ElFitz

I fail to see why I should care.

But maybe that’s just me?

Same with shoes if you walk or run significant distances multiple times a week. Different models, different brands, different types of soles.

It both gives time for the soles to recover, lessening the permanent compaction over time, and helps reduce motion repetitiveness, which all your joints will be thankful for.

Alright. An industrial robotic arm. A Boeing 737 MAX's MCAS.

I think my dog has spent all weekend refactoring its own framework to integrate Jev.

The article mentions bomb-equipped drones with image recognition, where before launch operators define a set of target types, a desired confidence level, a "kill box" (geographic area in which the drone may engage autonomously), and set the drones loose.

Of note, the drones can’t distinguish friend or foe, nor civilians from combatants.

How about "Pockmark!... Freshwater swabs!... Bully!.." or "Amoeba! Bashi-bazouks! Chowderheads! Certified Diplodocuses! Nyctalop! Ectoplasm!"?

But everyone has to be in the office. For, you know, collaboration. And face-to-face interactions.

https://github.com/lightless-labs/pessimal: I got bitten a few times too many by my headless VM running out of disk space, so I had Claude build a minimal OTel agent sending metrics to an OTel collector (Signoz for now), and an iOS / macOS app to query it at regular intervals.

Works nicely with my existing Signoz, costs nothing extra, but I’ve yet to test it against Honeycomb and Clickstack. Apps are in private TestFlight for now.

https://github.com/Lightless-Labs/willikins: a tool to deterministically provision new projects’ git repository, secrets vault / config, build pipelines, etc, based on the project’s needs (cli tool, web service, mobile app,…).

It’s getting tiring to wire all those by hand or with cobbled together script. And I don’t trust agents to give them read and write access to all the resources they’d need to do it. Very much in progress, but it already forced me to clean up some config. I guess it’s like a bad Ansible, for project set up?

I’m also experimenting with a team / project / task-scoped blackboard (https://martinfowler.com/articles/exploring-gen-ai/an-accide...) tool for agents, to be used by multiple users’ agents (eg different devs working on different sides of the same fix or feature), and a self-hosted Q&A forum for agents intended to serve as an authoritative self-grooming source of truth on technical, domain, and business matters for a team or org’s agents.

And there I was, honestly thinking Terry Pratchett had come up with it all on his own.

Thanks!

There was a line in Discworld about some people in Ankh Morpork taking to breeding rats to collect the rat catching fee.

> Copyright is serving exactly nobody today.

As a budding wannabe fiction author, I certainly hope you are wrong.

Maybe I'm just 2026's Dropbox guy, but... yes, and?

Maybe, I wouldn’t know. I just find it fascinating that other people have such a radically different way of experiencing thought. No idea what it’s like.

Some humans apparently only or almost only do non-verbal thinking. It’s called anendophasia.

Thanks!

You seem remarkably confident about vast, complex topics.

I’ve been to quite a few deserts that get the occasional seasonal torrential rain.

They still remain deserts nine to ten months out of twelve.

The water saturates the soil, can’t percolate to underground aquifers where it could otherwise be stored, and runs into (ephemeral) rivers, washing away some of whatever is left of the soil with it.

I wouldn’t be surprised if two places could have roughly the same yearly amount of precipitation yet look completely different depending on how the rainfalls are distributed over time.

I have that in CC when I resume a session.

Suggests either starting from a summary, a new conversation, or keep going.

Or words, and language, just provide a sort of sensory input, which provided an environment and a selective pressure can shape an organism or neural network into predicting and producing the kind of reaction or output required for its continued existence.

And no, I’m not implying any sort of survival instinct in LLMs. It’s just that we don’t keep around the weights that produce gibberish.

No, not everyone. A few simply don’t have an inner monologue.

Many only know that way of thinking, and can’t begin to imagine what it would be like to think, or be, without one.

Plus all the people somewhere between those two, whether or not they are themselves aware of it.

Makes sense. Probably how one would approach it during a system design interview.

> How many times do you buy things during a day? Once? twice? Maybe thrice?

None most days. But then some days eight or nine subscriptions go through. Guess it averages out.

Might be spiky though, around the beginning and ends of each month. If paying out wages monthly, and at the end of the month or beginning of the next one is prevalent across countries where a significant share of the population has both access to a bank account and sufficient enough income to regularly spend either by card or online.

Did this turn into a systems design interview?

Anyway, joking aside, I was mostly amused by what my frame of reference for throughput / service load has become.

And yet anendophasia is a thing. Some people think without an inner monologue, without words[^0]. And they seem to do so quite well and quite successfully.

[0]: https://johnrmiles.com/life-without-an-inner-monologue/

And even the ones who trained it can’t afford to offer it at a price covering their training capex, since any good open-source model will see other providers selling inference without any training expenses to cover.

Didn’t see it, but it’s obvious in retrospect. Thanks!

Ah! Makes sense. Thanks!

> At a payments company, 500 payments per second is considered high throughput and 1000 payments per second is a huge amount of volume.

I realise I live in a strange world when my first thought reading this is "Worldwide? That’s low. Fits in the basic one region, one account concurrent AWS Lambda quota. No need to even ask support for the 10k."

> and providers don't have to cover training capex.

How so?

I guess the only option nowadays is to buy an old used car.

Soon some of us will be back to using typewriters.

What makes it even more ridiculous is that one could think grown fish are a fine solution over-fishing.

But it turns out, many fish farms use caught fish for (part of) the fishmeal and fish oil used to feed the fish they grow. Because many fish we eat turn out to be pescetarian.

Meanwhile, ChatGPT correctly diagnosed what was wrong with my plant from a single photo, identified which leaves I should cut, and annotated the picture showing where to cut and what not to touch.

I honestly expected a made-up useless generated image that matched the idea but not the actual thing.

Guess I’m still living in 2024.