florianherrengt

4 ms·

1.
by florianherrengt 28d ago
Works with GeForce RTX 20 series and newer, RTX PRO (Turing and newer), DGX Spark and Apple M4 or newer. It does not pool memory or split one inference request across machines. Adding machines increases parallel throughput but won't le
2.
Nvidia Personal AI Router (Pair)(nvidia.com)
4 points by florianherrengt 28d ago|1 comments
3.
DeepSeek v4 Flash Vision Exp is now open-weight(huggingface.co)
2 points by florianherrengt 1mo ago|0 comments
4.
by florianherrengt 2mo ago
I’m a big fan of simple things but I always go back to using a library or framework because it’s much easier to create a good, consistent UX
5.
Chain-of-Thought Reasoning in the Wild Is Not Always Faithful (2025)(arxiv.org)
66 points by florianherrengt 2mo ago|38 comments
6.
by florianherrengt 2mo ago
This paper puts words to something I’ve noticed repeatedly with LLMs, particularly Qwen3.6. When I read its reasoning, it appears to recognise the mistake and then carry on as if it hadn’t noticed it at all. > models often determine thei
7.
by florianherrengt 2mo ago
> While a human may say “aha” to indicate exactly a sudden internal state change, this interpretation is unwarranted for models which do not have any such internal state, and which on the next forward pass will only differ from the pre-a
8.
by florianherrengt 2mo ago
I was hoping for Fairphone support but they said it's not happening. > Fairphones lack the updates and hardware-based security features expected by GrapheneOS. https://x.com/GrapheneOS/status/20899372466191
9.
by florianherrengt 2mo ago
> PostgreSQL Replacing Your Microservice I've done that before and the code was a mess. It works at the beginning but APIs do much more than piping data from the database. When you start dealing with ACL, external calls, code reuse,
10.
by florianherrengt 2mo ago
I have a Pinephone Pro. It's nowhere near usable as a daily driver.
11.
by florianherrengt 2mo ago
I'm surprised it took them that long. I always found it odd that they would support Google phones only.
12.
by florianherrengt 2mo ago
It means the company is measuring the wrong thing. I would much rather have someone on my team who ships less but whose work I can trust than someone much faster whose changes leave me wondering what problems we’re going to discover later.
13.
by florianherrengt 2mo ago
Then I’m not sure what you mean by calling them a bottleneck. If the limiting factor is the rate at which experienced engineers can understand and validate changes, you have three options: generate less, find a genuinely better way to valid
14.
by florianherrengt 2mo ago
Corrupted or lost customer data, security vulnerabilities, incorrect financial transactions, regulatory breaches, production outages, irreversible migrations, important product change effectively impossible without a full rewrite. Every fea
15.
by florianherrengt 2mo ago
Thanks for letting me know. I had no idea. I’ll follow up with the moderators.
16.
by florianherrengt 2mo ago
The difficulty of producing code WAS one of the circuit breakers. We had tests, CI, code review, QA, architectural reviews, etc. None of those disappeared. But they were designed for a world where producing such a large amount of change was
17.
by florianherrengt 2mo ago
What you’ve described solves a different problem. You’ve put a lot of effort into making sure generated code conforms to your standards. That’s all good. But do the people still understand what the system is doing and why? You can have code
18.
by florianherrengt 2mo ago
I tend to find that this perspective comes from people working on relatively small or isolated projects. On a large system, the customer being happy today isn’t enough. You need other engineers to be able to understand the system. Have you
19.
by florianherrengt 2mo ago
For the vast majority of 25K-line PRs, especially AI-generated ones, I would bet money that the person opening it does not actually understand all of it. If you genuinely understand the resulting system, then that's fine. That’s not th
20.
by florianherrengt 2mo ago
They’ll care when nothing works, nobody seems able to fix it, building new features takes forever and every change breaks something somewhere else. This was already happening before. But now, a lot more companies that previously might have
21.
by florianherrengt 2mo ago
I do have a genuine question though: what value are you adding at that point? If you don’t understand the system, don’t know what the code is doing and you’re mostly prompting an LLM to make the decisions and implementation for you, what ex
22.
by florianherrengt 2mo ago
I’m hardly a Luddite and you’re basically making my point. You’re deliberately constraining the output so changes stay small, human-reviewable and reversible. That’s good engineering culture. It isn’t all or nothing. I use AI heavily but th
23.
by florianherrengt 2mo ago
This is extrapolating several steps beyond anything we actually know. Whether people will buy AI-generated software is a different question. The problem I’m describing already exists inside businesses with paying customers. I’m talking abou
24.
by florianherrengt 2mo ago
Debt is debt. Some of it is absolutely worth taking on. I don’t have any issue with intentional debt when you understand the trade-off and have a clear payoff plan.
25.
by florianherrengt 2mo ago
The industry doesn’t really owe people opportunities. Even if you think it should, that’s not how companies are going to behave.
26.
by florianherrengt 2mo ago
If the people trying to understand the changes and guard quality are now considered the bottleneck, the answer isn’t that the reviewers need to move faster.
27.
by florianherrengt 2mo ago
I wrote about this before. I don't think the vibecoders will have a job for a long time. At best, it will be minimum wage. See "AI Won't Take Your Job, Someone Using AI Will" > There's no lucrative middle ground
28.
by florianherrengt 2mo ago
I use AI heavily, so this isn’t about refusing to change. You still have to understand the change yourself if you’re going to take responsibility for approving it. That work is fundamentally much slower than generating the code, even with t
29.
by florianherrengt 2mo ago
Agreed. But it’s the difference between crashing at 30 km/h (19 mph) and crashing at 200 km/h (124 mph). AI massively increases the speed and scale at which bad engineers can do damage while fixing it is still slow, difficult work
30.
by florianherrengt 2mo ago
The fact that a machine can do something better than humans doesn’t automatically make learning it pointless. We still teach arithmetic and algebra despite computers being vastly better at calculation. We teach spelling, grammar and essay w

More ›