hex4def6

I'm not convinced he's done the calibration, at least in his demo video. It's pretty bad, imo.

I think he's just running it on the raw unrectified images.

For anyone wanting to play with this, OpenCV has a number of tutorials on this using stuff like stereoBM (which also uses SAD if I'm remembering my algos).

The first step is calibration -- intrinsics (basically figuring out the equations that can cancel out the lens distortion) and extrinsics (how far apart are the two cameras in XYZ, tilt, etc).

Regardless of the validity of the terrorism charge, it seems pretty reasonable to call what he did destruction of evidence.

Now, if he had taped "pass:1234" on to the inside of his phone case, and the CBP official had typed that in I'd argue it wasn't his action. But the act of giving false information that led to the auto-destruct is 100% equivalent to him typing it in himself.

The lesson imo, is to have something similar to trucrypt / veracrypt, in which multiple passcodes can unlock different sections of an encrypted drive (and the empty space is encrypted as well with a random key). That way, you have much more plausible deniability ("I gave you the key"), and it's not actively destroying evidence.

What about that chinese cat feeder whose app requires a suspicious number of permissions, and requires max version of android 2 versions behind the current one?

Or maybe your security camera has a "free" plan without motion detection, and you just keep the app running with a screen grab on top to do your own AI motion detection.

Or my eero router, which doesn't have a web interface... etc etc.

There's a million cheap hardware gadgets that are going to lose sw support at some time, and stop being functional. This will help a lot with that.

I hope sincerely hope they don't currently use "proprietary technology" like:

Wolfram Mathematica ($890/yr)

Magma ($2500/yr)

Maple ($680/yr)

COMSOL ($1500,yr)

Matlab ($500+/yr)

Seems like a very strange position to take, in my opinion.

Why does the field of mathematics suddenly now need to be "fair" and give everyone access to the same tools? Has that ever been the case in academics? It's always been a competition for name-recognition, grants, institutions, etc.

Macsyma / Maxima was an MIT developed CAS system back in the 60's that was proprietery until they sold it off to IBM for a tidy sum. Magma actually has free access if you're in the US, otherwise you pay. That's not to mention proprietary MATLAB toolboxes or specialized Stata modules.

Likewise, a lot of the above packages have pretty sweet site-wide deals with R1 universities. If you're at a smaller, foreign one, you're out of luck.

sure?

The point was that there are a number of recognizable brands that have fewer stores than pandas in existance.

If you do want by to go by 'sales volume', it looks like the cost to rent a panda is about $1M/yr, for a total of 2.5B/yr. So you drop down to about position 43 if you want to sort by that metric. Still bigger than Midas, Jiffy-Lube, Baskin-Robbins.

We're pretty deeply in the apples vs oranges comparison at this point though...

That's the charitable way of interpreting it. The less charitable way is that they're bitter that open-source has basically been used to train their replacement, and that they want to turn programming into effectively a guild-based profession to protect the jobs of those presently in the industry. Similar to the button-making guilds of the past, for instance.

You then decide to ask it to vary the unbleached:white flour ratio. You ask it to suggest flavor combinations based on 14th century english cookbooks. What happens if the dough is knead it for 3 minutes? 30? Substitute the water for oil and egg whites?

You ask it to try putting the mushroom in the pizza dough. What about cauliflower? What if the tomato sauce / cheese / mushroom order is changed? What if I add the vegetables in a second cooking step?

You're not a cook, but I'd say that at a certain point you might be considered a head chef in this example.

Doing something for fun is all well and good if that's the goal. Finding the best pizza dough recipe doesn't actually need you to get your hands in the dough if you can get the AI to do it for you. That's no different to delegating it to a subordinate cook.

You don't write in assembly. Or, probably, C. You don't write your own graphics driver, or GUI toolkit, or OS kernel, or typeface...We're all just climbing the abstraction tower. It's just that AI is a larger jump than most.

As a EE by trade, AI coding has been life changing. I can read code, but since it's not my day job, trying to get a meaningful understanding of multiple codebases is frankly weeks of time.

I no longer have to beg and plead the SW team to implement driver fixes or feature ads; I can freakin' do it myself.

Likewise, being able to explore e.g, a new vision algo piped directly into the software stack allows for setting up a demo in an afternoon.

I try to stay very cautious, and thoughtful prompting / checkpointing / getting it to first search for similar PRs as a guide have helped make sure it stays 'boring' and doesn't break architecture.

So far, I haven't been bitten... :)

Looks like 1200 people in the last year. 1200 / 350e6 = 3.4 micromorts. By comparison, getting out of bed at age 45 is about 6 micromorts.

The UK have about 13,000 arrests per year over social media etc. Without litigating all 13,000 of them for validity, that's about 185 microliberties.

Birth control is about 1-5 micromorts.

So you're 30x more likely to be arrested for a social media post than death by getting out of bed, and 55x more than being killed by police. And it's in the same range as taking birth control.

Realistically, these are population estimates. Your actual risk of dying by police is highly correlated with the number of violent interactions you have with them. If you're a getaway driver or a bank robber vs SW developer, you probably have significantly different risk profiles.

On the other hand, edgy social media posts are potentially much more diffuse across populations.

Think you have an error -- it's t_hot / (t_hot - t_cold)

With those numbers, ideal carnot would be 500/(500-357) = 3.5.

Multi-stage could potentially get you to a COP of 2 or so. So 0.5MW.

Would you?

In the UK, they can't force you to speak, but not answering questions can be interpreted as evasion / "adverse interference".

They do not need a warrant to search your phone. Not providing them with a PIN can be a separate charge, with up to 2 years imprisonment.

They can hold you up to 36 hrs if the police decide so. Up to 14 days without charge under modern broad "anti terrorism" laws.

> ...the mere possibility of them getting broken is no rational reason to abandon rules altogether.

Of course I agree with that. The problem is the enforceability / detectability of this on a global scale hovers around zero, when we're talking literally trillions of dollars on the line.

You would literally have to have every data center account for every cycle of compute it generates.

Again, the nearest comparison feels like bioweapon and nuclear bans. But the difference is the incentives for building those feel significantly weaker, so treaties that reduce and ban their creation are pretty easy to get consensus on, broadly speaking. By comparison, this is something that might provide a 10x competitive advantage to your country's economic, intellectual, and military might.

I think the only way we would get broad global consensus on this is for something Really Bad to happen, so that it moves it out of the fuzzy "sci-fi theoretical" into concrete existance.

The greater the competitive advantage, the greater the incentive to "cheat" the rules.

I am highly skeptical everyone will play by the rules, even if you could get people to agree on paper.

The difference between something like "unlicensed AI training" and, e.g, Nuclear weapon non-proliferation is that nukes are realistically only a deterrence against invasion. There are no economic benefits to having them in a stockpile. They're hard to make, and the manufacturing and testing steps are pretty obvious to an observer.

This is opposed to AI training in a data center that might host remote-stream video games, hospital infrastructure, protein folding, etc. How will we be able to police that in any realistic fashion? Will China or the US allow unfettered visibility into all data centers data streams to outsiders?

And again, the incentives: Imagine having Astra v3 or Opus 8 while everyone else is stuck with a lobotomized version of GPT 4o.

What about the military applications? The weapon of the future is autonomous drone swarms. If I'm a superpower, I'm pouring hundreds of billions into that.

About the only realistic way this is going to work is if we are able to solve alignment in a way that doesn't significantly hinder AI development. If it imposes even, say, a 20% handicap, people will ignore it.