2026-10-06 01:32:49
Git is used everywhere in software development. And this very blog is also stored in Git. Obviously, I use Git a lot. I even have a regularly updated blog post with various tips and tricks.
Git uses SHA-1 hashes to address contents it stores. The most important word in this sentence is address. Cause the aformentioned hash is not intended to provide security or verification.
The upcoming Git 3.0 is going to switch the default for commit hashes from SHA-1 to SHA-256, the newer more secure algorithm. It is being done to improve the security, however there is no real attack vector to abuse the weakness of the SHA-1 algorithm. And there are other ways to improve the security, for example, by introducing additional content hashes, not used for addressing.
The planned switch would create a dilemma for the whole community. The repository can have only one addressing, either with SHA-1 or SHA-256.
If we believe that SHA-256 is better (whatever it means) we must convert our repositories to the newer standard and rehash all objects (files, commits, etc.) But this operation by itself can be insecure and there are no tools to validate the conversion. This also would render all existing links, for example to GitHub, obsolete and make the repository incompatible with older versions of tools.
And if we won’t convert the existing repositories, it could seem as though they are no longer secure.
Git 3.0 may repeat the mistakes of Python 3.
2026-10-05 04:25:35
Since I moved to Munich in 2011 I haven’t missed the largest folk festival in the world — Oktoberfest. Two events in 2020 and 2021 were cancelled due to the pandemic. So today was my 15th time there. And this one has seen the highest attendance in history, there have been more than 7.4 million guests.
But I personally am no longer excited about it.
The beer price has doubled, the beer itself hasn’t changed. And as I no longer drink alcohol at all, it makes it even worse being there in a tent with thousands of drunk people. And the preliminary statistics report shows that the beer consumption continues to grow. The good thing is that, according to the official report, the non-alcoholic beer share grows year over year.
The food there has never been great anyway, and its price has doubled as well.
I remember fun times going for a ride on the Olympia roller coaster, but this isn’t that thrilling anymore. Probably, I’ve been to such attractions too many times before.
And for the sake of tradition, it would be great that the festival would be smaller and more folk, rather than being permanently overcrowded.
I respect the Oktoberfest’s contribution to Munich’s budget, but I’m very likely skip it next year.
(All photos are made in 2011 for my Instagram)
2026-10-02 15:23:57
I use LLMs as much as I can for software development. Bug fixing, analysis, various automations. But there’s one thing no one should use AI for: writing texts to be read by humans.
I learned it the hard way. No, I have never generated a full multipage blog post from a single prompt and posted it without reading. Would never do that.
I just used a single phrase suggested by an AI agent.
That was enough to trigger AI slop detector many of us now have. People considered the whole article generated, despite it being me who typed every word there except that one phrase.
So, the basic rule is simple: never accept a word suggested by AI. Use it only to correct grammar. Be careful with punctuation though, AI tends to overuse commas, for example.
I’m not alone in this stand against generated text, see more tips in “How To Write With LLMs” published on sockpuppet.org.
This text was typed by me.
2026-10-02 04:11:11
Marcin Wichary has posted today an article “The left hand should know what the right hand is doing” on his blog Unsung. It shows several scenarios across different apps and websites, in which one part of the app suggested something useless or impossible, while the other should already have been aware that it makes no sense.
I immediately remembered two similar cases.
This one is maybe borderline, but I suppose not many people decide to buy a second refrigerator immediately after they have bought one.
The same is also valid for Amazon’s selection of similar products, where they often show duplicate things, which people usually buy just a single item.
This is my favorite and it repeats every other Friday for me.
I often start the day by listening to the Release Radar playlist or going directly to the New Releases section and playing new albums. Some of them I have pre-saved, but it doesn’t trigger a notification when I first open the app on the release day.
It’s only in the afternoon when Spotify finally reminds me about the album I’ve already listened.
2026-09-29 01:03:36
I’ve seen three blog posts about major performance improvements recently. Two of them are from Cloudflare: How we saved 100 terabytes of memory by optimizing 1.1.1.1’s DNS cache and Saving another 100TB of RAM with math (and Rust).
The third comes from Anthropic: How we made claude.ai 3x faster in two weeks. And they couldn’t be more different.
Starting with the title, which in the case of both Cloudflare’s posts gives a (pretty big) hint on the area of the performance improvement. They go deeper and explain in detail, what and how they optimized, explaining memory layouts and alignments, cache strategies, and so on. Overall, it is an excellent read, from which everyone can learn something. Or at least be amused by the scale of Cloudflare operations, where every byte saved on some structures stored in memory can lead to gigabytes and terabytes of savings across their datacenters.
Anthropic’s blog post has a very different goal. You wouldn’t learn much from it. Instead, you will repeatedly hear how they tasked Claude (without disclosing any prompts) and how it improved something (without disclosing exact solutions or results).
What worked best from the marketing perspective? A SEO-optimized post mentioning Claude in every other sentence or a real optimization story where one actually learn how the system operates?
For me, the answer is pretty clear.
2026-09-27 02:16:57
The Steve Jobs Archive has published the second book after the Make Something Wonderful, a collection of Steve Jobs’ speeches, letters, and interviews.
I enjoyed that book and expected the same level of inspiration here as well. But Letters to a Young Creator didn’t keep up with the promise.
It features some big names, but only few of their letters worth mentioning or quoting. Here are some:
…don’t do the right thing, avoid it like the plague. So, what am I suggesting you do, the wrong thing? Of course not. Do the brave thing.
—Lee Clow
Many people are precious about their ideas. But ideas are cheap. It’s good ideas that are hard to find.
—David Kelley
“The days of thoughtless design for thoughtless consumption are over”, I also wrote that years ago. Unfortunately, this wish has not yet been fulfilled. I pass on my wish to you. Do less, but better things.
—Dieter Rams
“Do what you think is right and treat everything as a learning experience.”
—Alice Rawsthorn
The crazy ones are sometimes the ones who do crazy impossible things.
—Larry Brilliant
If you can learn to harness your fear—to let it push you forward instead of holding you back—it’s incredibly powerful.
—Jimmy Iovine
Don’t think about building a career; think about making something every day.
—Mona Simpson
Besides the quoted letters, I’d recommend the ones from Samuel Ross and Jon M. Chu.
★★★☆☆ short, and worth it only when you have nothing else
