I think it is clear that planning before coding is generally a good idea - it's true for human developers and even more so for AI agents. The question now is more: what is the right way for an AI agent to form a plan and share it with the user? Maybe the dedicated plan mode in the agent harness is no longer needed, but I would like to make very sure that the AI agent has a plan upfront and that this plan is compliant with my requirements, my general expectations and that it is consistent and clear.
I have zero interest in manually creating software for solved problems - no text editors, no games, no utility tools. I like to write software for new problems and I love to solve these problems using code. I am not keen on typing source code by hand. If an LLM helps me speed up code creation, I will happily use it, because it allows me to solve my (business) problem faster. We are talking about efficiency here.
If other people like writing code by hand and enjoy it, good for them. But don't complain that your hobbyist approach is unsuited for your career.
Although I read that this web site was made using AI, I still feel entertained by it. The topic requires precise language and that's something that an LLM is surely excelling at.
Yet I feel that the concepts in the text are a bit repetitive; the write-up could have been shortened a bit.
Consent-O-Matic is a bit of an edge case because their last update was more than a year ago and their Github repo looks pretty dead. This makes me not really trust it anymore.
What do you think all those IT engineers who were made redundant in the last months are doing? Some of them might prefer founding a start-up to joblessness. It's not the first time that a job market crisis increases the rate of start-up registrations.
I haven't lost faith in my career and I've been an IT specialist all my life. AI will elevate my game to a new level. I see zero evidence that it will make me redundant in any way. I guess the role of IT will shift - specialized tech companies are less important, while business use of IT will thrive. (Yes, the code-monkey style of work will disappear, but nobody is going to miss this anyway, I hope)
It's always good to see improvements around WSL2, but especially this one is not so relevant IMHO, since it only affects WLS2 file access to Windows file system. If you store your dev environment in WSL2 anyway, this won't help you.
I am trying to build a simulation that lets a simulated organism come up with its own small language, purely learned from sensory input: https://github.com/JoergStrebel/VirtualZoo/blob/main/compute...
I would like to implement the ideas put forward by Stevan Harnad in his symbol grounding problem paper (Harnad, 1990).
Banking apps are the deal-breaker for me. I only do business with banks that offer alternative ways of securing transactions e.g. eTan / ChipTAN / PhotoTAN with a separate reader / generator (see https://www.bsi.bund.de/EN/Themen/Verbraucherinnen-und-Verbr...). This is probably a pretty European thing to do, but at least it avoids being locked in and being tracked.
I also love it. Finally, I am no longer constrained by syntax errors or forgotten API details. I can focus on the feature. It's like taking programming to a higher level - programming in English (instead of Java).
How does your framework compare to spec-driven development e.g. https://github.com/github/spec-kit? In my experience, spec-kit produces a lot of markdown files and little source code.
Well, in a world of finite resources, I think I would need a better reason to invest time into this topic than just "for the challenge". I mean I just think that I have ample opportunities to do something more sensible with my time.
Climbing a mountain at least gives you bragging rights; I don't think a bootable floppy disk is impressing anyone these days.
A very nice video. It shows that computer games are glamorous on the outside, but once you look behind the scenes, they just look like normal software. I was also surprised to hear that the team did not only rely on computer graphics textbook algorithms, but built their own pathfinding algorithm in a pragmatic manner.
Ok, impressive, but - why?
No current computer has a floppy disk drive anymore.
The Web Page claims building such a disk is a learning exercise, but the knowledge offered is pretty arcane, even for regular Linux users.
Is this pure nostalgia?
Lefties sympathizing with criminals, sharing their wealth distribution fantasies, agitating against competing political views.
You've come a long way, CCC!
The initial ideas was political, but with a clear focus on freedom of information, and the power to govern your own personal data.
In all fairness: human senior devs see AI-written source code with some disdain, as it usually does not match their stylistic and idiomatic preferences (although being correct and fully working).
I don't think that untested code is the problem here - you can easily measure test coverage and of course. every CI/CD pipeline should run the existing unit and integration tests.
I am certain that LLMs can help you with judgment calls as well. I spent the last month tinkering with spec-driven development of a new Web app and I must say, the LLM was very helpful in identifying design issues in my requirements document and actively suggested sensible improvements. I did not agree to all of them, but the conversation around high-level technical design decisions was very interesting and fruitful (e.g. cache use, architectural patterns, trade-offs between speed and higher level of abstraction).
The publicly funded media (radio, TV) obviously use this finding to claim that they need more money and/or a tighter regulation of AI companies' products. Sounds a bit self-serving to me...
If you want to do research, usually the first thing to do is refine your research question up to a point where it becomes relatable to the scientific state of the art and where it becomes clear how to test / evaluate it.
I don't think you are there yet.
I think this "argument" has always been flawed. I don't need to justify what information I would like to share especially with state agencies. In Germany, this is even encoded in a legal principle called "Informationelle Selbstbestimmung" (informational agency). It's not about the information, it's about my right to decide about sharing it.
Impressive setup, but I would assume it to be very operations-intensive because of the high number of deployed components and their complex configuration. Plus, if you are serious about self-hosting, you would need the facilities and infrastructure to deploy it: server rack, redundant power supply, smoke detectors, fire extinguisher... I would never let my PC-grade hardware run unsupervised in my home.
And if I understood correctly, you would still have to have some server on the Internet for running your Headscale VPN, so you need your own dedicated Internet connection - ADSL, dial-up, cable modem would not be enough.
I think there is a misunderstanding - the whole point of my comment was that LLMs are lacking sensory input which could link the neural activations to real-world objects and thus provide a grounding of their computations.
I agree with you that purely symbolic AI systems had severe limitations (just think of those expert systems of the past), but the direction must not only go towards higher-level symbolic provers but also towards lower level sensory data integration.