krackers

>what happened in Ilya's mind

I thought Ronan Farrow's investigate essay answered that pretty satisfactorily? https://www.newyorker.com/magazine/2026/04/13/sam-altman-may...

I thought one of the main issues is that during sleep you lose muscle tone and control, so no matter how much you do myofacial therapy while awake it doesn't really matter much in preventing your tongue from falling back when asleep.

>basic technique of "mewing" is very similar if not identical to something that's been practiced in yoga for thousands of years

Can you elaborate more on this? I also assume by "mewing" you mean putting the tip of your tongue on your hard palette right behind your upper incisor, and letting the rest of the tongue suction against the hard (and soft) palette. Surprisingly for something that seems to be so "popular" in "broscience" I find it hard to find any canonical definition or technique.

See https://news.ycombinator.com/item?id=48233034

That was for this particular technique, but the comment suggested that a similar construction can (but is unproven) to beat erdos and at least gives some flavor of what this looks like

https://x.com/i/status/2057490144546927046

No, there's quite a bit of evidence that they last much longer than that.

> The only thing I've really been a bit tempted by is Lidar on Apple

Maybe in that vein, one thing I wish phones would do to differentiate themselves is just add more sensors. I want my phone to be the tricorder from star trek. iPhones should have first party support for generating point clouds and measuring distances using lidar. Their microphones are probably already calibrated, why not expose that as a decibel meter. Same for light sensors. Phones used to have IR emitters, why not add those back in?

Also the iPhone still only has 240fps slow motion, I've found samsung's 960fps really useful in capturing transient phenomenon or even measuring mundane things like LED flicker.

Conversely the Pixel seems to be the only one shipping with an IR thermometer, and they'll probably remove it given most people don't seem to care. That's something I would've found useful in ad-hoc situations where I've had to make do with the back of my hand.

Air quality detection (especially pm2.5, CO2, and CO levels) would be great but I don't know if those sensors can be miniaturized enough to fit.

I'm not sure what you mean, tetanus shots actually last 30+ years.

I wonder if anyone has taken the source of TextEdit [1] and added some minor niceties to just make it slightly more convenient for text editing (like adding line numbers).

[1] https://developer.apple.com/library/archive/samplecode/TextE...

Both good points, thank you! I hadn't appreciated how complex it was, reframing it as something closer to spectometry makes the difficulty clearer.

>He often needs to ask people to smell him during the day, which leads to some hilarious situations sometimes

Smell and taste seem to be the last two senses/modalities we can't really work with using technology. Vision (cameras) and sound (microphones) have existed for a long time but it's only within the last decade that they've become ubiquitous in the form of a smartphone, and only within the last 5 years that ML is good enough to work across them (ocr, stt).

But for some reason (maybe lack of easy commercial opportunities) we haven't even come close to making "artificial noses" to record raw input. Maybe as part of the push for "embodied humanoid AI" (e.g. Figure) we'll find a way to do that.

It's not going to break, it'll selectively bend to the gravitational pull of wealth like it always has. You won't be able to use anna's archive to "download" that out of publication book, but companies will happily train on all that data and charge a subscription for you to prompt out a summarized version of it.

They can simulate "driving out of a raging fire" but not a flooded street? This seems like an admission that the fancy "world model simulation" doesn't actually mean much

https://waymo.com/blog/2026/02/the-waymo-world-model-a-new-f...

If one only found meaning in life through external factors like work (no matter how "intellectually rewarding") then it seems like a life destined for eventual disappointment.

>Google SERP results have been gamed by SEO slop for economically valuable search terms

So why not use AI to find the obvious spam and SEO? Sites like pinterest, "geeksforgeeks", stackoverflow clones were notorious for ranking highly on almost every technical search and you didn't even need AI to deal with those. Using AI to provide distilled summaries and making it harder to get the actual links is going to make things even worse for the user, because you have to cut through two layers of slop. And there's also not using the smartest model for the responses anyway.

I'd like to see a sequel where the fractional fourier transform is used for image restoration

It seems plausible given that people have been using off the shelf 5.5 xhigh to decent success with some erdos problems. There is likely still some scaffolding around it though (like parallel sampling or separate verifier step) since it's not clear if you can just "one shot" problems like this.

Saving bytes on the wire?

>fine boo, but you need do something about it

Well they are doing something about it, just not the way the speakers had in mind.

It helped that he actually used and believed in the products he was pitching.

I don't think there is a difference today. Stocks used to be about long-term value but now the same gambling-esque mentality has applied to stocks as well, see things like GME. The modern stock market is basically a prediction market for attention/virality, in the same way that cryptocoins are.

>It's just a question because we don't have a good way of explaining how/why it occurs.

It's that you can't even measure it, since the way it's defined as a subjective experience, no external measure could ever capture it. This is what gives rise to the p-zombie argument.

To get rid of that you have to accept "functional qualia" as basically equivalent to qualia, which solves the p-zombie issue and resolves half of the hard problem. From there, explaining consciousness is no "harder" than explaining other scale-depedent phenomenon in complex systems like LLMs: still hard, but at least tractable with scientific measurements and experiments.

I don't think AI safety/psychosis/alignment issues are on most people's minds. All of the other ones are basically downstream of AI furthering wealth inequality. Job market for obvious reasons. Environmental concerns are mostly downstream of the fact that cities are suddenly accommodating data centers while they didn't care about promoting growth of infrastructure before, and citizens are asked to foot the electrical/water bills. (People would not care about environmental impacts as much if they weren't immediately impacted, say the DCs were built in Africa).

Also missed is the pushback against AI art: the further devaluation of talent, and an associated loss of meaning many people have. I think this is probably still downstream of it threatening jobs though, since people would not react as violently if they could truly treat art as a hobby instead of as a profession.

I personally feel "happiness" is more correlated with agency (or at least perceived agency), and in that measure civilization has been regressing since the industrial revolution. The amount of long-term planning required has increased and it's less possible to live "in the present", moment to moment.

It's also basically what was written about in that infamous manifesto.

How about "permanent underclass"?

There's a talk by Filippo that explains this nicely https://www.youtube.com/watch?v=0DV8WnqhH2Y

> which can be unwritable for a human

Unfortunate, those types of refactorings are my favorite, since they're tightly scoped, easy to verify correctness, and it's like a little puzzle. Bonus points if you write your own collector or use some more obscure parts of the stdlib.

> But how do you get 'Paris' into the value vector in that case?

Ok wait I think I see what you mean. Although maybe it's not getting paris _into_ the value vector that's hard, but isolating the residual stream to _only_ that instead of things like other capitals.

So as a naive example maybe at the very first layer consuming your tokens: Q{France} would have high inner product with K{capital} and so our residual would now mostly contain V{capital}, which maybe contains embeddings of all the capitals of all countries. You need some way to filter out all the other stuff, but can't do that without a FFN + activation.

Just throwing in a relu by itself won't help since that would still work on all the elements uniformly, you need some way to put weight on "paris" while suppressing the others, i.e. mixing within the residual stream itself.

Although maybe if you really stretch it, somewhere in a deeper layer you could have 1-hot encoded values with a "gain" coefficient so that when you do the residual addition it's something like {<paris>, <tokyo>, <dc>} + 10000*{<1>, <0>, <0>} and then if you softmax that you get something with most of its mass on "Paris". But it seems like this would not be practical, or it's just shifting the issue to how that the right 1-hot vector is chosen

I guess this had always been bugging me. I get while you need activation/non-linearities, but do you really need the FFN in Transformers? People say that without it you can't do "knowledge/fact" lookups, but you still have the Value part of the attention, and if your question is "what is the capital of france" the LLM could presumably extract out "paris" from the value vector during attention computation instead of needing the FFN for that. Deleting the FFN is probably way worse in terms of scaling laws or storing information, but is it an actual architectural dead-end (in the way that deleting activation layer clearly would be since it'd collapse everythig to a linear function).