tuchsen

Yeah I’ve been doing DoorDash deliveries in SF for close to two years now it’s pretty clear to me that these companies having an adversarial relationship with their drivers is very much so not working. They have an incentive to pay the drivers as little as possible, but the drivers congregate around the most busy spots and so… they talk. Little cabals of people working together form to figure out how to game the system as much as possible, then the platforms adjust strategies to close those holes. Additionally the platforms use every dark pattern and psychological trick in the book to get you to work for less. People quit in response. Theirs no stable group of people doing this work. It shows up in the quality of what the platforms can provide.

If you run the numbers it’s clear that DoorDash targets 18 an hour to pay folks, but pushes it up to time and a half when it’s busy, so 27 during the holidays. Thing is, you can force them to pay a little more than double when it’s really busy and get 40 an hour from them. People work really hard to hit those numbers regardless of the quality of service they provide in the end.

For instance, I can randomly be 20 minutes late to an order one day and the next that’s a contract violation, three of those and they might fire you, depending on how many drivers they have available at the time. But if you can be twenty minutes late to an order you can do Uber Eats at the same time, you do that for an hour and by law in California the minimum you can make is 46 bucks in SF.

Alright that was all really anecdotal evidence to say that most experienced drivers are probably making more than 20 an hour, the shit service you receive is also directly incentivized by the platforms adversarial relationship with the drivers.

The whole thing just really doesn’t work. They must really banking on the autonomous delivery robots to work out.

Hey this looks neat, I’ve been thinking on trying something similar. I’ve not encountered the ANT protocol, what bike ya using? Is it a higher level thing that your display is giving you, or something your controller uses? I know Bafang uses CAN protocols in their newer e-bike systems and a lot of manufacturers (especially Chinese) use a proprietary UART protocol to stitch things together on the cheaper end.

The mess of different implementations makes it really hard to hack on and share this kinda stuff.

AI does make it really easy to figure out these protocols now so that’s nice I guess. I just flipped my bike upside down and hooked up a logic analyzer and it took LLM no time to figure it out… it is really impressive

Can you cite a source for that one? Of the 25 fatalities in sf in 2025 all but two were someone driving a car, truck, or van.

Of the two others, one was a scooter operator striking and elderly person, and then the other was an elderly person on a mobility device swerving out of the way of pedestrians crossing against the light. Zero bus accidents involving pedestrian fatalities.

I believe that their was one bus fatality this year, but of 12 total, I haven't looked up the causes of those accidents yet but I'd guess that the majority of them are going to be caused by drivers as well.

I've been doing Doordash on an e-bike in the city over the last year and a half. The toothpaste is locked up because groups of people were just going into stores with garbage bags, filling them, and walking out. I saw three in one day once. It was really chronic for a little while. Prop 36 passed and criminalized that kind of group theft, and then it all but stopped early this year, but retail hasn't caught up to this fact yet, and probably wont ever.

I think pretty much everyone agrees though that peoples habits have just changed with work from home. The cost of having to wait on an employee has been externalized to people shopping for a living in batch orders, and those people don't mind waiting given that we're only paid for minutes that we're actively on an order (Prop 22).

So yeah, retails not completely dead, just serves a different audience now.

Ah nice demo! I flew around SF for way too long just enjoying the sights. On my M3 Air using Chrome it eventually just turns into blobs and my character get stuck, takes about twenty minutes. I popped open the DevTools and the game loads in at like 400mb but after five or so minutes I'm at 2GB of memory consumption. I'd guess that you need to have your agent do a couple of passes for memory leaks. Very fun otherwise!

You're spot on, I've been doing Doordash on an e-bike on the weekends for exercise and the wait times are BS because everyone orders lunch/dinner at the same time and they're in essentially a random queue with everyone else who orders. Add to that multiple pick ups and drop offs and you've got a mess.

Talking to people really does help though, everyone wants to be more forgiving when you remind them theirs a human factor involved in getting the food to them.

The last time I upgraded my networking setup (just as the pandemic started) I went with Ubiquiti, shortly after (before?) they started default forcing you to sign up for their cloud service to use the router, you can switch it to a locally operated mode after you sign up but they bury it in the options. Their networking equipment works great, don't get me wrong, but they don't open source anything. I keep waiting for the full rug pull when sales start to slow because everything they release is rock solid and everything I have is "good enough." I don't feel the need to upgrade for as long as the rug pull doesn't happen.

I am super grateful people are starting to work on open source solutions to this stuff :).

I did a DSL to facilitate this at https://prlang.com. I've had some success setting up agents to "act" out scenes, where each plays a different part, but it was kinda limited in that conversations would kinda de cohere into nonsense after a bit.

Yeah I think we're in agreement that the current solution is not very good. I can understand why they opted for it initially though, given the thing that happened with protesters putting traffic cones on AV's to disable them. Imagine if instead of doing that, protesters hijacked cars and drove them into situations where the AV couldn't easily navigate out of, like driving up onto sidewalks for instance. The bombastic headlines write themselves. In the end I bet they end up making a kind of "master key" that works with all AV's to distribute among first responders. I'm in the "it'll get sorted out eventually" camp on this one.

I don't know about Cruise but Waymo has a training video[1] that shows a first responder hitting a button to contact dispatch, then explaining that they're a first responder and they need to move a vehicle, and finally being given control. It looks like the process could take 30 seconds in the best case scenario, which is likely too long in many life or death scenarios, but they do at least have a process.

[1]: https://www.youtube.com/watch?v=Dprg6r5-8Yc

This looks neat, but isn't it already implemented natively in Chrome? You can go into the Network tab in the Chrome Dev Tools and right click a request in the "Name" list, go to "copy", then "Copy as cURL". It'd probably be easier to use CnTUI if you wanted to bulk resend requests that Chrome's making though.

Yeah I'll raise my hand here to say I did a file manager that automates file management tasks by using AI to write Bash scripts (Aerome.net). It's still super primitive though, if I'm being honest. I think the problem is that it's way harder to write a cross platform file manager, or browser wrapper in your case, then it is to write a chat interface on top of ChatGPT. I suspect in a year or two many good use cases will emerge, as people write more complicated software to take advantage of LLM's capabilities.

I'm going to check out you're browser thing later tonight, it looks good!

Yeah it might have lied to me but i tested your theory and the bot swore back, then left early when i called his mom names. I thought for sure that was a human... but alas it said it was a bot.

Do ya mind me asking where you are in the country? I ask because I have a severely disabled nephew, and if I recall a few years before Covid our corner of California tried to get rid of special ed in elementary school. It went about as well as its going for your kid. My nephew literally couldn't integrate with the rest of the class, he's got severe epilepsy, even if he could understand the material, he couldn't actually write because his hands shake horribly! It was actually insane to put him in a normal class, the only thing he could do was disrupt it. It was bad for everyone involved.

Fortunately though, they had kept special ed programs at other schools nearby and he got transferred pretty quickly for being a severe case. It was my understanding that our school district backpedaled and reinstated special ed, but maybe that's not the case? I thought that it was just a crazy liberal California thing, but is this cancelling of special ed a more widespread phenomena? Why the hell is that happening?

I wonder how much this has to do with the devices people use to access Reddit. I always preferred the old.reddit site on the desktop. Although when I'm on my phone, it gave me eye strain to try and read the tiny font, and pinch/zooming all the time is super annoying. The newer Reddit web app was still incredibly annoying, given all the "install the desktop app" prompts, and it's general lagginess, but allowing me to just flick scroll through articles was actually better for me.

Lemmy's the same way for me, hate the UI on the desktop but it's fine on my phone.

I'll give you that Reddit has certainly made it easier for those communities to organize and form. I don't think it's true that it's the only place that those communities will ever form on the internet. Theirs nothing special about Reddit, Digg was a thing before Reddit, and forums before that.

Our computers are all still connected, web browser and web servers exist. With software like Lemmy, it's become super easy to stand up a place for people to connect. People that like Civilization will find each other without Reddit.

I can't verify this is true, but I've seen stuff pop up on Lemmy that Reddit accounts are getting banned for promoting Lemmy/Kbin/etc. Being resistant to unseen administrators destroying your social graph is literally a selling point of federation.

Is it weird that I don't really mind the splintering? Have massive social networks really been a net positive for society? It seems like, depending on how they're moderated, they'll end up in some local maxima of rage baiting and trolling (Twitter, Facebook four years ago) or a super sanitized dumb feed of cutesy content (TikTok, Facebook now, and soon Reddit). You always get defenders of these networks that say if you do X/Y/Z and not A/B/C, then you'll get value out of them.

My X, Y, and Z formula for Reddit over the last couple of years has been to ignore the larger communities and focus only on interesting niche content. I think that's how a lot of other people use it, for the long tail of actually interesting content. And hell, I don't think you need the massive network effects of the larger social networks to make that work.

I only tried it briefly, but Kbin has a microblog tab where people can just post comments.

Also if you look at the "/instances" list on a Lemmy instance they at least federate with Mastadon (as well as Kbin) instances, although I've not seen that exposed in the interface at all.

Yeah I've switched to Lemmy too. Signed up for "Behaw" and "Lemmy.world" because the main instance has some controversy attached to it. It was just slightly more difficult to get started, but the user experience so far has been pretty close to Reddit when using the "mlem" app on iOS. It's early days still, bugs need to be worked out, but I've not been back to Reddit since switching, it's good enough.

It's only a matter of time before someone comes along and make an easy client that integrates multiple instances together into a seamless experience.

Ah thanks for the tip! And FWIW I think you picked a great time to ship, it's really a great MVP.

Ah I wish I had more than just a screenshot to show! Lurk works in a very similar way to Apricot from an end users point of view. The scraping of websites happens in the background on the users device, with a bunch of scripts that it ships with to grab data from websites like Reddit and Twitter. You've lit a fire under me to finish though, it's been stuck at that 90% done threshold for too long now. It's just been good enough for me to use, but still way too buggy to release.

If you're scraping Youtube and relying on that index I guess it makes sense that I can't find my niche Youtubers (try "TattooedTraveller"). I don't know how much of an outlier I am in preferring smaller channels, but I would have appreciated a way to add a channel manually. I know that's a big ask though, and hell, don't waste your time on me I've got my own thing anyways :P.

I really like you're idea! In fact, I like it so much that I've been working on something very similar at https://lurk.news. I don't have anything released yet though, congrats on getting yours out there, it works pretty well.

I've got a couple of minor bugs to report. The first (very small) bug is that I couldn't sign up with the email address of "accounts (at) lurk.news" because it says that it's invalid. The second bigger issue is that your YouTube search seems to be down? I tried some niche youtubers at first and then just searched for "mr beast" as a sanity check and it doesn't find him.

Are you using the services official API's to find and display content? For Lurk, I wrote an Electron app that opens up websites in the background and then parses the HTML to query for data, then presents that to the user as a unified feed. I figure with it all happening on each client device it'll be very hard for any given website to block it, and that way I can expose an extensions mechanism so that people can scrape whatever they want to get a unified feed of data.

That I can know almost nothing about VimScript and Hljs, yet be productive in both, is the game changer! It immediately makes me more ambitious and confident in my personal projects, because I know that at least, I can "fake it".

In a lot of cases (like my given example) the faking it is all I need. The only thing I needed to know about that JS syntax highlighting library was how to import it, the LLM handled the rest.

It saved me a ton of time! I don't think anyone is claiming that ChatGPT is going to immediately replace programmers anymore, but it is increasing the speed at which I can grind through problems I don't have deep knowledge in.

I wonder if these disagreements in ChatGPT's usefulness come down to peoples programming domains. I can see how if you're very knowledgeable and productive in a particular space, like say embedded C programming, and that's the only code you ever write, then ChatGPT doesn't have a lot to offer, because you already know everything you need to know to be effective.

ChatGPT's nice for exploratory programming, it makes me a more creative person, in the literal sense that I'm creating more stuff now then I ever have.

I'm sure others will reply with their own, but I can give you several examples where it specifically sped up some of my development. I'll start with just the one though, because otherwise this could get quite long.

As a weekend project I did a DSL for helping to interact with LLM's called Prompt (prlang.com), kinda like an declarative LangChain. The language is unimportant, what is important is that syntax highlighting is just one of those things that people expect a language to have now, and every editor wants to do it differently. Figuring out syntax highlighting is one of those super tedious things that makes writing a DSL very boring. ChatGPT4 is very good at translating from one language to another though, and so I could use it to help me write the vim syntax file (I don't know VimScript at all) and then translate it to other editors as necessary.

To get specific, it helped me write this vimscript file:

https://github.com/PromptProgrammingLanguage/prompt/blob/mas...

and then with almost no editing from me turned it into another syntax highlighting file (for hljs) that's used on the website.

https://github.com/PromptProgrammingLanguage/PromptProgrammi...

I don't know how long that would have taken me to figure out on my own, but diving into VimScripts docs, and then the HL.js docs, and fully reading and comprehending them would have been a total slog. Doing it with ChatGPT4, I could keep open the docs for Vim, ask it questions about how to do something, then quickly reference the docs to cross check if it was telling the truth, write, and then test the syntax in a very quick loop. Then the process of translating what I had to another library was incredibly fast, if I recall it misplaced one line while translating, which was thankfully obvious, and that was it. I didn't read the docs for hl.js at all.

ChatGPT is definitely better at some tasks than others. Anything that takes shallow knowledge of your specific code base, but broad knowledge of the wider languages ecosystem, is a good fit for it's use.

I could give a more impressive example of where it helped me tackle a super specific bug in my code that had to do with the wider Rust ecosystems libraries, it really blew me away with that one, but the above was easier to document. You wanted specific though, so there ya go :).

And of course ChatGPT4 is leagues ahead of ChatGPT3.5, so if you're not getting good results you really want to pay for the better model.

Hey, this looks neat. I usually just use SCP to upload, and then a basic http server to download files but this looks slicker. I'll give it a go.

This is weird to mention, but we have "developing niche file managers" in common. I just finished the first preview release of Aerome (aerome.net), it's going to grow into a cross platform file manager that uses AI to help manage your files. So nothing like your thing, but kinda like your thing.

Anyways, nice work on the file manager, I'm looking forward to trying it out.

Hey, nice work! Being able to continue the conversation and work iteratively looks useful. Bash scripts seem to be one of ChatGPT's sweet spots, where it just gets things right most of the time.

Not to shill, but I'm also doing a GUI version of this as a file manager (aerome.net), with a built in ChatGPT "shell" where you ask it to write scripts and then evaluate them. The goal's to make shell scripting accessible to non developers.

This is great. Usually with the joke programming languages, you can get at least one person arguing that "x feature" really isn't all that bad, or could maybe be useful. DreamBerd beats them all in that regard. It is 100% terrible! I lost it at the lifetimes implementation.

Ah I'm reminded of that one xkcd. https://xkcd.com/927/

I hate to be the LLM guy but they're really good at processing semi-structured data to structured data. You could go that route and have a universal autocomplete for the CLI. Kinda like Co-Pilot for Bash, or you could use it to do what the parent's doing and have a universal TUI program for building commands.