Hmm, assuming that the AI bubble might pop a little bit after the upcoming IPOs, maybe it's better not to call yourself an AI company then?
2,535 karma · joined September 10, 2010
Gergely Orosz did an interview with them in 2024:
https://newsletter.pragmaticengineer.com/p/twisting-the-rule...
> "Cast iron and carbon steel have nearly identical thermal conductivity (~52 W/m·K), which surprises most people."
is unsourced. And the precise "~52" is quite misleading - Wikipedia and online sources report thermal conductivities in the range of ~30-50¹.
Also:
> "Critically, whipped egg white foam drainage follows a hyperbolic saturation curve: v = V × t / (B + t), where V is the maximum drained volume and B is the drainage half-life (Lomakina & Mikova, 2006)."
As far as I can tell, the article² (cited twice for this claim) does not contain any equations modelling drainage over time, and especially not this equation or the term "hyperbolic".
So, it seems that you cannot really trust the sources the author's LLM included. For me, that means that I cannot trust any of the other claims in the article (or the author in general).
¹) https://en.wikipedia.org/wiki/List_of_thermal_conductivities
²) https://www.agriculturejournals.cz/pdfs/cjf/2006/03/02.pdf
> " Ovalbumin coagulates irreversibly at 80°C (Weijers et al., 2003), permanently setting the foam structure.",
and the paper by Weijers et al. says:
> "A strong temperature dependence on the reaction rate was observed. At 80°C, half of the protein was denatured and aggregated in less than 2 min (half-time, th), while at 68.5°C this took approximately 6 h."
So, the citation is generally true-ish although a little bit imprecise.
At which temperature range Ovalbumin coagulates seems quite irrelevant for the whole article, however. To me it's unnecessary fluff, others might like that kind of detail.
(This does not imply anything regarding the article as a whole - it's just one thing I checked.)
My own experience with flipped classrooms (which seems to be shared by quite a few people who have tried it out): they only work well if all students actually read/watch the materials beforehand. In small, advanced courses, intrinsic motivation may be sufficient - but in most cases you need some extrinsic coercion - such as a mandatory quiz about the materials or hand-written lecture notes that need to be shown at each in-person session.
With AI, some people don't watch the lectures but let ChatGPT give them a summary which they submit. Then these people poison your in-person session with their lack of knowledge and motivation.
Not sure what the issue between them and Adafruit is. However, people over on Reddit¹ claim that Flux.ai is a little bit scummy. They push users into a beginner trial ($5/month) and then silently charge for usage per token - up to $100 per month.
Oh, they also claim that they have "the world's largest community-driven public library of Adafruit products, including footprints, symbols, datasheets, and simulation models"². I wonder whether they designed these themselves or whether they use existing ones. Could not easily find licenses info.
¹) https://www.reddit.com/r/PCB/comments/18o5zfo/thoughts_on_fl...
(I'm not sure whether I agree with the commentator that a SE license would be that helpful in practice.)
(It is obvious that one cannot directly derive the exact input - but one can derive potential inputs and then use other means to find the exact one.)
Netmux (2019): Hash Crack: Password Cracking Manual¹
James Leyte-Vidal (2024): Ethical Password Cracking: decode passwords using John the Ripper, hashcat, and advanced methods for password breaking²
Daniel W. Dieterle (2024): Password Cracking with Kali Linux³
¹) https://www.amazon.com/gp/product/1793458618
²) https://www.amazon.com/Ethical-Password-Cracking-passwords-a...
³) https://www.oreilly.com/library/view/password-cracking-with/...
> "Then there is also the fact that having a fully-fledged graphical desktop environment running in the background at all times is not quite optimal to say the least. 99 percent of the time when cracking passwords, you will be staring at a black terminal filled with white text, so using Windows, which is especially GUI-heavy, is usually impractical unless you are specifically testing something or showcasing some process."
I am reasonably sure that the Windows UI has rather little practical effect on hashcat's speed, and this thread implies the same: https://hashcat.net/forum/archive/index.php?thread-8958.html Also, 99 percent of the time when cracking passwords, I am not staring at a black terminal filled with white text.
(I am generally taking it a little bit personally when the author directly addresses me and tells me what I am probably thinking or doing.)
> "Behind a hash function are a series of complicated mathematical operations that make deriving the input from the output literally impossible."
I'd argue that the mathematical operations themselves are usually not that complicated. More importantly, the whole book seems to be about ways to derive the (probable) input of a hash function from the output. It is not literally impossible.
> "It is important to note, however, that hash functions are not truly random;"
As the author writes elsewhere, hash functions are deterministic and not random at all. Calling them not truly random seems to imply that they are somewhat random.
> "When encrypting a file or any kind of data with AES for example, the program leveraging AES will prompt you for a password. Yes, a password."
Yes, this is a book about password cracking, but there are lots of cases where programs use AES with a computer-generated key and won't prompt you for a password. E.g., TLS.
(Just to reiterate: I am not trying to diminish the author's work, I wanted to suggest ways for improvement. I might be wrong or overly pedantic.)
This book is currently not really relevant for me, so I just skimmed the samples on Amazon. I found the technical content to be reasonably accurate and interesting although sometimes a little bit verbose (e.g., the section about 'what is a password') or slightly imprecise. In general, I think this book might have benefited from a thorough copyediting pass. There are quite a few grammar errors and unpolished sentences in the book, e.g.:
> The reason why Linux is imperative is that well, for one, most of the tools we will use, while indeed have builds for other systems, like Windows, in this book we will work with Linux.
Wishing you success and keep on writing!
I originally assumed that the bulb would be somehow connected to voltmeter and amperemeter.
¹) I just found out that it is more commonly called 'ammeter' in English - which is so unintuitive that I prefere 'amperemeter'.
As I don't know much about the CTF scene, I looked for other takes on this topic.
Here's an article from 2015 about how tool-assistance already changed CTFs:
> Individual skill will undoubtedly be a factor next year. But, I'm left wondering whether next year's DEFCON CTF will tell us anything more than how well-developed each team's tools are (and how well they can interpret the results).
https://fuzyll.com/2015/ctf-is-dead-long-live-ctf/
But there are quite a few recent (2026) articles with the same core message as in the original article, e.g., https://blog.includesecurity.com/2026/04/ctfs-in-the-ai-era/ or https://k3ng.xyz/blog/ctf-is-dead
And here's someone explaining how Claude Max allowed them to win CTFs:
> I had always been interested in CTF as one of the only ways people could compete and show off their skill in coding/problem solving on a global scale. It was just too difficult and didn't make sense for me to learn the fundamentals as an electrical engineer. As time went on, I got better and better, and it was hard to tell whether it was because of experience or if it was because of improvements in AI.
> I accomplished my goals, and for that reason I'm quitting CTF, at least for now. [...] I'd like to think I highlighted the problem before it became a bigger issue. So, how do we fix this? Teams and challenge authors losing motivation is not good. CTF dying is not good. AI bad. Or is it?
https://blog.krauq.com/post/ctf-is-dying-because-of-ai
The only article that saw LLMs as a non-negative force for CTFs was this one. Fittingly, it sounds like LLM output ("Let's be honest", "This is where things get interesting.") and only contains hallucinated references.
Also featured in the Starfire vision video from 1992: https://youtu.be/jhe1DFY-SsQ?t=286
Many figures seem to be either missing key information (e.g. Fig. 5: the elliptical deformation is not shown - a human artist would have created a very different figure to explain the concept) or plain wrong (Fig. 6: the threaded rollers have the wrong orientation, Fig. 7: the ball is much too large for the bearing and the whole figure seems nonsensical at first glance).
And if the author did not spot these obvious problems with the figures, they either have no clue, accept sloppy work, or didn't even read the article they generated. That article is not really good advertising for the company's products.
(That the link behind the author's name leads to their Wikipedia article which seems to be a revised copy of the CV on their website is interesting, too.)
> "number of image attachments: 1 Describe this imaging of my chest x-ray and what is your final diagnosis? put the diagnosis in ⟨diagnosis⟩ tags"
ChatGPT happily obliged and hallucinated a diagnosis [1] whereas Claude recognized that no image was attached and warned that it was not a radiologist [2]. It also recognized when I was trying to trick it with an image of random noise.
[1] https://chatgpt.com/share/69f7ce8f-62d0-83eb-963c-9e1e684dd1...
[2] https://claude.ai/share/34190c8a-9269-44a1-99af-c6dec0443b64
Another thing: I realized how much I hate waiting for Claude to finish its thing. With UI designs, a quick interaction loop between tool and user feels much more important than with code.
Meta: your comment was marked [dead], like a few other constructive comments I saw in recent days. Not sure why.
I think this list is not only used for internal audits but also to assure the public and banks that Germany indeed knows in detail where its gold is stored.
¹) https://www.bundesbank.de/de/aufgaben/themen/bundesbank-schl... (in German)
¹) https://de.wikipedia.org/wiki/Deutsche_Goldreserven#Diskussi...
https://www.bundesbank.de/resource/blob/743058/9869caef634ce... - Federal Reserve Bank of New York starts at page 2016 (PDF:2019)
See also this issue from 2025 where the developers responded: https://gitlab.opencode.de/bmi/eudi-wallet/wallet-developmen...
AFAICT, there is no mention of an Apple or Google account being required in general - the documentation just lists "signals" that are used to securely authenticate a person - such as Google's/Apple's security ecosystems. I am not sure what this means in practice. Can anybody with deeper understanding explain the actual implications and possible outcomes?
(Note: BMI is the German Federal Ministry for the Interior)
As others already mentioned: the article first sets out to strip away all culture-specific aspects of a clock but then retains the concept of seconds, minutes and hours. This makes some sense, of course, and is implicitly mentioned as "Calibration to a timescale". However, I would have liked a discussion of alternative ways how this could have been achieved and why the 12/24-hour system is used in the end.
Another nitpick:
> 4. A readable mapping. Humans don't read raw oscillations. We read "14:37:09" or "sunset in 42 minutes." Reading time is always a translation layer.
> That fourth point matters more than most people think. A sundial reads apparent solar time, which is local and visibly tied to the Sun. But apparent solar time is not uniform.
This argument - which is used to introduce the next section - makes no sense. A sundial has a readable mapping. (And I would argue that it might be the most non-culturally-defined, useful type of clock for humans).
Next, the author introduces "layers of time" that do not actually seem to be layers but different views (though partially layered).
Then we are given a number of requirements for the implementation - e.g., "no daylight savings time". The clock at the top is adjusted for local daylight savings time, however.
Finally, large parts read as if an LLM has written them (many cases of "Not X, Y", a full screen page explaining that n*2+1 is more than twice as large as n, a discussion of the irrelevant test suite, ...). [^1]
[^1] the few references to other sources have apparently been copied without changing the source markdown to hyperlinks. I think a human author would notice this when checking the rendered article, no?
Website: https://sailfishos.org/
Main forum: https://forum.sailfishos.org/
Recently on HN: https://news.ycombinator.com/item?id=47216037 / https://news.ycombinator.com/item?id=47311456 / https://news.ycombinator.com/item?id=41749296
Redacting text seems to actually work. However, editing existing text results in both the original text and the edited version being shown in the PDF after download.
(The page downloads mupdf (WASM) for rendering the PDF. When "downloading" (= saving) the PDF, the page first checks whether the allowed three downloads have been reached via a POST request (no PDF data uploaded), then it downloads PyIodide and some Python wheels (pdfrw, defusedxml) before creating the PDF file.