persolb

Along the lines of presentation layer being transient,I’ve been experimenting with LLM and accessibility and screen reader to generate cleaner UIs and APIs. Haven’t spent too long on it, but parts of it seem to work. Speeds the biggest issue.

I haven’t tried this with new LLMs yet.

How schedules are crazy, but the results are in the right direction. Current FSD does seem safer than humans. Reusable rockets are now a thing.

But hundred of miles of tunnels, and driverless cars, and Mars base… not yet. Probably not ever for Mars.

I do engineering on these systems and have some expertise on this particular class of event. We’ve had tabletop red teams, and the general outcome is that rail systems being designed to ‘fail safe’ is an easy target. Failing safe at an individual actual ‘fault’ is still the best option, but it could be abused at scale.

It’s near impossible to cause two trains to run into each other (unless you are physically operating one of them), but it is really easy to cause all trains in an area to stop because a cross check or a bit has flipped. In this case, they literally flipped a bit by short circuiting the track circuit.

If done at scale, it is still possible to bypass and run trains safely the old fashioned way. It is less safe, but would probably be the decision if this was happening at multiple locations everyday.

Maybe won’t work for your site; I’ve been leaving flat files open, putting anything that needs processing (like search and login) behind Anubis.

There are at least dozens of us doing this. My $5k bloated corporate laptop sits in my office 24/7. The 25% of time I’m traveling is with a $300 Chromebook with Linux Mini, a Bluetooth mouse, and logged into to a Windows VM.

Now that everything corporate is forced into the cloud, using a VM doesn’t carry all the extra downsides it used to.

and the bonus is that, when the workday is done, I have a machine I actually own and can use without breaking policy.

I think the root of much of this are money laundering rules. The feds want to be able to punish bad guys without actually proving the crime they are suspected of. There are now a large set of rules which the government isn’t efficient enough to enforce, so it has actually become the banking industry’s legal responsibility. Once such things are normalized, they get stretched.

Everything about this looks fake, and Tesla confirmed it wasn’t real. Not finding great sources right now: https://www.carscoops.com/2025/08/rapper-fakes-video-claimin...

I would be amazed if there are not interlocks to prevent an update while in drive. On normal updates there is definitely an interlock, plus a timer, but theoretically it is possible.

In order to consistently output the same fake prompt, that fake prompt would need to be part of GPT’s prompt…. In which case it wouldn’t be fake.

You can envision some version of post LLM find/replace, but then the context wouldn’t match if you asked it a direct non-exact question.

And most importantly, you can just test each of the instructions and see how it reacts.

It is doable but highly location based and still really tight. I’m in a tier 2 US city. Our interns wage would equate to $58k annually. A car isn’t needed if in the city. Most people have a roommate, which works out to just over $1k a person.

58 (pay) - 21 (tax) - 13 (rent) -13 (transit) leaves $900/month or $30/day.

Full time grads are around $20k higher….so roughly triple the amount left over.

Or you commute further. I’m mid career and commuted more than hour most of my career because it let me get a better/cheaper apartment while affording a used car.

State governments do this too. NY under Cuomo just dictated a 10% cut to invoiced labor. It was widely understood that saying no meant no more work in the future.

Potentially this is like giving an axe murderer an uzi. They don’t need it, but they’ll probably be more effective.

Alon Levy being brought up on this topic always tweaks my “but somebody is wrong in the internet.” I’ve been on several of the projects he talks about. He’s right about the macro numbers and the general vibe, but often wrong when he starts talking about he details.

The main issues are, in general: 1) increased regulation, which includes internal self-regulation. Lots of rules that are preventing potential minor problems, but have a lot of overhead to follow. 2) large projects are treated like a Christmas Tree. Everybody expects their vaguely adjacent hobby horse to be addressed by the project… so scope keeps growing. There is ALWAYS something you can point to that has a good cost/benefit; and always addressing these ensures that the project never actually finishes. 3) lack of decision making. There is a general analysis paralysis and fear of making the wrong call. It’s often cheaper to just move ahead and risk rework. By not moving ahead, change orders are being incurred anyway.

As much as a hate saying it, the best thing for any large project in these orgs is being run by a semi-dictator who has enough political capital internal to the org, and who strongly objects to anything outside of scope.

I really like this idea; I’ve made crappy versions myself a few times…. Maybe this one will stick.

(For others, it’s a method to follow people across multiple services without being a normal feed. A person who updates only shows up once.)

Is AmEx any better? I’m planning to cancel my Mastercard with this gatekeeping as a reference to why. It seems to be the most effective lever most of us have.

It’s done in various places. NJ Riverline is an example. There are a bunch of others.

The bigger problem is the freights just have no interest in sharing the tracks with passenger trains, and requiring heavier and more expensive passenger trains is a convenient way to price the project to death.

As a mid size company that does work with government agencies, it’s near impossible to use anything ‘better’ solution. Cybersecurity requirements are getting so onerous that Sharepoint is too commercially feasible of an option to use anything else for a shared file store between organizations.

The fact that Sharepoint sucks* doesn’t matter… because anything else is seen as a risk.

* folders with lots of files are hard to scroll through because each page is lazy loaded, the automation functions are buggy, logins between different M365 tenants breaks and is not correctable by a normal site admin, human readable URL paths aren’t standard, search is shit, tables/filters are buggy, the new interface hides a bunch of the permissions logic, some things like permission groups need to be managed via outlook, etc etc. I’m sure a bunch of my gripes are technically fixable, but these aren’t things that should need a web search in order to use/fix.

Having H1-B visa employees introduce overhead. Not all companies are willing to do it. We’ve recently decided to stop H1-Bs for anyone that isn’t a senior hire.

I wish they’d named FSD as Autopilot. FSD is amazing and I use it door to door wherever I drive. I got a LOT better about a year ago.

But it still makes some dumb mistakes. It is not ‘full self driving.’ It’s 99% percent Autopilot… but I still need to pay attention to the road. (And it enforces this more/better than it used to via the driver camera camera)

I don’t want to guess how long that last 1% will take to be a low enough level of risk. All I can say is that 4 years ago it would do something hazardous every couple minutes. Now I can go hours without disengaging for a safety issue. Most disengages I have now are because it is more cautious than a human driver, or because a scenario is something that I wouldn’t be able to react fast enough to fix IF it did the wrong thing.

I believe this is only used on one passenger line in the entire country. This is really a freight based system intended to transmit brake apply signals as speed of light instead of speed of sound.

Since passenger trains are usually short and often have a wired bus) they don’t really need this system.

Yeah. Exactly. The consequence of this club is either:

1) The brakes take an extra couple seconds to apply (note: this is only used on long trains… so stopping is over a minute anyway)

2) The emergency brakes apply. This is considered a safe condition, and for Positive Train Control is considered the ‘safe state’.

If someone tries to utilize this vulnerability, the EOT device will be shutoff. On the few tracks where it’s actually required, there are mitigations to still operate safely.

This would be really easy to annoy a single train crew. This would be really hard to do to geographically diverse trains.

It’s really frustrating seeing all this about something I know about.

‘Slamming in the brakes’ is the default safe condition for a train. The freight train Positive Train Control does this automatically if the operating engineer breaks certain speed/location rules. It probably happens thousands of times a day at speed.

This whole thing is FUD.

This isn’t how US trains work though. There is literally an air pipe from end to end. It leaking air is how the brakes are normally applied.

The system with the vuln is just an overlay.

I work on trains. This is FUD.

Except for 1 train in the US, no passenger trains use this function. It is only for long freight trains.

If you block it, the train still brakes…. Just the propagation is at the speed of sound instead of speed of light. Functionally, it doesn’t matter.

You can theoretically cause the brakes to apply, but then this system just gets cut out anyway. It’s not really required.

I've actually used this fact in a related way, for wayfinding.

Old school Open-CV was able to see tracks well from an onboard monocular camera, but calibration and scale was annoying. Track width is accurate enough that I was able to use it to input a bunch of head-end video to map the tracks.

It was mostly just a modified edge detect where the tracks approximately would be. Once finding the tracks, you could automatically calculate the camera's height, lateral location, and angle.

Substack simply has better quality.

I've been involved in some things a handful of times that made it into the paper. Technical laws being passed, corruption, complains about a system failure... In every instance the only thing that was really correct was the simple facts (law X passed, thing Y failed, person Z arrested). Anything more nuanced tended to be 'technically' correct but was phrased in a way that often would make you think the opposite of what actually happened.

A coworker of mine was punched on NYCT. Went to the cop in the station, who did nothing, and was sore for a week. This is a case where it was ‘reported’, but the cop said they couldn’t do anything if he wasn’t seriously hurt. My other NY coworkers have similar stories of other people they know.

Maybe random minor violence is up, but major violence is down… but the number of people I’ve heard of getting assaulted has definitely increased since COVID.

The number of ‘we can’t do anything’ stories I’ve heard have skyrocketed. If anything pre-Covid cops were to quick to go after people. I got jumped after high school by kids I didn’t know and just got a black eye; the cops went and roughed up the gang of guys based purely on me saying ‘that’s them’ as we drove past.

I was going to say “I’d be surprised if organ support is a large issue to an adult. We already have long term astronauts and people who are bed ridden for a long time during recovery. I would expect an issue during development though.”

Turns out there is research in this direction: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC2379624/pdf/can...

It seems to say that there is a long term impact on connective tissue hardening… but it blames lack of stretching; not lack of downward force. Any regular movement would seem to fix that.

However, internally, organ tissue probably ‘bounces’ more due to gravity during movement… so less gravity means less flex of connective tissue.

TLDR: you seem to be right

Exactly this. Facebook makes money by network effects. They are incentivized to grow network engagement, more than they are to make direct money off new network members.

Is the air gap actually an issue in this case? If I'm understanding correctly, the suggestion is to put up additional cable instead of thicker cable. The two ends of the cable are both still connected to the same electrical phase, so there should be no arcing between them.

I think this somewhat makes intuitive sense. I think most people caught Covid. Covid was most risky for people who had other illnesses. In the universe without Covid, many of those people would have died anyway.

The people who survived Covid were on average healthier than the people it killed... so the life expectancy should go up post-peak-Covid.