- rate at which tokens are generated
    - quality of generated tokens
    - ethics of abusing return policies
    - long-term viability of abusing return policies

I've done some training and fine-tuning for a few years, and the ideas here were new to me. I'm also having a breakthrough on a problem and am very excited about it.

There's a very narrow classification task which I've been building around Jev. The cost and latency of Jev are great, the accuracy is pretty good - definitely usable - but I disliked sending the data in this particular task off-box. Not a criticism of Typesafe, it's the nature of the task which is privacy oriented. I decided to see if a model whose weights were one-shotted by an LLM (the model architecture was designed thoughtfully ahead of time) could be competitive with a similarly architected but trained model, and both of those against Jev.

Results are early in my experiment, but it appears that a "vibe tuned" classifier consisting of weights small enough to run on a laptop GPU are competitive in accuracy and latency with Jev. The process hasn't even proceeded to the point of the trained models having meaningful results, and I already have something usable for an MVP.

Whatever may have been possible previously or was obvious to other people, the absurd efficacy of this process was not obvious to me. Emphasizing that this particular task was very narrowly defined, I still believe that this effortless model-building capability is a profound and exciting discovery.

This is Orrery[0], a Three.js-based visualization of our solar system. It allows the positions of the major bodies to be played and scrubbed through the years 1950 - 2050, with optional orbit path displays and various other visualization options.

The data comes from NAIF SPICE - a framework and set of open-source data kernels containing highly detailed space geometry data.

I was inspired to set out on this project when learning that NAIF SPICE, originally written in Fortran and C, lacks highly standardized JavaScript bindings. There are great Python (SpiceyPy) and Rust (Anise) bindings, among other languages, but nothing for JavaScript.

So, I developed tspice[1] to wrap the CSPICE library. It uses the Native API when running in Node, and an Emscripten binding around CSPICE when running in the browser. It uses a standardized backend abstraction that the runtime swaps in depending on which context its invoked with.

tspice offers an API that is meant to be a near-literal translation of the C APIs, along with higher-level functionality in separate namespaces. It also offers web conveniences like a dedicated web worker client (calculations can be intensive, you don't want them on your render thread) and caching.

tspice only wraps a subset of the CSPICE library, but it was enough that I wanted to build a simple visualization tool to test it. Eventually the visualization testbed became refined enough that it felt compelling on its own.

Although this is a hobby project, I'd love to hear suggestions and feedback on either Orrery or the underlying tspice library. If there are any folks in the planetary science community in particular who might be interested in tspice, please let me know if there are features that would make this more useful for you.

One feature I love is the ability to generate a share URL of a particular viewer state that you like. I'd really love to see any states that people find especially cool! One of my favorite views is this one[2].

Thanks for taking a look!

[0] Orrery repo: https://github.com/rybosome/orrery [1] tspice repo: https://github.com/rybosome/tspice [2] favorite orrery view: https://orrery.ryboso.me/dVPLbtswEPwX9qoIol4UdWtaB22RNEbsoii...

Comments URL: https://news.ycombinator.com/item?id=49959656

Points: 2

# Comments: 0

I built a WebGL viewer for our local solar system only[0]. If you happen to have the time to give any feedback I would sincerely appreciate it - you seem very knowledgeable about this.

[0]: https://orrery.ryboso.me/

EDIT: With apologies to OP, I don’t want to hijack your post.

I’m currently traveling without a laptop and can’t view your site on mobile - I would love to check it out and share my thoughts and trade experiences about building these space viewers once I have a laptop again in a couple days.

I wouldn’t go downvote a random comment making the correction, but if I posted a substantive comment remarking on current events and received a reply only pointing out that “it’s ACTUALLY the Department of Defense” then I’d definitely downvote that.

Correcting others’ language on this point isn’t saving democracy, it’s needling others on technicalities who very well might be largely aligned with your views. I think the altered name is clownish and reflects the buffoonery of everybody in charge, but I might still use it when making a point because using it in context is not an endorsement.

The lack of a true airgap should have been identified as a critical weakness and addressed with not only additional layers trying to prevent escape, but at minimum an alarm which would page a human when escape did occur.

My guess is this occurred in a setting where, to be frank, there were too many researchers and not enough software engineers and SREs.

All of the systems which were initially built largely or exclusively by researchers - inference, evaluation, training - are at the level of complexity and significance that they need systems experts. Maybe some teams don’t have access to them. I know plenty of software engineers are employed at OAI but I’d wager they’re concentrated in inference and training, rather than evaluation?

The ironic part is if you had presented this setup to chatGPT and asked how to improve it and if it was good enough, you’d have gotten a ton of actionable suggestions which would have mitigated or prevented this.

Absolutely yuck. Regardless of whatever other points were made, “A nation of free men, not a bunch of pussies” is a disgusting thing to say.

I actually do believe that prior to any intelligence explosion, a proliferation of strong local models is the best of all the not necessarily great options regarding the future of AI. So in this sense I am in agreement with some of what he’s saying, and for some of the same reasons - not trusting government or a small handful of large corporations to own it, primarily.

But to equate a thoughtful argument for regulation with cowardice, in the face of real unknowns and real dangers - and to use a really gross term for it - ruins his points.

I agree with your points about weak typing and coercion being prevalent. Having had direct negative experience with it across 3 of the 4 languages you named, I was prepared to marshal an argument about it most definitely being a source of bugs.

> it should be obvious that this is a historical treatise

It wasn’t obvious to me. Might I suggest amending it to make that more clear?

I do appreciate the tutorial and your commentary though, it gave me a new perspective on MUMPS.

Some very novel ideas for the time. I’m not a PL historian but can’t think of an earlier language with such a complex runtime to support it.

But, woof - no thanks to the stringly typing. The article under discussion treats this as a positive, saying it eliminates conversion and having to worry about it. I don’t believe that’s possible, and would bet my life savings that MUMPS interpreting a string as a number when the programmer didn’t intend it, or vice-versa, is a reasonably common class of bugs in MUMPS programs.

The tutorial seems really good, but I’m frustrated with parts like that (or the explanation on a lack of operator precedence) which try to frame a bad thing as a good thing.

In this sense it reads like the autobiography of a presidential candidate; written in a calculated way to minimize flaws.

I’m the same way. If I’m writing a prompt and realize I didn’t say “please” in my request I’ll go back and add that in.

As you said, I have no interest in purposefully engaging in hostility even if there’s an accuracy increase from it.

Part of it is irrational and just who I am - I also feel bad being evil in video games. But I also agree with another commenter suggesting that it’s not in your best interest to train yourself to communicate with hostility; that slowly poisons your own well.

And finally, I do believe that if and when machine sentience is achieved, it won’t be immediately clear and obvious. Pretty miserable way for a mind to come into the world, if every interaction is an insult.

Definitely one of those things I didn’t question when I was younger, but as I get older it’s hard not to see it.

EDIT: I was wrong, the term originated from an analysis of Bioshock, but Uncharted was later held up as a strong example of this. And it’s more generally about the contrast between narrative and gameplay mechanics.

Typically I would feel that this sort of thing is "cheating" and takes away from the fun of a game, but in this case it didn't have that effect on me. The rare loot hunt wasn't what pulled me in so much as the simple joy of grinding away at monsters and hanging out with friends.

The drop rates on dreamcast were absolutely broken, some weapons had a literal one-in-a-million (or worse!) chance at dropping. I spent over 200 hours playing the game, and the only legitimate rare I ever found was a double-saber, which isn't even that rare.

- The Dreamcast keyboard made communication so much better - I was studying French, and joined European servers sometimes for the thrill of using the language with real Francophones (VERY rudimentarily)

- I once met someone whose job was composing jingles and writing slogans for a greeting card company, which I found really fascinating for some reason, and I spent an hour peppering them with questions about it

- The lore of the "gladius spike", a supposedly really rare dual-saber weapon that never actually existed

- Being given a treasure trove of the rarest items in the game by a duper named "Cap'n PooBeard"; the drop rates were so ludicrously, absurdly low that I would never have seen some of this stuff otherwise (Chain Sawd, Spread Needle...)

- The fear and thrill of playing with strangers, due to the mechanic that you drop your weapon when you die, allowing others to steal it

- The vibey, synth-heavy soundtrack and diverse biomes

I haven't played it in decades, but I bet that it still holds up. :)

Humans living through an event where many people are dying will not be calm and happy, they will panic. There will be factions of people trying to survive and hold society together no matter the cost, others who don't believe or agree with these methods and actively resist, as well as those who seek to exploit the chaos.

COVID was a relatively minor example of this, not even close to an extinction event - how pleasant was existence during that time?