Hacker Timesnew | past | comments | ask | show | jobs | submit | TekMol's commentslogin

    AI isn’t replacing me. Like a toddler, it
    needs to be constantly coached.
Like a toddler, it will grow up.

Humans are really bad at noticing trajectories. They see the current situation. They know what the situation was 5 years ago. But for some reason they do not believe that there is a trajectory. They view the present state as the final destination.


Sure, just like AI enthusiasts seem to be unfamiliar with the concept of local maxima...

It’s been basically the same for 3 years now. Are you sure we’re the ones who can’t see trends?

Your experiences must be much different from mine.

Three years ago, AI was barely able to provide sort-of reliable command completion.

Two years ago, it could extrapolate a single function from a docstring - but the docstring had to be so verbose that it wasn't practical to use in that way.

A year ago, I was tinkering with Devin to try to find a way to get it to reliably implement small, isolated features from verbose Jira tickets.

Six months ago, I started using AI to generate the majority of my code output. Most of my time was spent reviewing, and I was ecstatic to reach ~2x output because I could run the next task while reviewing the last.

Now, at work I'm managing a half dozen Claude Code instances, Devin sessions, and orchestrating a review loop between Claude, Devin, and CodeRabbit. It's not uncommon for me to be working on four or more discrete features at once. My output is approximately 15x my pre-AI baseline - and I've not sat down and written a line of code directly in six months.

At home I'm managing a Hermes agent that can spin up a whole fleet of purpose-tuned agents for whatever purpose I'd like. I've implemented spec-driven development a'la Acai, and extended it to the point that my agent creates specs from text or voice conversation, I review them, and it handles implementation end-to-end. The code itself is an almost disposable artifact - useful primarily to ensure no regressions have been introduced between rounds.

... I simply don't understand how you can assert that "it's been basically the same for 3 years". It absolutely has not.


It sounds like our experiences are different. My software work isn’t on products where code can be disposable, since it affects people’s lives in material ways. I’m not sure why you’re launching fleets of agents at home, either.

Cmon - cursor has been out for like 3.5 years at this point. AI was still in its infancy but it was definitely able to complete tasks, albeit smaller ones.

Not disputing the overall trajectory, yeah it’s gotten better. But it was definitely capable of more than just command completion 3 years ago.

I reach for it more frequently. But personally, it’s at the point of diminishing returns for my work. It’s capable enough now to handle most of the things I want to throw at it, sometimes it’s wrong, sometimes it’s right.

I’m not doing cutting edge deep tech work - and I also don’t have the motivation (or salary increase) to be 15X more productive, if that’s even measurable. We are so busy because the CEO hears these “15X” statements and then the pressure is on to match or exceed that, and I’m not playing that game.


> Like a toddler, it will grow up. Humans are really bad at noticing trajectories.

Yourself included??


head in the sand

For an open source project, is there any reason to still accept code contributions?

Feature requests are valuable because they tell you what users want.

Error reports are valuable because they tell you under which circumstances the code fails.

But the code that implements those features and fixes those errors can now be written by AI. AI follows all the rules for how code is supposed to be written in your project. Is already producing very high quality code. And soon it will produce a quality that no human can match.


It was alluded to in the post - contributors turn into maintainers. Someone who contributes has a small but plausible chance of sticking around.

For an open source project that isn't a business, that's really the only way to recruit people


But why recruit people, now that we have AI?

Couldn't an agent monitor feature requests and bug reports, reason about them, and then implement and fix the ones it deems important?


For the same reason software engineers still employed. AI is not yet capable of autonomous software development. From my perspective we're nowhere close.

The status page says it is fixed, but the server I host with them is still down. It even says so when I log in. Not available "Due to a current error".


Hi there, Please use your customer account to create a support request. Document the issue you are facing in as much detail as you can, and our team will do their best to help you. --Katie (Hetzner)


Would running Codex in a container on its own fix such vulnurabilities?


No, that is not how ELO scores work.


As far as I understand, this is exactly how ELO scores work. If a more capable show up and starts beating all the other models, it literally takes ELO points from everyone else.

https://en.wikipedia.org/wiki/Elo_rating_system


    If a more capable show up and starts
    beating all the other models
There is an instance of this in the chart. In 2025-06-24 when Gemini-2.5-pro shows up. As you can see, the ELO of the others do not drop.


Depends on the test design; is an agent competing against other agent in a given match, or against a test? Plus! Does the test's ELO fluctuate?


It's a fitted Bradley Terry model, scaled to familiar Elo scores, anchored to wins against Mixtral-8x7B at 1114 (at least last time I looked at it). When you fit the model against historical data, and then you add another month of time that contains newer models, the relative strength of a given model might decline even if its absolute ability remained fixed.


Yes, that is in fact how Elo can work[0]. There are quite many ways Elo systems can work.

[0]: https://en.wikipedia.org/wiki/Elo_rating_system


It depends what you use as an anchor. If the anchor is a fixed model, you’re right. If the anchor is updated to a better model over time, then the elo of historical models degrades, right?


Here is my perspective as someone who has not started 3d-printing yet, but is interested to give it a try:

I'm a confused about the whole "3D printer sends prints to its manufacturer's server" issue. Because I wouldn't want to connect hardware device like a 3D-printer to a network in the first place.

Can I buy a Bambu Lab printer and just never hook it up to any network?

Will I be able to print from sd-card just fine?

Can I update the firmware from an sd-card?

If these two are possible, I would not have any problems with such a device. If they are not, I would not even think about getting such a device.

And when it comes to slicing software: Can I use any slicing software and all I have to do is load the hardware info of the Bambu Lab printer I want to use? Or do I have to use Bambu Lab Studio or a fork like Orca Slicer for some reason?

And while we are at it: Does command line slicing software exist? I wouldn't want to dabble with a GUI. I would want to define the parameters of a print job in a yaml or json file and then slice it like "./slice.sh config.yaml myobject.stl"


You hook it up to the network to use monitoring features like a remote camera view to make sure your print is going smoothly.

SD cards work but it's extremely less convenient than just printing straight from the slicer.

You can use any slicer you want but Bambu wants only their slicer to directly connect.

CLI slicing is not something you want in general. Visual confirmation of the toolpaths is very important to making prints as successful as possible.


> You can use any slicer you want but Bambu > wants only their slicer to directly connect.

Here it looks like you can connect Orca?

https://www.youtube.com/watch?v=PMtkIGf8xOs&t=240s


Only on LAN mode, which disables remote monitoring.


Could it be that this is a reasonable protection so that the average user is not being spied on via cameras in the printers when their computer gets compromised?

I guess you can do remote monitoring via Orca when you set the printer to developer mode?


I still do not support IPv6 on my servers and I think I will skip it and wait for IPv8:

https://www.ietf.org/archive/id/draft-thain-ipv8-00.html

Avoiding a dual-stack and making IPv4 a part of whatever superseeds it seems like the right choice to me.

IPv6 always seemed to me like throwing away all existing telephone numbers, just to support longer numbers.


  ::203.0.113.42 (tunnels to 203.0.113.42 over v4)
  64:ff9b::203.0.113.42 (translates to v4 at nearest NAT64 point)
  ::ffff:203.0.113.42 (opens a v4 connection via an AF_INET6 socket)
What are these then?


Maybe you didn't read the draft. It's either a delayed April's Fool joke or AI slop. Even if you take it seriously, it will require updating every v4 device in existence.


I wish there was a system that lets users put up a donation that is released once a specific bug is fixed or a specific feature is implemented.

Wouldn't that be cool? The company would have a list of tasks with a dollar amount next to it.

I for one have been dabbling with a bug in ThunderBird for days now that drives me mad:

I recently created a folder in Thunderbird and called it "archive". No way would I have expected that this will lead me to a bug and will take hours out of my day: There seems to be no way to get rid of this folder anymore.

Things I have tried:

"Keep message archives in" in "Copies and Folders" is disabled. I tried temporarily enabling it, setting it to some other dir and disabling it again, that did not help.

I have disabled it in "subscribe".

I cannot rename it.

There is no "archive" folder in the web interface of my email provider, so if it Thunderbird somehow created it on the server, there seems to be no way to see, let alone delete it again in the web interface.

I tried deleting archive.msf on disk. That makes the folder disappear after the next start, but it is recreated after about a second.

I deleted folderTree.json and folderCache.json, that did not help.


You can do that. It's called a restricted donation. If you make a donation with a cover letter or a check memoizing a specific purpose and the nonprofit accepts it, then by law they're legally obligated to follow through and use that money for that purpose. With bugs it's probably easier because you can just write the bug ID on the check.


MZLA Technologies, the organization that these donations go to, is not a non-profit.


There are also a couple of bug bounty websites out there for exactly this kind of thing: you and others throw some money into the pot for fixing a given bug or implementing some feature, and coders can claim that bounty once they've written the code.

I've seen a few of these sites over the years but I can't remember the name of any RN. Search engines are your friend.


Swift reminds me a lot of Flash back in the day.

While the Flash guys had to use a native development environment and compile their stuff, I could just edit JavaScript in a plain text file and hit reload.

20 years later, and some of the same friends now swear by Swift. And have to use a native development environment and compile their stuff. While I still prefer to just edit JavaScript in a plain text file and hit reload.


Given that Swift has one of the slowest compilation times, this is a valid criticism.


I only skimmed the article, but the proposed solution seems to be that the authority (the "issuer") sends data to a device the user owns but has no control over. Like an Android or iOS phone.

The data is of such form that the phone then can pass challenges of type "are you of at least x years old" without giving out any other information.

And the user cannot share that data with other users because their phone will not let them.


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: