zxexz

I went a little further than I did previously, thanks to it being 2026 and I can antagonize half a dozen frontier LLMs :) My original was bolted onto a Transformer-XL that was trained from scratch.

https://github.com/zxexz/kgmem

I don't have the original code but I can throw something together.

Seems like a bright young guy. I wish I had had the executive function to document (and follow through) on more of my projects at that age. Not saying I do now, aha.

EDIT: just noticed I was responding to the poster :)

I’ve done it by hardcoding a knowledge graph nn and adding a few “query” layers. You can freeze the graph and swap it out and update it during training so you can update the knowledge at will during inference. You just need to keep the format the same; the model learns an arbitrary graph query language. If you register a buffer it can update you can teach it to keep scratchpad memory very quickly.

> So, AI would need access to tens of millions of dollars of HW and Software in order to build and verify a chip design.

So, a seed round?

Live feedback without additional expensive and buggy test tools. May be marginal, but if you are trying to guarantee a certain income/payrate, it’s easy to take shortcuts in a system you’re comfortable in.

Yeah I’m really confused why France is in that list (also, less so, but China - they have come a great way, further than almost any western country save France, I’m just unsure I’d the metric here applies evenly).

I’ll stand by the only reason nuclear power is unprofitable (in most cases) is because we decide it is unprofitable.

This whole thing reminds me of DeepMind’s Variational Bayesian Last Layers[0], which never gained much traction in the broader “AI” world, but is a remarkably useful tool. And a relatively obvious one that anyone with experience in SVI and with transformer pretraining, seems to independently rediscover (including me) before finding this paper.

[0] https://arxiv.org/pdf/2404.11599

the word is the last segment of the url. very readable

Huh, looking at the photos, they really don’t do the board justice. It’s a gorgeous thing to look at and one of the only “bare pcb” tools I’ve used that still functions after years of being thrown in a backpack. It is Fully passive I think (the newer BLE one does emarker queries etc.). But it does the job very well. I haven’t had it fail to identify the capabilities of any cable yet.

This is the thing I almost, but don't quite, get.

Throw away what doesn't work - either because it is complicated or...its protocol is needlessly contrived.

Next step: do whatever you want. Upstream it, downstream, whatever.

Somewhat recently, on a wild night out on the town, I ended up buying a USB-C <> USB-C cable tester[0]. Using it, I managed to find a few amazing cables amongst my collection. Gave away the others with context. I want something like this for every type of cable.

[0] https://caberqu.com/20-42-c2c-caberqu.html

How is that open source? There’s not a single line of code. No CAD. Nothing. Sure, there’s a “coming soon” CrowdSupply link, which I guess is more than a KickStarter. But open source?

I've heard it as shorthand, in-context. Never heard it being chat-as-in-kleenex. I've supposedly been in the industry since before ChatGPT. Maybe that's where I went wrong :D

Socrates was a man. Socrates was creative. Ergo, Socrates is on HN. Ergo, we’re all dead.

Used to regularly take the shortish flight from Boston to Chicago, and then a bus >3x as long to Madison to see family, because the puddlejumper was too expensive or not even running.

Depending on the cost, and availability, 125 miles gets you within easy public transit range of NYC from Boston.

Now, of course, most of the hassle of getting to NYC from Boston is transit to and from the airport. And of course, transit from JFK, Laguardia, etc into the city.

There are at least 5 capable airports/airstrips closer to me than Logan or T.F. Green. Only one of them, Worcester, offers NYC flights. Worcester is almost an hour from me on average, and a bit more expensive. It’s an hour from takeoff to landing (I’ve arrived 10 minutes before takeoff, at the PARKING LOT many times without it being an issue).

If they can get this in and out of even smaller airports, like in Stow, MA, which is beyond capable, it will be a gamechanger.

GP is referring to GGP’s last paragraph. 150k t/m, yes, and 128k context.

As a user, I’ve not thought too much about this before now. I agree with you mostly, but the keyup on space behaviour actually feels so innate I’d hate any change to it. Keydown on space is “jump”; nothing else.

Oh, wow

https://learn.microsoft.com/en-us/lifecycle/products/windows...

Windows i686, tier 1, what? I’ll do some digging when I’m home but I feel the sudden need to understand this.

He also posted it on a public blogging platform - https://xcancel.com/moxie/status/2091218652133732491

People there still behave the same though

I used to scrap to make an extra buck, it was super fun. I did it for about half a year as my primary income. That was wicked fun, and exhausting. I’ve never been anywhere near that strong at any point in my life, before or after and did not work out once. Did not pay tech wages, but if you’re Smart about it and not unlucky, or if you’re just lucky, you can make a fair bit more than people here seem to expect. I’m back mostly to doing it here or there, mostly for fun, when an interesting opportunity presents itself.

Every scrapper is different. How they source, tracking yards or private buyers, how much they strip down what items (how do they value their time, with a modifier of how much they enjoy it). I’m sure it varies a lot geographically. I enjoy scrapping retired public/municipal electrical infrastructure/equipment municipal very enjoyable, and if you ask around enough you’ll find auctions or even towns willing to let you do their dirty work.

Always be scrapping.

> Edit: Does anyone else notice the switching between dashes and em-dashes between paragraphs? Tells you a lot about the man, doesn't it.

That he doesn’t know how to use basic word processing, or even agent SKILLS.md or whatever it’s called now. Or maybe it’s just the next generation of vagueposting.

I had floaters for years, they slowly disappeared. Occasionally start coming back again and disappear. When they disappear, I initially kinda miss them, because I start using them as stopwatch. Just spent 5 mins checking; I don't have any right now. I really think I would enjoy having one or two of different sizes and speeds, permanently. I guess that's just wishing for a superpower.

Not relevant at all, but I find that interesting and funny. As a rule of thumb I tend to accept unknown callers after 3 missed calls in rapid succession. Everybody I know closely knows this, and its come in useful. Though tbh I’d it’s a recognizable localish area code I usually pick up on the 2nd call in rapid succession.

I think in this case - where an editorialized title is better representative of the article, but the published title is too long - it’s usually left to someone like you to give a suggestion, publicly, which is then seen by @dang or their kith, who can then actually make spot adjustments. I personally approve of your editorialization, but that’s worth naught but a canker in a hedge

I don’t want to be a downer here, but the structure of this repo and the verbosity+language of the docs feel 80% vibecoded. Not that that’s wrong; I just feel kinda gullible for even clicking on this.

One uses xargs or parallel only a few times before they remember some of the quirks that but them. And then they become cautious. And then it’s muscle memory. And if it’s not an often occurrence, they learn to check the man page.

Anyways, in a world of “vibecoding” why add another “tool” to the mix when the LLMs have been trained on all our stackoverflow-posted grievances to begin with?

I wish this form of Veganism was more well-known. I know a few Vegans who went down a similar path and who were told by a few people they were no longer Vegan. They took the easier path and stopped calling themselves Vegans, which I thought was sad given their objectives when becoming Vegan were clear from the get-go.

Caffeinate is burned into my muscle memory at this point. I attribute that to the realization that you could repeat opt flags. There's something catchy about

caffeinate -dimsum