Love the idea, I was thinking about this problem a few weeks ago, but never got anywhere with it. I'm intrigued by the latency savings in particular, that sounds great.

One concern I have is that right now each session gets fresh "eyes" on the problem. Right now I find I get a lot of mileage out of a combination of long-running sessions and fresh ones. I worry with a single generated concept graph that gets only incremental refreshes will become stale slowly, and in subtle ways that are hard to detect. That could lead to semantic drift in the graph from reality, and every new session will take the drifted form as gospel. Have you run any long tests (weeks or longer) on this to make sure that this doesn't happen? My understanding is SWE Bench is only a point-in-time evaluation.

Also the graph is stored in the repo, right? How mergeable is it? I know I wouldn't want to do conflict resolution on that myself, and even Opus struggles to keep all the references correct (especially when comments are involved) when there's an B->C, A->B symbol rename.

For the staleness specifically we are using hooks on claude code and codex, where on every turn, or post edit or pre-tool use, we sync the graph. so that graph never goes stale.

We have run tests on DeepSWE as well which are long running tasks, we got 20% better accuracy on the tasks where sonnet 5 failed. didn't want to post that numbers yet as I think we can do better on DeepSWE and on a cheaper model like gpt-5.6-luna or grok-4.6

The graph is pre-filling the context that does the sync, right? Doesn't that create a bias towards what's already present? If anything hallucinated does make it in at some point, won't the sync just reinforce that hallucination if some new component touches the bad info, just like a long session rabbit-holing? New sessions, in contrast, have no contextual assumptions and often catch those problems.

I only know one person who has ever found phenylephrine effective. It's definitely not for me, but they've done single-blinded self-studies (with help) to see if it's a placebo effect, and it's pretty clearly not.

DXM is also not a placebo, although it might be specifically for cough.

I don't especially want the FDA to ban them, but requiring separating out the acetaminophen might not be the worst idea.

Phenylephrine only works as a nasal spray. It doesn't do anything when taken orally, because it is broken down in the digestive tract.

I'm not much of a physicist, but this rates a 9.5/10 on my personal BS mater. A single citation, two lines of analytical mathematics, a promise to resolve the Hubble tension without actually doing so quantitatively (or indeed, even conceptually), and I'm pretty certain (> 99%) that the no-hair theorem applies specifically to black holes, not other massive objects.

I know a little bit more about orbital mechanics, and I know that the internal structure of a planet matters for the stability of an orbit. While at larger distances the effect becomes small enough to ignore, it never really goes away. Whereas a black hole is indeed thought to behave like a point mass.

This was planned as a submission to the Gravity Research Foundation for 2027.

The Dirac Spinor equation is extremely standard, and GR would be the more famous one. The only "leap of faith" here is having the metric to be the functional of the matter wave, and not just the stress-energy tensor (which is the mainstream semi-classical approach).

Simply sharing here to prove authorship + sharing something interesting to HN.

edit: Though I note that I should have quoted Birkhoff's theorem instead, thanks for the pointer.

> (> 99%)

Yep, 100% it only applies to black holes (after enough time, so they can radiate the weird parts, but this time is short in human scale)

Eh, there were plenty of games that were coded for a particular clock speed, and then once the SE came out, had an update that included a software version of a turbo button, let you select which of two speeds to run at. They run FAST on an SE/30 or Mac II and unusably fast on anything newer.

I didn’t encounter too many of those back in the day, I think because there was the VBL task mechanism for synchronizing with screen refresh that made it easy to avoid using instruction loops for timing.

Much more common in my experience was the assumption that the framebuffer was 1-bit, but such games would still run on my IIci if I switched to black & white—they’d just use the upper left 3/4 of the screen since they still paid proper attention to the bytes-per-row in its GrafPort.

Could be that by the time I was using a Mac II though that all the games that didn’t meet that minimum bar had already been weeded out.

Yeah, there were a bunch of floppy games which only ran on an original Mac or maybe a Plus. No go with even my Mac SE.

Yep, that’s definitely the question. The article says that there are caches of recently used graphs for use in large loops. Presumably those are pretty fast to swap, but I have to imagine programming a whole new graph in isn’t fast. But maybe the E2 or E3 will have the ability to reprogram partial graphs with good AOT dataflow analysis.

For sure. Pretty similar problem space, very different implementation and target market. Apportable wasn’t a transpiler (and boy were we proud of that); it was a reimplementation of the iOS frameworks on top of a replacement NDK (Android’s libc was bad-mediocre at the time. Real nasty behaviors in its (dl)malloc for instance). Similar to WINE. It was always targeting games, so there was never a lot of effort to port any of the UIKit controls, but OpenGL CoreGraphics were supported. It also had a compiler extension that let you access the whole Android SDK in ObjC, and a tool for generating ObjC APIs from JARs. The goal was to make ObjC the one true mobile dev language. I applaud the effort to do something similar for Swift, even if it does involve transpilation.

While the large sample size and many controlled variables are good, and make the results statistically significant, the effect sizes are awfully small. Only two of the emulsifiers, tripotassium phosphate and guar gum, increased the hazard of an individual developing diabetes by more than 10% (11% and 15% respectively, with very large error bars in the latter case). These are much smaller effects than previously known Type 2 diabetes risk factors.

For instance, the HR for family history of diabetes is 4.46 (vs 1.03-1.15 here, where 1.0 is no change in hazard).

No. None of the ones with an increased Hazard Ratio are or contain them.

The researchers also did not control for PUFA consumption, only SFA consumption.

> Even Super Mario World[10] got the treatment (I can't remember slowdowns but I was only twelve can then).

Yoshi’s Island 4 has a slowdown in some circumstances (have Yoshi, get Starman and hit P-Switch), as does another level I can’t recall, exactly… it has a bunch of Monty Moles that explode all at once. I think it’s on Chocolate Island. I think there might be a third with two Sumo Bros. and an Amazing Flying Hammer Bro. onscreen.

I remember a ton of lag in Outrageous

This again? Last time this came up, it was calculated that in order to get to the levels (mg/kg body weight) found to affect fertility in study rats and pigs, you’d have to eat something like between 2/3 and 2x your body weight/day, every day (because it’s easily excreted), of chlormequat-contaminated oats. The studies on rats and pigs used much higher concentrations than are found in oats.