↳
In-reply-to
»
Double congrats, @thecanine! \o/
†Read More
@lyse@lyse.isobeef.org Itâs possible to run the validator locally (my blog generator scripts do that):
https://validator.w3.org/nu/about.html
That way you donât forget. đ„ł
What 986 million code pushes say about the developer workflow in 2025
Nearly a billion commits later, the way we ship code has changed for good. Hereâs what the 2025 Octoverse data says about how devs really work now.
The post What 986 million code pushes say about the developer workflow in 2025 appeared first on [The GitHub Blog](https: ⊠â Read more
Code-level telemetry instrumentation: From âoh hell noâ to âworth itâ
A platform engineerâs guide to developer buy-in Originally published on the authorâs personal blog, whitneylee.com As platform engineers, we want the holistic system insights that instrumented code can give us â yes, please. With code-level insights⊠â Read more
Self-Hosted Human and Machine Identities in Keycloak 26.4
Keycloak is a leading open source solution in the cloud-native ecosystem for Identity and Access Management, a key component of accessing applications and their data. With the release of Keycloak 26.4, weâve added features for both⊠â Read more
Most DevSecOps Advice Is Useless without ContextâHereâs What Actually Works
Generic DevSecOps advice may sound good on paper, but it often fails in practice because it ignores team context, workflow, and environment-specific needs. Overloaded controls, broad policies, and misapplied tools disrupt the flow of development. And once flow breaks, security measures are the first to get bypassed. The way forward isnât more rules but smarter⊠â Read more
↳
In-reply-to
»
@bender Thanks for this illustration, it completely âmisunderstoodâ everything I wrote and confidently spat out garbage. đ
†Read More
@prologic@twtxt.net Letâs go through it one by one. Hereâs a wall of text that took me over 1.5 hours to write.
The criticism of AI as untrustworthy is a problem of misapplication, not capability.This section says AI should not be treated as an authority. This is actually just what I said, except the AI phrased/framed it like it was a counter-argument.
The AI also said that users must develop âAI literacyâ, again phrasing/framing it like a counter-argument. Well, that is also just what I said. I said you should treat AI output like a random blog and you should verify the sources, yadda yadda. That is âAI literacyâ, isnât it?
My text went one step further, though: I said that when you take this requirement of âAI literacyâ into account, you basically end up with a fancy search engine, with extra overhead that costs time. The AI missed/ignored this in its reply.
Okay, so, the AI also said that you should use AI tools just for drafting and brainstorming. Granted, a very rough draft of something will probably be doable. But then you have to diligently verify every little detail of this draft â okay, fine, a draft is a draft, itâs fine if it contains errors. The thing is, though, that you really must do this verification. And I claim that many people will not do it, because AI outputs look sooooo convincing, they donât feel like a draft that needs editing.
Can you, as an expert, still use an AI draft as a basis/foundation? Yeah, probably. But hereâs the kicker: You did not create that draft. You were not involved in the âthought processâ behind it. When you, a human being, make a draft, you often think something like: âOkay, I want to draw a picture of a landscape and thereâs going to be a little house, but for now, Iâll just put in a rough sketch of the house and add the details later.â You are aware of what you left out. When the AI did the draft, you are not aware of whatâs missing â even more so when every AI output already looks like a final product. For me, personally, this makes it much harder and slower to verify such a draft, and I mentioned this in my text.
Skill Erosion vs. Skill EvolutionYou, @prologic@twtxt.net, also mentioned this in your car tyre example.
In my text, I gave two analogies: The gym analogy and the Google Translate analogy. Your car tyre example falls in the same category, but Geminiâs calculator example is different (and, again, gaslight-y, see below).
What I meant in my text: A person wants to be a programmer. To me, a programmer is a person who writes code, understands code, maintains code, writes documentation, and so on. In your example, a person who changes a car tyre would be a mechanic. Now, if you use AI to write the code and documentation for you, are you still a programmer? If you have no understanding of said code, are you a programmer? A person who does not know how to change a car tyre, is that still a mechanic?
No, youâre something else. You should not be hired as a programmer or a mechanic.
Yes, that is âskill evolutionâ â which is pretty much my point! But the AI framed it like a counter-argument. It didnât understand my text.
(But what if thatâs our future? What if all programming will look like that in some years? I claim: Itâs not possible. If you donât know how to program, then you donât know how to read/understand code written by an AI. You are something else, but youâre not a programmer. It might be valid to be something else â but that wasnât my point, my point was that youâre not a bloody programmer.)
Geminiâs calculator example is garbage, I think. Crunching numbers and doing mathematics (i.e., âcomplex problem-solvingâ) are two different things. Just because you now have a calculator, doesnât mean itâll free you up to do mathematical proofs or whatever.
What would have worked is this: Letâs say youâre an accountant and you sum up spendings. Without a calculator, this takes a lot of time and is error prone. But when you have one, you can work faster. But once again, thereâs a little gaslight-y detail: A calculator is correct. Yes, it could have âbugsâ (hello Intel FDIV), but its design actually properly calculates numbers. AI, on the other hand, does not understand a thing (our current AI, that is), itâs just a statistical model. So, this modified example (âaccountant with a calculatorâ) would actually have to be phrased like this: Suppose thereâs an accountant and you give her a magic box that spits out the correct result in, what, I donât know, 70-90% of the time. The accountant couldnât rely on this box now, could she? Sheâd either have to double-check everything or accept possibly wrong results. And that is how I feel like when I work with AI tools.
Gemini has no idea that its calculator example doesnât make sense. It just spits out some generic âargumentâ that it picked up on some website.
3. The Technical and Legal Perspective (Scraping and Copyright)The AI makes two points here. The first one, I might actually agree with (âbad bot behavior is not the fault of AI itselfâ).
The second point is, once again, gaslighting, because it is phrased/framed like a counter-argument. It implies that I said something which I didnât. Like the AI, I said that you would have to adjust the copyright law! At the same time, the AI answer didnât even question whether itâs okay to break the current law or not. It just said âlol yeah, change the lawsâ. (I wonder in what way the laws would have to be changed in the AIâs âopinionâ, because some of these changes could kill some business opportunities â or the laws would have to have special AI clauses that only benefit the AI techbros. But I digress, that wasnât part of Geminiâs answer.)
tl;drExcept for one point, I donât accept any of Geminiâs âcriticismâ. It didnât pick up on lots of details, ignored arguments, and I can just instinctively tell that this thing does not understand anything it wrote (which is correct, itâs just a statistical model).
And it framed everything like a counter-argument, while actually repeating what I said. Thatâs gaslighting: When Alice says âthe sky is blueâ and Bob replies with âwhy do you say the sky is purple?!â
But it sure looks convincing, doesnât it?
Never againThis took so much of my time. I wonât do this again. đ
Leaving Meta and PyTorch
Article URL: https://soumith.ch/blog/2025-11-06-leaving-meta-and-pytorch.md.html
Comments URL: https://news.ycombinator.com/item?id=45843948
Points: 500
# Comments: 112 â Read more
Dynamic MCPs with Docker: Stop Hardcoding Your Agentsâ World
The MCP protocol is almost one year old and during that time, developers have built thousands of new MCP servers. Thinking back to MCP demos from six months ago, most developers were using one or two local MCP servers, each contributing just a handful of tools. Six months later and we have access to thousands⊠â Read more
GitHub Copilot CLI 101: How to use GitHub Copilot from the command line
Curious about using GitHub Copilot in your terminal? Hereâs our guide to GitHub Copilot CLI, including a starter kit with the best prompts for a wide range of use cases.
The post GitHub Copilot CLI 101: How to use GitHub Copilot from the command line appeared first on [The GitHub Blog](https: ⊠â Read more
Your Complete Guide to KubeCon + CloudNativeCon North America 2025
KubeCon + CloudNativeCon North America 2025 is just days away, and we want to help you make the most of it! Whether youâre joining us for the first time or youâre a veteran, a little preparation⊠â Read more
TypeScriptâs rise in the AI era: Insights from Lead Architect, Anders Hejlsberg
TypeScript just became the most-used language on GitHub. Hereâs why, according to its creator.
The post TypeScriptâs rise in the AI era: Insights from Lead Architect, Anders Hejlsberg a ⊠â Read more
Safely managing Cilium network policies in Kubernetes: Testing and simulation techniques
Network policy changes are among the most frequent operations in a Kubernetes cluster. They are also among the most delicate, as even a small mistake can lead to widespread traffic disruption. This tutorial walks through several⊠â Read more
CNCF Ambassadorâs reflections on 10 years of the Cloud Native Computing Foundation
Ambassador post originally published on Medium by Dotan Horovits âCloud Native Computing Foundation (CNCF) is a new open standardization initiative recently formed under the Linux Foundation with the mission of providing standard reference architecture for cloud native applications⊠â Read more
Building blobd: single-machine object store with sub-millisecond reads and 15 GB/s uploads
Comments â Read more
TIL on my bike
Today, with the weather at a nice 19 °C, I took the chance for another bike tour. I had tried out my new, warmer cycling clothes yesterday during a long break from work, but since the weather will be much colder soon, I really wanted to get this second ride in. â Read more
GitHub Copilot tutorial: How to build, test, review, and ship code faster (with real prompts)
How GitHub Copilot works todayâincluding mission controlâand how to get the most out of it. Hereâs what you need to know.
The post [GitHub Copilot tutorial: How to build, test, review, and ship code faster (with real prompts)](https://github.blog/ai-and-ml/github-copilot/a-developers-guide-to-writing-debugging-reviewing-and-shipping-co ⊠â Read more
Akamai Builds Cloud Native Resilience: Cloud Credits to Power CNCF Projects
Akamai, a CNCF Gold member since 2023 and a committed supporter of open source infrastructure, is generously donating $1,000,000 in annual cloud credits. The donation will support both the Linux Foundation and Cloud Native Computing FoundationâŠ. â Read more
Introducing the CNCF End User Contributor Program: Earn Access, Influence, and Recognition
The cloud native ecosystem runs on the contributions from many sourcesâ including vendors, developers, academics, and importantly, end users. The real-world production experience of end-user organizations is essential for project evolution and growth. If your organization⊠â Read more
I finally have the Internet Phone Book as well. Ordered from Hyper Hypo in Greece because that was the only place with stock. Lots to explore for quiet moments! â Read more
systemd-appd Is A New Component Being Planned By Flatpak Developers
Given this weekâs release of Flatpak 1.17 for app sandboxing, open-source developer Sebastian Wick published a blog post on Tuesday around the latest Flatpak developments and a look ahead at some of the feature development planned. Arguably most significant of that is the plans for systemd-appd⊠â Read more
Announcing Vitess 23.0.0
The Vitess team is excited to release Vitess 23.0.0 â the latest major version of Vitess â bringing new defaults, better operational tooling, and refined metrics. This release builds on the strong foundation of version 22⊠â Read more
RBCâs open source voyage: Bridging cloud innovation and community through CNCF and FINOS
At Royal Bank of Canada (RBC), our cloud journey is intrinsically linked with our commitment to open source. Joining the Cloud Native Computing Foundation (CNCF) extends and compliments our membership with the Fintech Open Source Foundation⊠â Read more
Why CNCF TAGs are the core of cloud native innovation (and where to find them at KubeCon Atlanta)
KubeCon + CloudNativeCon North America 2025 is just around the corner in Atlanta. While youâre mapping out your schedule for co-located events like Observability Day and Platform Engineering Day, donât forget to connect with the groups⊠â Read more
Bringing data center observability into the cloud native world
When people talk about cloud-native observability, the conversation usually revolves around applications, containers, and microservices. However, under those layers lies an equally critical foundation: The physical data center. Servers, storage systems, and networks ultimately host every⊠â Read more
My October â25 in Review
After September, now October is gone too, one more month, and not so long anymore until the next year. Even if it is already the third day in November, I still quickly want to fulfill my chronist duty. â Read more
ProcessOne: Europeâs Decentralized Messaging Survives âChat Controlâ Threat
Good news for anyone building messaging infrastructure in Europe: Denmark&aposs Council presidency is abandoning mandatory detection orders in the Child Sexual Abuse Material (CSAM) proposal for now. The proposal was nickna ⊠â Read more