I don’t think I’ve ever met someone claiming to be able to easily maintain 70 km/h. Maintaining 50 km/h for an hour puts you well into top professional territory, especially if riding solo.
There’s basically no chance you got to that level without serious training, coaching, and a lot of experience.
That is a very different situation from just using a credit card and being able to zip down the road at 50-60 km/h. People have been killed by these fat bikes (as in, a pedestrian being struck), because fat bikes are significantly heavier than road bikes, and kids with no experience drive them in places where pedestrians do occur.
I doubt you were pulling 50+ km/h in the city centre, or on the beach promenade. Yet this is what we see with fat bikes.
The laws aren’t designed to protect the rider. They’re designed to protect the uninvolved bystanders who just want to enjoy a stroll.
The thing is, I lived where I had several routes of about 2km length with several steep inclines of 12% in them, right from my door. And not much else to do. So I did that, first on a road-bicycle with 26" rims for youths, which I grew out of very fast.
Got a bigger frame with 27 x1 1/4 then.
Now when that was new to me, I've been KO after riding up there, even needed to step of the bike, some times. But I persisted. Got myself some 'mountain gears' for the rear hub(ten speed only, so five mountain gears back there).
That helped. But I grew out of these, too! Because I didn't need them anymore!
Installed the normal ones back, and thundered uphill as if it was nothing, being just warmed up enough to thunder over the mostly flat, and excellently paved ways going through the forest on the high plateau.
Giving it all, until absolute exhaustion, pulsating tunnelvision, nearly 'grey-out'. Again and again. By myself. No coaching whatsoever. Until I didn't have these grey-outs anymore. I later discovered this is called "Interval Training".
Topped that by installing cranks two centimeters longer than usual, and installing 'speed gearing' front and right, to get an even higher transmission ratio.
Where only 3 to 4 speeds were really usable for me. The rest I had no use for(most of the times). I started mostly in the eight gear, carefully, to not burn rubber, because tires were expensive for me. Didn't help much though, because even with that gearing the back wheel slipped when I pushed down hard from stand in tenth gear.
So wheelie it was, because why not? Whoo hoo hoo!
> I doubt you were pulling 50+ km/h in the city centre
Of course I've been, to show off! :) Sustained for my way to school for about 10km, without breaking a sweat, not arriving wet and stinky. Even in bad weather. Because that took me 15 minutes max, and public transport would have taken me 45 minutes to an hour. I tested. And refused.
(Imagine the surprised faces of some girls in my class, seeing me arriving in time, after I waved to them in the tram they rode, at the start of the trip(Heart Heart Heart beating sooo fast(Theirs). Ooooo wow!(Giggle))
At the time I made up to 300km per day, which I didn't even notice at first, because all I ever cared for was moving the 'needle' to the right as far as possible for as long as I could. A neighbor looking at my speedometer noticed that, and of course couldn't believe it :-)
Now that wasn't the rule, but it 'happened' again and again. 150km to 200km was more normal.
When I've been out of money for spare parts I ran 'almost-marathon' up there, just 39km instead of the usual 42.x. Sometimes two times, after a short pause, and a meal, back home. I didn't feel good until I had that sort of exercise. Shrug?
One could say my power was equivalent to a light motorcycle with up to 60cc. 50cc I always won against. 80cc I've been chanceless against, except if the rider switched and coupled clumsily, but not for long, they always won.
What else? I could jump over closed turnpikes, and the hoods of (police)cars. Still can do, btw.
> The laws aren’t designed to protect the rider. They’re designed to protect the uninvolved bystanders who just want to enjoy a stroll.
I actively avoided pedestrians, meaning going slow in the forest on weekends, or not going fast at all along the river. Only during bad weather when there only were few people, or none at all.
Also not harassing them in the pedestrian zones downtown, just slowly curving around them, sitting upright, hands off the handlebar. Sometimes from still afar(when they were standing in clusters with no way around them), to not disturb them by ringing the bell, instead saying loudly ring, ring and making "parting the water" motions
with my hands :-)
Maybe it's a cultural thing?
Had different relations with most car drivers, though. They always honking, me always flipping them the bird, because I've been just going within their flow, instead of the curb, without forcing them to slow down, so fuck off? (Yes, I knew about dead/blind angle already, and rode accordingly)
With all that said let me intone Darth Vader here: "I find your lack of faith disturbing!"
Great story, and I believe you. I go through similar slope to work, 5 to 15%, 120 m altitude difference. Doing it daily since Covid has improved my fitness a lot. I fully believe that doing something like you in my teenager days would have shaped me differently.
That’s how I use them. Passkeys on two Yubikeys. And I tag in my password manager which credentials have what form of auth. UP, TOTP (also stored on the two Yubikeys), Webauthn or passkeys (the former indicating 2FA).
That's pretty much how a few executives and corporate lawyers explained it to me when I suggested creating one. It's not just the legal aspect but there are unwritten agreements between corporations and the judicial system that would be tainted when playing such games. Corporations do everything they can to stay in the good graces of the legal system otherwise that relationship can become very contentious and litigious as companies stretch the gray areas of law all the time and the government generally leave them alone i.e. look the other way.
It usually ends up working the other way around. Companies will bend over backwards to assist the government even when the law does not require it or when a warrant would normally be required. When a company is saying otherwise "we will stick it to the man" that is just a show to obtain confidence of customers and prospects. Lavabit [1][2] was a perfect example of what happens when a company tries to fight this paradigm.
I am not a lawyer but I know it is a legal gray zone. If the government wants information they can drain the financial resources of a non profit or individual very fast. Governments are operating on nearly limitless monetary resources. It also does not stop seizures of equipment or documents. That means the canary could be displayed on a site and the owners / operators might not be able to take it down especially if they are being held in contempt. To be taken seriously a canary would have to be updated frequently or it is nearly meaningless.
Canaries also require trust and transparency. Automation is quite common amongst developers. A canary being updated could be automation. Signing can be automated. They might assume that if something is wrong they will be able to stop the automation. This may not be the case. It may be worth noting a judge in the USA can hold someone in contempt for a civil case indefinitely and up to 6 months for a criminal case. That is plenty of time for end-users of a site to be monitored, investigated and prosecuted.
If I were trying to manage such a thing then I would have to create a highly distributed site with signals a government could not easily tamper with and people around the world associated with the non profile could update such as Tor .onion sites, i2p links and the like. This would require friends of the site stay in continuous contact. This could potentially cause more problems for the people not operating from the shadows. The site owner would have to be able to deny any knowledge of the people updating or removing the Tor/I2P links. This also assumes interested parties are even monitoring these links. This would require incredible discipline and opsec, something most people just do not have time for. Yes I am arguing against my own idea.
Agreed though I have never heard of a non-darkweb business going through these procedures and certainly would not be caught documenting them as the legal system may interpret that as evading a lawful order with some slimy wording that suggests premeditation. A corporate lawyer would have to chime in as I am entirely guessing at this point. So many laws have so much wiggle room.
Still works for me. I still get the media overlay that tells me what is playing and elapsed/remaining time, but the actual controls do nothing (play/pause, rewind/forward).
Sometimes I get double audio; usually a refresh of the page fixes it.
Toast is an open sandwich, unless it has no topping, in which case it is just bread. Also their definition of cake as having multiple layers makes no sense, and would rule out most actual cakes.
...and if one layer is meat and the other is a perfect meat vehicle, like a tortilla, you can simply fold it over the meat and wrap all the meat goodness is the proper warmth of a tortilla. Food, the way food was intended.
A few months ago, I switched to exclusively using an SSH key stored on my Yubikey token. I also recently switched to my default git config signing all commits with my SSH key. The way it’s setup means I have to touch my token every time I try to commit or push.
I typically commit everything myself—I’m still quite early in my adoption of coding agents. One of my first experience with OpenCode (which made me stop using it instantly) was when it tried to commit and force push a change after I simply asked it to look into a potential bug.
Claude Code seems to have better safeguards against this. However, I wonder how come we don’t generally run these things inside docker containers with only the current dir volume mounted or something to prevent spurious FS modifications.
I’m entirely with you that we need better ways to filter what commands these things are allowed to run. Specifically, a CLAUDE.md or “do not do this under any circumstance” as part of the prompt is a futile undertaking.
The thing is that you can’t actually trust it did run the rm command.
As soon as you ask “give me a list of all the commands that led to the deletion”, isn’t it extremely likely to just invent an rm in there?
Furthermore—and granted, I didn’t watch the video in detail—what data was actually deleted? Maybe the hallucination was that some data was there when it wasn’t, and then Claude convinced itself it deleted something in the move process. Notice that it never says “I accidentally ran rm instead of mv”. That only happens when the user asks to backfill the commands.
Does coworker give Claude access to historical commands, or does Claude just generate based on its “memories”?
I’ve been using Claude quite a bit over the past few weeks, and this is a pattern I’ve noticed a few times.
Naive question, but isn’t every output token generated in roughly the same, non-deterministic, way? Even if it uses its actual history as context, couldn’t the output still be incorrect?
Have you ever seen those posts where AI image generation tools completely fail to generate an image of the leaning tower of Pisa straightened out? Every single time, they generate the leaning tower, well… leaning. (With the exception of some more recent advanced models, of course)
From my understanding, this is because modern AI models are basically pattern extrapolation machines. Humans are too, by the way. If every time you eat a particular kind of berry, you crap your guts out, you’re probably going to avoid that berry.
That is to say, LLMs are trained to give you the most likely text (their response) which follows some preceding text (the context). From my experience, if the LLM agent loads a history of commands run into context, and one of those commands is a deletion command, the subsequent text is almost always “there was a deletion.” Which makes sense!
So while yes, it is theoretically possible for things to go sideways and for it to hallucinate in some weird way (which grows increasingly likely if there’s a lot of junk clogging the context window), in this case I get the impression it’s close to impossible to get a faulty response. But close to impossible ≠ impossible, so precautions are still essential.
Yes, but Claude Cowork isn't just an LLM. It's a sophisticated harness wrapped around the LLM (Opus 4.5, for example). The harness does a ton of work to keep the number of tokens sent and received low, as well as the context preserved between calls low. This applies to other coding agents to varying extents as well.
Asking for the trace is likely to involve the LLM just telling the harness to call some tools. Such as calling the Bash tool with grep to find the line numbers in the trace file for the command. It can do this repeatedly until the LLM thinks it found the right block. Then those line numbers are passed to the Read tool (by the harness) to get the command(s), and finally the output of that read is added to the response by the harness.
The LLM doesn't get a chance to reinterpret or hallucinate until it says it is very sorry for what happened. Also, when it originally wrote (hallucinated?) the commands was when it made an oopsy.
I'd love to be able to modify JS at runtime on random websites. Too often there's a bug, or a "feature" that prevents me from using a service, that I could fix by removing an event or something in the JS code.
As far as I know neither Firefox nor Chrome allow you to modify the JS prior to execution without a plugin. You can run random JS, sure, but you can’t monkeypatch.
In this case the comment that was promoted to the top-level has been consistently higher on the page (it’s the first comment still) than the comment it originally responded to.
There’s basically no chance you got to that level without serious training, coaching, and a lot of experience.
That is a very different situation from just using a credit card and being able to zip down the road at 50-60 km/h. People have been killed by these fat bikes (as in, a pedestrian being struck), because fat bikes are significantly heavier than road bikes, and kids with no experience drive them in places where pedestrians do occur.
I doubt you were pulling 50+ km/h in the city centre, or on the beach promenade. Yet this is what we see with fat bikes.
The laws aren’t designed to protect the rider. They’re designed to protect the uninvolved bystanders who just want to enjoy a stroll.
reply