spatular

The trick is to redesign the bricks for worse tolerances. With 3D printers you can print very nuanced springy elements that are impossible to achieve with injection molding. I got some reasonable bricks years ago on cheap printers with PETG, should work even better now with modern printers and ABS.

There is also prompt processing that's compute-bound, and for agentic workflows it can matter more than tg, especially if the model is not of "thinking" type.

Yes, 140W sustained, 160W burst (~10 seconds).

Clickhouse is a network server, duckdb and polars are in-process databases. It's like postgres vs sqllite.

You can take a look at orange pi 5 plus. It has M.2 for WiFi and M.2 for SSD, both PCI-E. It's in $100+ category though.

Orange Pi 5, "plus" version also has 2gen 1-lane pci-e (M.2 wifi), and 3gen 4-lane pci-e (M.2 SSD) and 2x2.5Gbit ethernet.

8nm, pretty power efficient. I've measured it to run at 0.7A@5V idle and 1.2A@5V with all 8 cores loaded with md5sum /dev/zero; iirc it had 1 ethernet connected, no other periphery. Running on Armbian.

You'd need to spend at least 110kWh to convert CO2 back to propane, maybe much more. No matter what you do, LPG -> CO2 -> LPG cycle has to be energy-negative.

To be precise, that'd be called MPU -- memory protection unit.

MMU allows linux to provide a "flat" memory space to each userspace process. This memory space is assembled from 4KB pages that can be randomly dispersed throughout physical memory. Say, you want to malloc() 1MB of memory on a system that's been up for some time. Physical memory may not have a fitting continuos chunk of memory, but virtual memory of a newly created process will always have one, provided that there is enough free pages available.

In other words without MMU all programs share the same memory space, and if it get fragmented, generally you'd have to reboot. With MMU fragmentation can still be an issue, but it's greatly reduced, since each process has its own memory map. And if memory of the process gets framgmented, you can just restart it. If runtime supports object compaction/relocation, then fragmentation may be not an issue at all.

Re access to memory locations -- mostly any direct access to instruction or data memory in userspace program automatically goes through MMU remapping.

Sorry, I meant to write "voltage regulator" about which parent asked about, not sure how I ended up with "linear regulator".

I believe that without qualification "linear regulator" could be just anything that regulate voltage. Usually you can easily tell from schematics around it if it's some switching IC, an LDO or a module. And if you have layout or 3d render of the board it'd be almost certain.

Otherwise the person may just ask: "where are caps, or is this thing a module or what?"

I've yet to see a voltage regulator IC with built-in caps. DC-DC modules sure, but author argues that this kind of info (module / ic) should be present in schematics. BTW, some linear regulators can't handle ceramic and low-ESR capacitors at output, and may be better with none if that's the choise.

I think he meant that in 3-rd world countries iPhones are luxury items bought by rich people in position of power.

Well racism connotations permeat everything in US, it's overheated/taboo topic there. Not so much in the rest of the world, see two top dictionary definitions from google:

To punish (a person) without legal process or authority, especially by hanging, for a perceived offense or as an act of bigotry.

(Law) (tr) (of a mob) to punish (a person) for some supposed offence by hanging without a trial.

Nothing about racism here. The only thing that doesn't apply to Stallman is that he has not been hanged. That's quite important one. Though words are sometimes used not in their literary meaning (like "I'm killing it" -- well hopefully not).

BTW from what I've gathered this entire thing happened because Stallman was pedantic and has written something along "let's be precise with words, I think what happened was not rape, but statuatory rape because of such and such".

Let's see: - judged by angry mob -- yes, - no ability to defend oneself -- yes, - killed -- no (though seeing polarization and level of anger here it's possible that he'll be eventually physically assulted), - person demolished, life ruined -- yes (Stallman got removed from the org he created, got fired from his work. Another poster here written that his friend got accused, defended in court with hard evidence, life is still ruined).

Comparison with lynching seems fair to me.

All of this got me thinking about creating, you know, accepting tech community where everyone is free to express their own opinions. I'm personally fine with my code being called shit and me being called dumbass for a good measure, just tell me what you really think and we'll sort things out faster instead sugar-coating stuff to not hurt my ego.

Um, I'm actually worried that I or other person will be punished for something that they didn't do. In the third world country I live in you'd have to accuse someone and then bribe the judge. Turns out that in US one don't even have to bribe anyone, mob will lynch them. You just need to accuse them of right things; e.g. stealing and homicide won't do. Is that the progress that humanity has made?

I don't have US cultural background and it looks insane to me, about level of witch hunts or lynch mobs. Like one accusation of peadophilia is enough to ruin a life, and lives of others that dare to voice opinions counter to that. Same thing with rape. It doesn't matter if it happened or not. Court of law? Nah, why? Clearly if they are accused then they are guilty.

I haven't tried it, but it should be doable. A lot of linux-based rust libs wrap C libs and link to them, or compile their own embedded version during build. Same thing here too, though there could be some problems regarding what C compiler is used and where it expects to find its libs.

AFAIK LM3S6965 is used in the book as a target for QEMU, and no one endorses to use the chip in actual hw.

Absolutely this. The same price and the Bluepill has much more powerfull CPU.

The problem is HAL support. Some periphery is supported, other is not. ADC and DMAs for UARTs were merged just recently. Stuff like SPI / I2C slave modes is missing, I think HAL interface is not fleshed out yet. Stuff like usb stack would probably take a lot time to complete.

For me imaginary input has less details and exists in a separate plane, and it never mixed up with real senses, in the sence that I always know which is which and to my knowledge I've never mixed them up when awake (only in sleep).

Visualizations are "ghostly", have few details, like 0.1 - 10% of what I usually see. Detail level depends on amount of attention allocated to internal imagery. Picture is always moving, changing, in flow. It takes considerable effort to pin a scene and stabilize it though it increases resolution. I've asked around and some people report having sequence of static, very high-resolution, almost real life quality pictures, either from real experience or synthesized. And images flip, replacing one another. Other people say that they have very fast low-res flowing "movies" like me, often consisting of several independent streams in different "parts" of awareness, typically with only one if focus. It's not that imaginary stream is overlayed on real sensory stream, it's more like they are different kinds of perception, like seeing and hearing is, though visual imagery is closer to seeing because objects it deals with have similar perceptive qualities, like having spatial component and colors.

In my case imagined sounds are more detailed. When I'm remembering or simulating a conversation, voices are rather close to that of real people. Music can be recalled quite well too, maybe up to 20% detail level of original. On several occasions a had some excerpts playing unvoluntary playing in my head with like 90% detail level of original. Quite often I got some music stuck in my head. When I'm underslept it's usually some bad quality pop music from 80s-90s that I haven't heard since my childhood.

Tactile, smell and taste sensations can be and are emulated too.

I wonder, do people with aphantasia have dreams? Dreams are fully emulated sensory input, though I feel like they come from a bit different place then imagined input does.

Edit: actually mind's eye is not restricted with field of vision of real sight, it can encompass say 1km sphere around me with very very low level of detail. I think that kind of capacity is used automatically in daily life when you walk somewhere and you have automatically constructed knowledge of what's behind your back.

Yes. You can also run "watch cat /proc/meminfo" in console and watch "Dirty" and "Writeback" fields to see dynamics (and how long you have to wait).

It should re-poll only when a new file is added. In op's usage scenario it won't be very ofthen.

The systemd solution may be even better, though it seems there is no way to ignore changes in temporary files that emacs likes to create.

In that setup entr only watches files that existed when systemd service was first started. So new files won't trigger backup. Something like this should work:

while true; do find $ORG_DIR | entr -d -r rclone sync -v $ORG_DIR $REMOTE:org; done

If on Linux, try running fstrim from time to time. More free space makes life of SSD's garbage collector / defragmenter much easier. I've anecdotally noticed that running fstrim reduces freezeups under heavy load from 1-2s to almost nothing on my Toshiba drive.

My first thought was: why don't they just filter the output? My guess is that perhaps LEDs are more efficient when being driven full ON or OFF rather than with a steady DC.

It's actually the opposite, efficiency goes down when current goes up.

I did a lot of stuff with Django and Flask. On current project I was free to choose a platform, so I went with Rocket.

The framework itself is similar to Flask in productivity and amount of boilerplate, though it's even better. It converts all input data to proper types for you using type signatures of handlers. So I don't need to convert strings into ints, or into json and then check if it conforms to a schema. If handler is called, input is parsed and converted.

Rust itself is more verbose than python. Typical data manipulation, like put something in dict, extract/count uniques, "transpose" a datastructure, load data from file, usually takes 1.5x-3x as much chars. For this price you get strict compile-time type checks that take lots and lots of mental load off code refactoring, C++ level performance, and real multithreading, including easy data parallelism with libs like rayon. Rust also has enums (aka algebraic types or tagged unions) that can help to represent problem space better and make fewer logic mistakes later.

It would be all unicorns and rainbows for me if a more concise database interface was awailable. Currently the go-to ORM is Diesel, and it's very verbose comparing to, say, python's sqlalchemy. Current solution is to move database access from views into methods of a model.

About iteration: code requires recompilation. Changes to non-compiled templates do not. I use Maud templates and they make html look suprisingly clean, they compile and are fast, but well, you need to recompile them every time. Currently debug build takes ~10s, though parsing and typecheking are done in the first several seconds.

Diesel doesn't support async yet, AFAIK. But async is usually used to save on context switches and stack space, allowing to support tens of thousands of live connections. You won't ever have tens of thousands of connections to database like Postgres. 100 connections is quite generous. 500 connections may be possible on the fattest DB boxes. Postgres is not async itself, it allocates a 4MB scratch buffer for each client from the start, IIRC. The only good reason to use async for communication with postgres may be if all of your other code is async and you don't want to deal with sync/async impedance mismatch. Though you'll likely want have connection pool anyway.

About IDEs I can't say much, I use emacs with LSP, it works most of the time.

Dell XPS 13 (9360) with Debian Sid, works flawlessly. Though some initial setup is required: turning off DBC (.exe available at Dell's support site) and using latest ath10k firmware from github. After that it's a laptop of my dreams ) Light, small, 13" screen in a typical 11" laptop chasis. The only drawback is soldered RAM, I have a 8GB version and Angular's reloader leaked memory, so I had to restart it every few hours.

Other than that, I've dropped it from my height on a tiled floor by accident, it landed on a corner. I've unbent screen part with pliers, and fixed cracked hinge with superglue, soda and steel wires, and that was it. Alternative would be to order a part of the chasis, but it would take too long. During the repair I've also weakened hinge resistance a bit, it was too strong for non-touchscreen version; should have done it from the start I guess.

atop as replacement for both top and htop. By default it dispays CPU frequences, disk i/o (read, write bw, average io time), disk io per process, basic network load. And it wastes no space as htop does.

I have 9360 with i5-7200U. I've just checked, fan kicks in after 90 seconds of sha1sum /dev/zero and stays always on lowest RPMs. 25C ambient. With 50% core load at 3.1GHz it stays silent most of the time...