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