Intermernet

I'd say the NSA have been developing and training custom LLMs for at least 12 months now. They have the means, and they have the history. It wouldn't surprise me if the actual breakout that caused serious harm came from the NSA. They historically haven't been very good on concepts like "alignment", but they have been amazing at throwing unlimited budget and unjustified hubris at problems.

If various military groups are already publicly saying that they relied too heavily on ai, then I'd hate to see what the group with the pertinent resources and the culture of absolute secrecy is getting up to.

I have 2 cats that are sisters. They have a very interesting dynamic. There isn't a fixed dominant sibling, they switch back and forth.

They antagonise each-other most of the time, but they also support each other in times of stress. If one goes to the vet and the other doesn't, the initial reaction is hissing and being a dick. Then, after a few minutes, it's licking and comforting.

I don't like to anthropomorphise too much, but they definitely look out for each-other. Equally, they definitely annoy each-other.

More like a bunch of CNAMES pointing to a single record, but in this case the CNAMEs all start with "charlesv3".

We have reached a point where it's possible to make a car that lasts 200,000 km reliably, but the economic incentives are to make a car that people want to upgrade every few years. The ability and incentives are hugely misaligned.

My goto example for this is Hilti powertools. They can (and have) made some of the best tools in the game, but they now make most of their money from companies going the "fleet" option, where, for a monthly fee, tools can be replaced almost immediately when they fail. The result of this is tools that work brilliantly for a short time but are non-field-repairable, and aren't really designed to last any more.

Economic incentives beat engineering ability.

As someone who started using Linux in 1996 (Slackware, of course), I learnt the hard way about compiling packages. I remember trying to compile the kernel, and doing "make config" and following hundreds of options, and not knowing that I then needed to do "make" or "make zimage" or "make install" after any of the config stuff. Obvious, dumb, rookie errors, but still understandable for a teenager coming from Windows. Doing LFS really taught me a lot about dependencies and complexity, as well as understanding of the actual boot process. All of this helped me greatly in later years.

I'm working on a bunch of hardware stuff.

I'm up to my 3rd prototype of a device that combines GPS with TERCOM style algorithms using a bunch of sensors for use in areas that have terrible GPS coverage (initial use was canyoning). Hopefully, when I receive the samples, I'll have ironed out most of the hardware faults!

I've also just (sort of) finished a TinyGo version of firmware that runs on a cheap Adafruit board in combination with Gaudi Lab's OpenThereminV3 shield to provide MIDI theremin capabilities for live music performance. That one is available at https://github.com/intermernet/theremin if anyone wants to play with it.

I have to send a daily email that always has the subject starting with "Work Diary". If I type "W", it doesn't come up. If I type "Wo" it comes up as the 3rd auto-complete suggestion. If I type any more characters, I have to type the entire two words. This is on Android, but I have a feeling it's no better on iOS.

I'd love to know the algorithm they're using for this. Fuzzy search seems like it was a mostly solved problem 20 years ago, and it's definitely worse than it was 10 years ago.

"after that apple started to dominate"

When did Apple dominate? I'm not trying to hate on Apple here, but there's this weird belief that Apple have had a lead in personal computer OS market share at some point in the last few decades, and it isn't really true since about 1984.

You can argue that they should dominate, but that doesn't seem to have happened.

IBM actually do a bunch of semiconductor research, which they license to other fabs. I'm not sure how much money they make (probably nothing compared to support contracts with banks etc), but they actually do innovate. At least, a small portion of one of their business sectors still innovates. The rest are probably engaged in aggressive IP lawsuits or taking CFOs out to strip clubs on golf courses or something.

I'm sure you understand the dystopian irony of your last sentence, but I thought I'd highlight it anyway. That's some "no fighting in the war-room" level commentary on the state of the USA.

Isn't that the one that was left behind in 2002?

This seemed to have been accidentally downvoted to death. Just vouched for it, as it's a normal comment

Astral navigation is probably accurate enough

Now I'm glad I didn't have a webcam. It would have just added a few more steps to my tale of woe ;-)

My Amazon account is, unfortunately, tied to AWS, Prime, Kindle as well as online shopping. I hate this, and I'm in the process of removing all of my stuff from all of these services.

Minecraft looks very similar to original Wolfenstien or Ultima Online. Probably a bunch of earlier games that used blocks of simple textures. It's not an original art style, and it's not gonna stand up in any fair court. Notch never claimed that he invented the graphical style of Minecraft. He deliberately used low complexity graphics for ease-of-use and the ability to create large, performant maps.

I recently had a perfect storm of non-friendly technology.

I made one purchase on Amazon while connected to a VPN.

Amazon froze my account and demanded proof of identity.

Amazon don't accept an existing photo of primary ID for their preferred ID verification check, it must be a photo taken by a camera on-device, and I don't have a webcam on my desktop PC. Once you select the secondary method, it seems you can't go back to the primary.

My bank can't issue a statement on the mobile app.

My bank, even through the website, can't issue a statement that has the required details (name, credit card details - blanked except for last four digits, and transactions) that includes pending transactions.

I had to manually request a statement from my bank (via a phone call) that included the pending transaction, edit it in an image editing app, submit it to Amazon via their web-site.

None of the apps (bank, Amazon, government ID, camera) had the flexibility to solve the problem.

The root cause was using a VPN to make an online purchase of a $12 piece of hobby electronics.

Not only are apps broken, the current state of the internet (and the random ways companies try to enforce "security") is also broken.

I wanted to like this, but the article is rife with Claudisms. I don't really care if you use LLMs to generate interesting articles, but maybe re-write it, or prompt it to not produce output like this?

1. The Guarantee: A Smooth Sweep Across Flatland 2. The Surprising Part: Existence Does Not Help You Find It 3. Why the Obvious Answer Fails: The Center-of-Mass Trap 4. It Gets Stranger: The Ham Sandwich Theorem 5. How I Turned the Math into a Game 6. Why This Is Harder Than It Looks

At least there's nothing loadbearing about it.

I still don't know which huge part of the training corpus led to this style. It had to be something massive that I never interact with in real life. Maybe marketing presentations? Pitch meetings? LinkedIn posts?

Where did the LLMs start to learn to talk like this? Are there actual examples from the pre-LLM era that anyone can point to that resulted in this style of output?

Not yet. I currently have one private repo with everything in it (hardware design, firmware, experiments, notes etc.). Want to clean it up before I release anything.

Weirdly enough, I'm working on a similar system for position tracking in canyons. It's hybrid, uses a combination of Kalman filters, particle filters and GPS (EDIT: and LIDAR based DEM) and is based on the obvious (in hindsight) realisation that the view of the sky to get an accurate gps fix is inversely proportional to the constraints of the terrain. If you have a low field of view of the sky, you can use sensor data to constrain your likely position. I should have hardware samples for testing in the next month or so. Hopefully my theory stands up!

x86 Software Reverse‐Engineering, Cracking, and Counter‐Measures By Stephanie Domas and Christopher Domas

https://onlinelibrary.wiley.com/doi/book/10.1002/97813942771...

Admittedly, that was probably the NSA getting their grubby hands into NIST. A good thing being almost permanently tarnished by a bad thing.

I'm usually not an Ai shill, but I found Fable to be the first LLM that could actually provide useful feedback on a KiCAD PCB design. I paired it with a KiCAD MCP, but it tended to reverting to raw KiCAD scripting. If you ask it to review a PCB design, and provide feedback relevant to PCB design best practices, it will probably give you something useful. I wouldn't trust it to do a full design, but it's pretty good at finding issues. I'd add a few clauses to the request regarding checking components against JLCPCB/PCBWay/etc available components, as well as checking copper layer thickness and distances from the main components to the supporting components.

I did this as an experiment on my current project and it found a bunch of minor, non-critical, but completely valid problems with my deisgn.

I'm in the middle of my first hardware project right now and I can recommend the videos on the "Phil's Lab" channel on YT. He mostly uses Altium, but he has a few videos on KiCAD which really helped me. He also has generally good throwaway comments in his videos with regards to PCB design / component placement, as well as the order of operations for good PCB design.

Rock? You were lucky! We used to have to hand pick our zeros and ones from sparse clouds of hydrogen and helium!

I feel like I shouldn't need to say this, but Simon is also the Author of the awesome PuTTY. Until fairly recently, a vital tool of anyone doing remote management from Windows.

When Netflix launched, you got the service without ads. That has changed. That's what's known as a rug-pull.

Flock provide more than LPRs. Check out the "Condor" cameras.

So why do you think increasing surveillance will decrease the stalking of children?

Benn Jordan did a video recently where he showed Flock cameras, which were hackable, pointing at children's playgrounds. Who is stalking the children?