okwhateverdude

I run the docker container locally. As far as I can tell, it doesn't call home or anything (from reading the source and from opensnitch). It is just a cgo wrapped llama.cpp that provides an HTTP API. It CAN fetch models from their library, but you can just as easily load in your own GGUF formatted llama models. They implement a docker-like layers mechanism for model configuration that is pretty useful.

But how much profit? And how much of that revenue was spent on spez's comp?

My trusty two-wheeled steed manages to take me home drunk after the bar just fine

Or Uber and Lyft could simply make less profit so they could abide by local labor laws. The labor law didn't remove the employment opportunity. Greed and shitty companies that don't want to abide by the law removed the employment opportunity. Don't shift blame for the scofflaws.

Schwartz basically rode his early fame into the ground. While the world was moving on from Perl (especially after the community completely fractured and started disintegrating during the early days of Perl6, now Raku), he doubled down, but also didn't really produce much or engage with community. The Perl work around the world was drying up. He ended up begging online instead of pivoting. Glad to know he finally did pivot.

>There are many ways to have OOP? Moose, Mouse, Moo, MooseX::Declare, Class::Accessor — you are pretty much guaranteed to pull in them all and have them coexist in runtime.

I think this is hyperbole. Generally, you won't be mixing Class::Accessor and Mo* based dependencies unless you are dealing with a shitty, barely maintained legacy codebase.

There was a progression to things, mostly because of performance reasons why there is a bit of a proliferation. Moose (and the underlying Class::MOP) were about providing a meta-object protocol for Perl which didn't exist before it. As such, it was focused on being correct over fast. A whole community sprang up from stevan's art. Then sartak's Mouse came along and made some performance improvements with some compatibility caveats. Finally, mst's Moo stripped away the "reflection" aspects of Moose to really squeak out even more performance.

MooseX::Declare was always a (beautiful) research project. I spent some time in that space welding POE and MooseX::Declare together. But these weren't suitable for any kind of production code. If you see this live somewhere, you should definitely rip it out lol.

As for TIMTOWTDI, this is sorta why project Andy Lester's Phalanx came about (it was trying to bless a chunk of modules as a sort of smoke test for Ponie (perl5 implementation on the Parrot VM (which was the first step toward a Perl 6, now Raku)). It was trying to provide an opinionated set of high quality libs for SWE.

Unfortunately, working in Perl really requires a bunch of skill and discipline if you want a long-term maintainable project. This means really looking at your dependency tree and making difficult decisions on which libs to use specifically to avoid installing all of CPAN. npm's installing the world behavior was foretold by Perl.

This signature is hilarious. It is the moral equivalent of "I'm rubber, you're glue..."

> The advanced version: it's not necessary to let them hit the ground. Stand over a bed or couch, so to pick them up, at least you only have to bend down to waist level.

lol, this is exactly what I did when I started because of the annoyance of bending down. Then I incorporated it into a fitness goal and did proper squats to pick them up. Now, I mostly will just pause and reset if I have a weird catch to avoid the next throw that would end up on the floor.

> My favorite version: juggle while standing in a swimming pool! Use an object that floats, like wiffle balls, and a missed catch just stays there floating right in front of you.

This is a great idea. Next time I go swimming I'll have to bring some balls that float.

I picked up juggling after doing a management training class that included a one hour lesson with a professional juggler. I had never tried before then and couldn't quite make it work within that one hour group lesson. But they gave us a set of balls to take home and over a weekend I got the juggling to click.

For me the trick was to stop thinking about it and to just let go. Our built-in parabolic trajectory prediction is magical, just need to throw the balls consistently enough for that prediction to become automatic.

I just fuck around with three balls and it is mostly to help me keep sanity if I am working somewhere with shit CI that takes forever. I was aware that there was a proper pedagogy and nomenclature, but never bothered to learn it. Really surprised to see my "repertoire" is all of the Level 2 and some Level 3 listed on this site.

> What does “tight and messy” mean here?

In music terms, playing tight means being exact. But I think in this context tight would be length of the song and how succinct it is. 70s rock could be power ballad-y with long solos. As for messy, in context, I'd argue it means the songs are less structured compared to 70s arena rock that are extremely formulaic (think of any AC/DC song and how it sounds just like every other AC/DC song).

> The same is valid for psychopaths and other bad people.

Just what we need. More soulless, avarice soaked, societal parasites.

Cool, then we're on the same page for other ecology management issues too, right? All of that "let's sell licenses for hunting other animals" we should simply stop, right? We should just let the world sort itself out since we've never had any success in managing populations of other animals either through protection or hunting, right?

Humans are not some unique, special animal. We're a very invasive, ecology modifying organism. We already have rules on where humans can and cannot live, where resource extraction can happen and how much. It may be uncomfortable to consider, but yes, we absolutely should manage our own population so we don't fuck up our one, and only, spaceship that's flying through a universe that is very hostile for us.

Metastatic growth is simply unsustainable. Running our civilization like a capitalistic business ignoring externalities and borrowing against the future that needs constant growth to sustain isn't a great idea. That will crash.

Might want to at least try to mitigate the prompt injection and definitely make sure none of the openai tools are enabled: https://i.imgur.com/DQCHMup.png

But also not that far away from that method, too.

> 2. Story Generator Agent - Write a story from a given outline (for educational purposes). Even at a very detailed outline level, and with a large context window, kept forgetting key points, repetitive language, no plot development. I just have to write the story myself.

You have to break it down into smaller steps and provide way more detail than you think you do in the context. I did an experiment in story generation where I had "authors" that would write only from the perspective of one of the characters that was also completely generated starting first from genre, name, character traits, etc. Then for a given scene, within a given plot and where in the story you are, randomly rotate between authors for each generation, appending it in memory, but not all of the story fits in context. And each generation is only a couple hundred tokens where you ask it to start/continue/end the story. The context contains all of this information in a simple key:value format. And essentially treat the LLM like a loom and spin the story out.

Usually what it produces isn't quite the best, but that's okay, because you can further refine the generation by using different system/user prompts explicitly for editing the content. I found that asking it to suggest one refinement and phrase it as a direct command, then feeding that command with the original generation, works. This meta-prompting tends to produce changes that subjectively improve the text according to whatever dimensions specified in the system prompt.

If you treat the composition as way more mechanical with tightly constrained generation, you get a much better, much more controlled result.

If you pull the llama.cpp repo and use their convert/quantize tools on the pytorch version of the models uploaded to huggingface, they will load just fine into ollama:

https://old.reddit.com/r/LocalLLaMA/comments/18av9aw/quick_s...

https://github.com/ggerganov/llama.cpp/discussions/2948

You can run ollama (and a web UI) pretty trivially via docker:

docker run -d --gpus=all -v /some/dir/for/ollama/data:/root/.ollama -p 11434:11434 --name ollama ollama/ollama:latest

docker run -d -p 3000:8080 --add-host=host.docker.internal:host-gateway --name ollama-webui ghcr.io/ollama-webui/ollama-webui:main

That particular webui will let you upload models (with configuration). Other wise, you can use the api directly (you'll need to POST a `blob` first):

https://github.com/ollama/ollama/blob/main/docs/api.md#creat...

Ollama folks just announced api compat with openai's stuff

https://ollama.ai/blog/openai-compatibility

So apparently, yeah.

More like an easy-mode llama.cpp that does a cgo wrapping of the lib (now; before they built patched llama.cpp runners and did IPC and managed child processes) and it does a few clever things to auto figure out layer splits (if you have meager GPU VRAM). The easy mode is that it will auto-load whatever model you'd like per request. They also implement docker-like layers for their representation of a model allowing you to overlay parameters of configuration and tag it. So far, it has been trivial to mix and match different models (or even the same models just with different parameters) for different tasks within the same application.

Seems like a skill issue, honestly. I feel like many of these arbitrary rules end up being made to account for low skill developers to curb their behaviors which seems like the wrong way to tackle that problem.

Man, I sincerely hope you had a jenny-throws-rocks-at-the-house moment later. Parents being shitty to their kids is the worst.

Are you really web scale if you aren't multi-cloud with accounts in GCP and Azure?

Dude wrote a CPU emulator with all the fixin's to emulate a particular ancient computer. That emulator is compiled then flashed on a microcontroller and then uses that emulator to run linux and ultrix which target this particular emulated environment. Also seemingly includes exposing a serial console via USB to allow interacting with the system. It is all a bit inception-y, but no doubt a master work.

The images of the PCB have silk screen explaining what happens when you plug it in.

Playing No Man's Sky in VR gives this vibe. Sitting is fine. The first star field motion might induce a bit of vertigo, especially if your machine struggles and you get frame drops and stutters as it is loading.

Man, I miss my N900. That slide out hardware keyboard made the difference. In situations where I was on-call, the ability to quickly triage production problems from my phone meant not needing to carry a laptop around.

I can't seem to make anything more complicated than "Divorce Hammer + Super Mothra Unicorn Batman". As in, that's the name of the item which was indeed the two components I attempted to combine. Trying to combine that item with anything else simply doesn't do anything.

Yes it is normal here, but it is also hilariously fraught with low-stakes peril. Dumpert always has vids of people busting their ass trying to make a turn or they hit an icy patch every year. I've had a couple of falls on ice here and I am not confident of my resilience long term doing that as I get older. So I avoid biking when it is frozen. But since that is only like a week or two out of the year, it is no big deal

I intuited the phrase in context to roughly mean that APIs are fair game for respectful contact by everyone including competitors. It wouldn't mean they need to provide a open api spec or make it easy, but it couldn't be a ToS violation or something forbidden to access.

Same. Pretty much any instance where he mentions prison groups or classes, it is very specific and emphasizes strongly that they have changed him. And he knows his mail will be read by the prison staff anyhow. The only benefit of the doubt I have here is maybe the groups/classes promote discussing topics like this precisely because they know a parole board gives them consideration. In which case, he'd be an idiot to not play along if he's angling for parole. (And if that is the case, it wouldn't surprise me if the prison system is being duplicitous in telling prisoners that so they can be demoralized when denied, ie. "I followed the rules and did what you tell me, but you still won't let me go?")

Basically, yeah. It is shockingly trivial to do, and yet like playing with alchemy when it comes to the prompting, especially if doing inference on the cheap like running smaller models. They can get distracted in your formatting, ordering, CAPITALIZATION, etc.

Presupposing completely optimal 50mW/m^2, could you origami that surface? Because then you could probably have something pretty useful.