unphased

great approach. I did that with my opencode based setup as well, it's neat and fun to tune skills and mcp loaders and stuff. Then i got fed up with opencode's design limitations. And then, my own harness work is on hold in favor of a harness-puppeteer paradigm, but that one has also been on hold! I'm mostly currently pulling on the thread of making it easier just to review the voluminous conversation turns!

So often we will burn 20% of limit in a single ill conceived agent tool call that we're simply not going to be able to or want to be able to intercept. Where I see a tool like this being a real step forward is to add a decision point. it does not have to bubble up to hard-require user to provide permission, but it can let the LLM have an intermediate checkpoint to say that it's about to get blasted with 30k tokens and here is roughly the shape of it and do you wanna adjust or whittle it down if you know what you're looking for etc.?

There is definitely tons of value to extract from this line of thinking.

haha, that analogy says more about whoever wrote it than it ever could to get the intended point across!

I think you might be vastly overcomplicating it because I didn't think there had to be any sort of "conservation of branching" in the MWI. each nondeterministic event (of which unfathomable quantities take place every moment) generates an infinite number of branches so to even conceive of the total geometry of all the branching (e.g. all that could ever take place, truly) is a bit of a mindfuck, and that's probably okay and the way it was intended. It's supposed to be comforting to know that regardless of how bad reality seems, if we could navigate arbitrarily through the branching space/time/universes then there would be unimaginable infinities of joyful utopias to visit.

could you elaborate slightly on what is meant by ancestor simulation? My best stab is that you're saying we're the unknowing entities that they created for fun to get to meet or observe their own ancestors? This still seems far fetched.

What is it about understanding the universe that makes it such an axiomatic global objective? Sure for many of us myself included it's as all pervasive as the air we breathe... But sometimes I do wonder if it is actually all that correlated with my well-being.

Security is one area in which you can't really afford not to go down the tin foil hat rabbit hole. It's been like this for some time, where it's not really possible to know just how much capability is possessed by nation state actors, but with AI and other tools those capabilities will expand and become the baseline probably way sooner than we predict. No longer will experts be needed to do careful tracing to deanonymize, one of the true superpowers of AI will be its ability to extract the full power out of the big data we've been collecting. So the cost then of figuring something out that now might cost millions will soon cost a couple bucks.

It's relatively fast (probably could be faster than NVMe if optimized, i did not rtfm) and it is available. it's not the most horrible of ideas. I am fairly fairly certain lots of applications will load assets into vram and then evict them from system memory. Why not leverage this for general purpose use if it makes sense (after taking into account the security implications of course)?

This article started off really clear and straightforward, then it dove into stuff that I clearly wasn't going to be able to follow, and upon skipping it I arrived at the explanation which was was plainly obvious from common sense.

We "bought" a Samsung fridge last black friday. Something extraordinary happened. Their system failed to deliver us the fridge. We had to reschedule the delivery since we were staying at the parents' for Thanksgiving a few extra days. It just never got delivered. It seemed like it was XPO's screwup, but it was hilarious to me that they could not get their system working well enough to even get the fridge to show up at our house.

Screw Samsung.

Great post.

Now you got me thinking about one thing in particular.

Suppose we will have quite soon portable and discreet technology that can, through monitoring your BAC, or what have you, warn you that you're about to be overly drunk. It would stand to reason that for many folks out there who have An Alcohol Problem that technology like this could materially enhance their lives. Now that I think of this, something like this could even become the first practical widespread form of cybernetic enhancement, not that we don't already have plenty of awesome ones in the form of prostheses.

Just makes me wonder what kind of hurdles a company developing such a product would face.

I think it could actually be a pretty good strategy to optimize your job search for the degree to which you might be able to go rogue and get away with it.

Mine's installed from homebrew and is a native process. 0.10.0-dev (1df7dc51) it reports.

The O would be capitalized if that were so. ;)

Pretty sure it does do this.

The ethical dilemmas we face now will pale in comparison to those of tomorrow.

This explains well why it's so satisfying to delete code.

And to add on to that, you're lucky if your computer just tells you no. Often you get a yes and something blows up much later with precious little for you to follow and you've got to put your Sherlock hat on.

> Maybe it's easier to create an endlessly gratifying simulation than it is to build a starship.

You say this with a maybe as though it isn't already a certainty.

Nothing can stop it from happening. There will always be brainpower available, willing, and capable of contributing to computer aided fantasy and isolation. Not that it has to necessarily be a bad thing, mind.

OTOH plenty of things could make it impossible to succeed with a starship.

We'll know for sure next week (or already just not me), but thunderbolt compliance means displayport compliance which means there should be no concerns along these lines. I'd be very pissed if I would be limited to 60 on these new machines.

It's funny that you did the German capitalization of all nouns in that broken English quote! It does add a delightful German feel to the quote.

The key is to have sophisticated software to "filter" down the content into reasonable geometry (rather than the grossly inefficient geometry you'd get out of typical photogrammetry) and then into LOD levels and very optimized streaming representations for that geometry. Anyway this is why unreal engine 5 comes together with current-generation console runtime SSD streaming, it's to facilitate this sort of streaming large quantities of geometry for virtual environments.

Obviously your result is only gonna be as good as your source imagery, but a large part of the magic is in how precisely the photogrammetry is able to solve and compute correct camera poses so as not to inject artifacts, combined with the automated generation of efficient meshes.

It seems like traditional asset pipelines never really had a need for crazy optimized LOD generation. There have been sophisticated mesh optimization tools, but usually it's only a small fraction of the manual effort to make some LODs compared to modeling the actual full detail model. It's usually done manually, but I don't know for sure, never worked in the industry. But you can imagine how this is a non-starter for when your full detail model came out of photogrammetry. I haven't been following it that closely but my take on it is that Unreal Engine 5 is a paradigm shift in that it will liberate you from even needing to consider levels of detail, much the same as you don't consider releasing memory in a garbage collected programming language. This is clearly possible once those concerns you state, "input is an absolute mess", "having to recreate models", become automated and reaches a level of quality that equals what a human can do.

I would expect in unreal engine 5 you will be able to import assets in traditional 3D formats (in addition, of course, to a bundle of photos) and it will recommend you provide it with the highest possible quality input. And you'll just be able to straight away instantiate hundreds of thousands of copies of it into your scene stretching into infinity and your framerate will be fine.

Thanks for sharing a piece of how it felt. I'm curious what the circumstances around that were, if you don't mind.

My comment really doesn't even do your comment justice, but... It really made me think about that scene from the Matrix where Agent Smith described humans as a cancer. There are just so many possible perspectives on it all.

Anyone work out the key bind for undo?

i guess it looks like a sphere but it's hyperbolic space and the opposite "direction" of curving compared to a sphere. But I dig what you're saying, like.. maybe allow us to drag our center of projection around so we can get a sense for the space.

To pile onto your point, webpack is clearly on its way out, it's all about snowpack now (afaik). I never got a chance to get into learning webpack, and I'm glad now. Throws hands up

I think this is reasonable unless SpaceX runs into a wall at some point. Settle has a fairly wide range of interpretation, but yeah it may take longer than this before a "permanent" settlement can be established.

I find AGI in 60 years to be more controversial than Mars. Not enough evidence yet. One can def hope though.

Pray tell, what's "st"? It's not googlable.

Hi. Sorry, this is a total non sequitur, but I was trawling my comment history and saw your reply to my comment. HN doesn't let me respond to a 60 day old comment, so... here I am, I found you somewhere else! You mentioned "infinte pan zoom graph based window manager" as part of your response. Thanks, I think you totally got what I meant.

I agree that it's essentially an OS scale problem to solve and I'd be chipping away at it endlessly (and that's if I have my way!). But I do believe the AR revolution is coming and it will wash over us like a tsunami and make the recent smartphone revolution look like a fad. I think we don't already have this kind of software yet because endless 3D graph-connected space is potentially clunky to explore on tablet/display.

It's already a foregone conclusion to me that this barrier will be obliterated by XR and it will be deeply intuitive and liberating to move beyond 2D interfaces. We actually must move beyond 2D interfaces in order to even reasonably support visual cues of dynamic relations. Only when your structures are represented as nodes in a shared context can you allow users to manipulate relationships between arbitrary things.