet, but I'll see if I can publish the basics for it this week.

In this context, "top 90%" means "bottom 10%". I did truly awful and got "top 99%", which sounded impressive until I realized what happened.

Nope! Every source I care about either has an RSS feed, or I've been able to make a custom recipe [1] to handle things like auth for paid sites and whatnot.

[1] https://manual.calibre-ebook.com/glossary.html#term-recipe

I've been building something similar for my own purposes. Basically, I use Calibre's "Fetch News" functionality to get epubs of all the news sources I trust, and then once a month (interval is configurable), I run a pipeline over all of it (some LLM magic, partly local partly API) to dedupe and rank stories, then compile it into an actual honest-to-goodness newspaper (using Typst for layout and stuff), then read that at my leisure.

> I asked it “do you think the model will generate a meaningful number there?”

This also isn't likely to give you good results, models don't have a good sense of their own capabilities.

This is also in the same camp of asking a model why it did something. It'll generate something plausible and likely totally incorrect.

A more charitable take is that we shouldn't tile the universe with consciousnesses that are in constant, unending agony. Torment Nexus and all that.

> I have the agent write a script that is aimed to be idempotent and then run the script.

You can take this one step further and have the agent write Terraform configs [1]. I did this (including having the agent import all the initial resources from the live device), works great and is generally more robust than a script.

[1] https://github.com/terraform-routeros/terraform-provider-rou...