What are the killswitches acting on ? Are they acting only on the dedicated "enable" or "poweroff" inputs of various chips ? Are they cutting the power supply of the chips ? Are they cutting both the power and logic busses, thus isolating the target chips (at least electrically) ?
1 Modem | Pulls Q1501 gate up (FET killing modem power) | "On" enables 2G/3G/4G communication and GNSS hardware, "off" disables it.
2 WiFi / Bluetooth | Pulls up CHIP_EN | "On" enables WiFi and Bluetooth communication hardware, "off" disables it.
3 Microphone | Breaks microphone bias voltage from the SoC | "On" enables audio input from on-board microphones (not 3.5mm jack), "off" disables it.
4 Rear camera | Pulls up PWDN on OV5640 | "On" enables the rear camera, "off" disables it.
5 Front camera | Pulls up PWDN on GC2145 | "On" enables the front camera, "off" disables it.
6 Headphone | Pulls up IN2 on analog switch BCT4717ETB | "On" enables audio input and output via the 3.5mm audio jack, "off" switches the jack to hardware UART mode.
With the microphone bias floating, what prevents some digital signal processing form recovering faint and fuzzy audio? I'm sure the microphone loses at least several dB of gain with the bias floating, but isn't it much safer to either disconnect the bias and tie it to ground, or else pull up/down the the digital output of the ADC?
I understand that with the bias floating, the microphone output will be a combination of radio and quantum thermal noise, but won't that noise still be slightly modulated by the microphone? Or is it that the noise being modulated will be below detectable by the ADC and the digital output will always be exactly 0?
Depends on actual hardware implementation, but those microphones are REALLY good at picking up audio. I once fiddled with some microphone and wondered why it works so poorly (lots of "digital" noise, faint audio), maybe cable broke or smth. Turned ot, it was "disabled" with hardware switch on cable, yet still picked up enough sound to "somewhat work".
Nice work! Your BOM looks similar to OpenPCR's. One fundamental advantage of these smaller PCR machines over traditional designs was running fewer samples (i.e. 16 vs 32+), which translates to less fabrication, less peltier devices, and less sensing/cooling, among other things.
The Bretton Woods Agreement pegged the dollar to gold ($35 for an ounce of gold). I understand that this can work if the central bank emits $35 for every ounce of gold they store.
But how does that account for destroyed/damaged/lost dollar bills ? The loss is very hard quantify and monitor, but needs to be compensated either through re-printing of new dollars, or through the destruction of stored gold to maintain the desired exchange rate.
Could someone with better knowledge than me explain how this works ?
> Could someone with better knowledge than me explain how this works ?
The short answer is it doesn't. We haven't been on the gold system for a long time, and BW was a sham, no one really did what the agreement said, there wasn't nearly enough gold to do so anyway. Breton Woods was a fictional agreement, basically. But once Nixon stopped pretending, we've been purely a floating currency like most others since.
Cash money is replaced at face value if it is damaged - presumably the owner of the cash has an interest in seeing it kept safe.
But unless the Treasury is notified and presented to their satisfaction with damaged bills for replacement, by definition, there is no way of tracking destruction/damage to cash money.
The Bretton Woods agreement never had anywhere near enough gold to account for all of the currencies pegged to it. The US held all the gold and agreed to the fixed conversion and everybody else agreed to never convert. It never had to balance presumed destroyed bills with the gold reserve.
Here is also a lesser known open source project that demonstrates the use of proof-of-stake for energy efficiency, combined with a fixed-width directed acyclic grah with transaction sharding for parallelization.
Furthermore, "final" and "stale" blocks are dynamically forgotten to increase storage efficiency.
It achieves 10,000 tx/s on chain and transaction times below 1 minute.
There's also Credits which claims to be the fastest chain at 0.1 second transaction times and to be capable of over 1 million transactions a sec. Though it you run it that fast you use a lot of disk space. https://credits.com/
Differentiable does not mean easy to optimize.
One could imagine implementing sha-256 using differentiable operators, and yet the system as a whole would not be optimizable at all.
It would be interesting to have compilers that optimize the "optimizability" of differentiable programs tho...
Also, here are two interesting examples of differentiation through physical systems for classification:
Would it be possible to add one or several memory components to the LLM model itself while training it ? Example: https://en.wikipedia.org/wiki/Differentiable_neural_computer
I guess doing so would make the training less parallelizable ?