It doesn’t hurt that American frontier models are phenomenally huge and cost an exorbitant amount per token in and out while the main source of real world value created by llms in the past year has been through agentic/harness/other words for massive context systems that show a real benefit from simply using more tokens.

Dictating a post while doing fifteen over on the way to get busted by a tcap yt channel type beat.

A team of professional engineers couldn’t detect terahertz radiation without grown nano scale antennas any more than they could use the coils of the size implied in your picture to differentiate between two cars of the same make and model let alone cars that aren’t in wildly different weight classes.

It’s not a question of pointing enough clever people or computing power at the problem, it’s a question of trying to dig a hole with a ball peen hammer in one swing. The tool is inadequate for the job, uses the wrong motion and doesn’t have enough time to do anything.

and for the example you gave where analysis isn’t definitive but could be paired with automated readers to clarify a picture, simple presence detection does the same thing and costs much less!

And come on, you know that just to cut those little octagons in the asphalt, drop in the coils and tar over em they’d send a ten man crew in four diesel trucks and shut the ramp down for the whole business day.

If you have any of the releases from those days or even scene releases form the before times, or can rip and upload, they are still desired on red and oph.

No, now that “cognition” of a sort can be smeared on any task at the cost of electricity the future belongs to those who control the generation of electricity.

Local models are much less efficient in terms of both power consumption and cost per token.

I am currently running a local model. It’s not efficient.

I didn’t get into energized coils and how they can be used to recognize objects because there’s much more to cover than just relatively accurately explaining how a record player cartridge and speaker work.

On the speaker end, and you’ll be familiar with this if you’ve built a metal detector before, the coil size can make it much more sensitive but that means that, and you’ll be familiar with this if you’ve ever built an am radio before, you also need to be selective in order to pick up precisely what you’re looking for. The old radio shack Forrest Mims books had compatible metal detector and am radio selectivity circuits iirc. With a change in the dropping resistor they could be daisy chained together and with an adjustable component or two in the bp filter you could stay over the same area and adjust your selectivity filter to reject all the rocks and red clay and show that one little nail buried four feet down.

The problem with using a selectivity circuit is it needs to be high order and in order to not fuck up the phase portion of the detectors chirp circuit, relatively phase linear. It’s okay if it changes the phase across the band as long as it changes it extremely consistently. That’s hard to engineer correctly and requires expensive parts so it’s less common than you would hope (thirty years ago when I was a little kid fucking around with shit in the garage).

Okay so metal detectors exist and big coils take extra circuitry, controls and training to be useful, who cares, the point is could coils like the ones pictured be used to reliably identify specific vehicles?

No. You could probably id the general weight and makeup of a car, if you knew all the other environmental factors. You gotta consider the sample size. You’re looking at a fraction of a second sample size and it’s gonna be a blip shape (just like it’s a blip sound when you sweep the metal detector coils over a nail), so you’re limited to analysis of blip shape and superimposed patterns. Blip shape is gonna just tell you general makeup, how much aluminum versus steel is there and where in the passing object, but you’d have to have an exhaustive database of different model/year/options to be able to correlate your reading to some kind of car and a good understanding of how each model ages to be able to account for the difference between a rust bucket and barn find (severely corroded pennies sound different than bright ones!).

In terms of superimposed patterns, you’re in even more difficult territory because it’s gonna be influenced by engine speed, driveshaft speed, wheel speed, and the mass of those components themselves. Go over the coils in a different gear than the last time and suddenly the reading is different enough to not be correlated.

All of the above analysis and more would require the computation power of a logging oscilloscope btw.

So if your goal is to track individual vehicles, coils are a bad choice. You need extensive knowledge, datasets to test against, repeatability, a system capable of at the very least high frequency data collection and you still get defeated by someone on the off ramp in third gear of hauling a trailer.

Or you could just use a tag reader and actually know exactly what car went where with a timestamp for likely less than a tenth of the price and no blocking off the ramp for a week while they install the coils.

I am a Unix jihad, gpl em all and let god sort em out, rust programmers are compradors type motherfucker but a company not wanting to use your software because of the license is a barrier to use.

Np. It’s easy to just choose the license that is easiest or “fire and forget”, but too many programmers don’t consider what kind of freedoms they want their speech to have and that the license is how you define that.

If you want to know something feel free to ask.

The point of different licenses is the thing you ought to be looking at.

The point of mit is to get the licensed thing used as widely as possible. It breaks down any barriers to use. A good example of something that should be mit licensed is an encryption algorithm. A good example of something that shouldn’t be mit licensed is anything else.

The point of bsd3-clause is to prevent the licensed likeness from being used to imply the creator shares some view or idea. A good example of something that should be bsd3-clause licensed is the royalty free music on youtube. A good example of something that shouldn’t be bsd3-clause licensed is a creative work the author wants to have their name known for.

The point of the various gpl licenses is to create a digital commons that all people can access. A good example of something that ought to be gpl licensed is a game engine you want people to be able to freely use and modify but never make secret changes to sell. A good example of something that shouldn’t be gpl is a game engine you want to eventually sell the code of or make secret changes and sell the code of those.

If my language sounds biased, it is because I have a bias.

As a person who has some expertise in a related field, let me explain:

If the covered area represent coils, those coils are not appropriately sized or positioned to detect differences from car to car. They are appropriately sized to do what the other poster said and provide a presence input that can be used to feedback into a traffic control system or a simple counter to analyze road use.

A current is induced in a coil when a ferrous material is vibrated near it or when an electric current flows near it. The frequency, intensity and wave shape of the current is directly proportional to the movement of the ferrous material or flowing current. The degree to which the current is induced is modified by coupling strength and the structure of the coil.

A coil that size would be really bad for receiving induced currents that reflect various relatively small ferrous moving parts like pistons and wheels whose particular signatures could be used to identify a vehicle and also bad for receiving induced currents that reflect electrical activity which could also be used to identify a vehicle.

It is the right size for receiving induced currents that indicate a large ferrous object like a car has passed near it.

Which is of course what a coil like that is used for in civil engineering.

“Who could have ever predicted this?” We all said.

You’re right, and there’s two reasons I don’t host email or password vault.

I don’t host email because it’s out of scope and a ton of work for no real gain. Email is incredibly insecure by design. You can’t make it secure. Internalize that. Once you need to talk to someone else’s email server (to, you know, exchange mail) you throw all your security work on your server out the fucking window.

There used to be a bsd that advertised no exploits in the default install since <some year in the 90s>. They changed that recently I think, but it was a funny joke to everyone in the know because the default install had no services enabled. It’s pretty goddamned easy to have a secure default install when you don’t turn anything on!

The point of that digression is to show that for the security component of privacy, the best way to rest easy is to just not do certain things. You can’t cultivate a secure child left alone at home. The state of being is insecure.

Well, what if you’re willing to accept insecurity to be away from the prying eyes of Google! Okay, from a technical perspective setting up and configuring an email server in a way that will not immediately get flagged as spam and trigger blacklisting is a much higher bar than just downloading some docker image and pushing the go button. The stakes are higher too because failure to recognize and investigate failures results in mail simply being silently undelivered by other mailers.

So it’s more complex, heavily scrutinized and when you screw it up you silently lose the function of the server from a side you have no control over.

What should you do for email, then? Just use some other service. Let them handle dmarc etc. how will you know they aren’t scanning your emails? You won’t. You can’t. You can’t even know that the other party has a secure and private setup. When that starts to bug you, look into pgp.

I don’t host a password manager because of the same reason you brought up, it’s a single point of failure, what happens in a disaster?

Well just recently I was in a disaster. There wasn’t any power to run my home server, my offsite was unpowered and swept out the door into a ditch by flooding and there wasn’t any internet to reach my cloud backups.

Because I use a service for password management I was able to securely use all kinds of communication from public access and disaster response computers and store the credentials and secrets given to me in the process of beginning recovery (and oh boy were there a lot of em).

From my perspective a trustworthy service is miles better than hosting for credential management.