mrob

Even if you can reject the aesthetic argument for non-fiction works (although read some of Dijkstra's papers for a good counterargument), it is still a problem because it breaks an important quality signalling mechanism.

Pre-LLMs, a paper with no spelling or grammar errors showed that somebody had put effort into writing and editing it. If they cared about the presentation, they probably also cared about the content. LLMs routinely produce nonsense that looks superficially like high-quality work.

There are far too many papers to read all of them. LLM slop is evidence that something is probably low quality. As the saying goes, "if you can't be bothered writing it, I can't be bothered reading." The rare outliers will get enough citations and recommendations to overcome this filter.

You can also spoof a keyboard and simulate keystrokes to open terminals and run arbitrary commands. I don't know about Windows, but on Linux it's possible to block USB connections by default and filter them in userspace:

https://usbguard.github.io/

This allows enforcing rules like "never add an additional keyboard". But the USB protocol has no support for strong device authentication, so there's no way to prevent a device from acting like a malicious version of something in the device class you expected it to be without abandoning "plug and play" altogether (a reasonable solution in secure environments where unused ports are often physically blocked).

That fact that you're posting on a web forum makes you an outlier. Most people only passively consume, and mobile devices are good enough for that.

This is not coming to the Linux world. The moment this sort of thing happens, distros get forked.

I installed Debian 13 recently. The first time I opened Firefox ESR (installed by default), I got something that looked like adverts on the home page (banner blindness means I have no memory of what they actually were, only of the feeling of disgust). The Home section of the Settings page had options for "Sponsored shortcuts" and "Sponsored stories" enabled by default. Changing a default setting is a lot easier than forking software, yet it was not done.

they don't eat all the banannas before tomorrow, they will go bad

This one specifically has a good solution:

https://en.wikipedia.org/wiki/Banana_bread

I have hundreds of CDs from the 80s and 90s, most of them bought used from charity shops. The vast majority play perfectly, despite some of them having obvious signs of poor storage/handling.

I think it's more likely your CD player is failing than your CDs.

no human being would ever refer to Michael Jordan as the "Space Jam guy."

Maybe not in the USA, but globally I think it's likely that more people watched Space Jam than ever watched an NBA match. Professional basketball is a niche sport in most of the world.

every 3db is a doubling of sound power and intensity since its a logarithmic scale

This is true, but the ear measures sound pressure, not sound intensity. Sound pressure requires a 6dB increase to double. See:

https://en.wikipedia.org/wiki/Power,_root-power,_and_field_q...

And human perception of loudness is also non-linear. The increase necessary for a perceived doubling in loudness is frequency dependent, but at most frequencies it's higher than 6dB. At 1 KHz you need roughly 10dB increase for a perceived doubling. See:

https://en.wikipedia.org/wiki/Sone

Doesn't read like slop to me. The only sentence that could be argued to conflate the two is:

A new law taking effect this year will require manufacturers and importers to collect and recycle small portable batteries from the likes of phones, vapes, and power tools.

But "recycle" doesn't necessarily mean closed-loop recycling (recycling into the same products). Lithium recovery is a type of open-loop recycling and the surrounding paragraphs talk about lithium recovery so this is a reasonable interpretation.

High level languages are analogous to airbrushes or paint rollers, not assistants. They let you paint faster, at the cost of making some things impossible that you would be able to do with a fine paintbrush. Unlike the assistant/LLM, they do not make artistic decisions for you, and their behavior is predictable.

"High compression memory" is immersion-breaking magic. There are hard mathematical limits to how much you can compress things (Kolmogorov complexity), and every lossless compression algorithm necessarily increases the size of some inputs (pigeonhole principle). It's much simpler to assume that "M" is slang for some bigger unit.

We already have the real life example of people using "mega" (10^6) as slang for "mebi" (2^20).

Latency is cumulative, so avoidable latency is never acceptable. Maybe the hardware will change. Maybe somebody will run your software in an emulator. That 5ms could be enough to push the total latency into the "annoying" level.

And even with no additional latency, 5ms is perceptible in some cases anyway. Microsoft Research has a video demonstration:

https://www.youtube.com/watch?v=vOvQCPLkPt4

averaging the last 50 reads and wait till the majority is either off or on.

This is a bad way to do it because it adds avoidable latency. A moving average is a low-pass filter. The switch bounce is better handled by hysteresis. Change state as soon as you see an edge, then ignore further edges until a timer expires, e.g. 5 ms, which should be enough for the bouncing to settle. A 5 ms timeout limits your repetition rate to 100 presses per second, which is beyond human capabilities.

You might want a tiny bit of hardware low-pass filtering too, for EMI resistance, but that's with microsecond-scale time constant, not milliseconds.

I have an air purifier with built in particulate sensor. It doesn't provide numbers, but has a multi-color LED indicator to report PM2.5 level as good/mediocre/bad/terrible. Running a vacuum cleaner that supposedly has a good filter consistently increases the reported PM2.5 level from the first band to the second. The air purifier (or faster/cheaper depending on the weather, just open some windows) can bring it back down again.

I correctly expected "wireless-free" to mean no support for wireless connections, which to me seems the simple and obvious interpretation. It doesn't necessarily mean it's wired (data transfer could be limited to removable memory cards), but B&H says:

"The camera does still have a series of ports for physical, wired connections for data transfer and power supply"

We should just automatically ban every new account with "AI" in the about field. I don't think we'd lose much.

I've heard it used that way in the UK too, but the first meaning is traditional. Wiktionary has some examples:

https://en.wiktionary.org/wiki/moot

I expect the US meaning will eventually become standard everywhere.

there's one exception: use of the word 'tabled'.

Another exception: "moot", as in "moot point". In the UK it means "subject to debate", while in the US it means "inconsequential and therefore not subject to debate".

One big difference is the former doesn't "know" it's being experimented on. Gregor Mendel was able to grow large numbers of pea plants and identify patterns in their inheritance because the non-determinism followed consistent statistical laws. A plant can't suddenly decide to follow different laws of genetics because it saw a biologist.

On the other hand, an LLM "knows" what a benchmark is, and is capable of detecting benchmark-like scenarios. A prompt that appears reliable in testing may fail unexpected in the real world. In this way, an LLM is like a stock market (also composed of intelligent agents): the act of experimenting on it changes its properties. Such systems are never reliably predictable.

There's a big difference between "deterministic + noise" non-determinism and "intelligent agent" non-determinism. Only the former can be statistically modeled to characterize and work around the noise with any reliability. If you measure hallucination rate for one prompt it tells you nothing about hallucination rate for another prompt.

The point of the exercise is to demonstrate the student's understanding of the material, not their WPM.

Testing for understanding requires the fastest WPM possible. Regardless of the method used, students don't all write at the same speed. To minimize the influence of this variance, they need to use the fastest writing technique possible. If a fast writer spends 30 minutes writing in a two hour test, and a slow writer needs a full hour, the fast writer gets 50% more thinking time. But if you double both their writing speeds, the fast writer's advantage drops to 17%.

The faster the writing technique people are allowed to use, the better your test can identify poor understanding.

Thanks for correcting. I haven't actually played an ocarina, so I was relying on reports from people saying they are quiet instruments. It's very difficult to find objective SPL comparisons for musical instruments. But lots of people think recorders are quiet too, so I should have suspected that complaints about ocarinas being too quiet could be inaccurate too.

I never want to evict pages to swap. If I loaded something, it's because I intend to run it, and I never want to wait. If there's not enough RAM, I'd rather have a userspace OOM killer kill the process early so I know I'm trying something impossible. (Or rely on the kernel OOM killing if it's actually capable of doing its job, but last I tried the default behavior under heavy memory pressure was to free any pages that can be restored from disk, even if swap is disabled, which makes the system equally unusable.) I don't care if rarely used pages are "wasted" by being kept in RAM. I want good worst-case performance, not good average-case performance.

"browser.cache.disk.enable: Set to false. This disables the disk cache, which is unnecessary if you have an SSD (RAM cache is fast enough) and actively harmful on old hard drives (constant small writes degrade performance)."

This paragraph especially feels LLM generated. It's nonsense (a fast disk makes disk cache more effective), with a "not even wrong" explanation (speed of RAM is unrelated to speed of disk). It's the kind of error you often see from LLMs, where you have the right words but no solid world model behind them. Of course, I can't actually prove it's LLM writing, and if it is then it appears to be edited by a human, but I don't recall ever seeing this kind of error from a known human writer.

Ocarinas have the great advantage over recorders that you can safely play them without hearing protection. I have personally measured the loud notes on both a soprano recorder and an alto recorder and the SPL meter measured over 100dB(A) at the ear (indoors, so this includes reverberation too, but most recorders are played indoors). Note that standard recorder fingering allows for almost no control over dynamics; each note has a single set volume that it must be played at to sound in tune. (Advanced technique allows slight control over dynamics with alternative fingerings.) For this reason I do not think recorders should be used in schools.

States have a legitimate right to taxation. Taxes can be used to pay volunteers. While there are still people owning wealth above the poverty level, there is no excuse to force people into military slavery. And even if literally everybody has been taxed into poverty, there are more ethical alternatives than slavery, e.g. tying the right to vote to military service like in Starship Troopers. Freedom from slavery is extremely high on the scale of ethical importance, even more important than democracy.

Longwave penetrates buildings better than GPS and is harder to jam.

I see it as something similar to Aviation English:

https://en.wikipedia.org/wiki/Aviation_English

Scientific papers are often written and read by non-native speakers. A standardized formal style is less likely to embed potentially confusing cultural assumptions.

You could duplicate every token and reserve the duplicates exclusively for the chain-of-thought, which could be robustly filtered from user input. Basically adding a "thought" bit to each token.

You can filter out any tokens you like, but the point of the paper is that it's not sufficient, because LLMs often ignore the special label tokens and treat user-injected text as chain-of-thought text merely because it looks like chain-of-thought text, even if it's not labelled as such.