Go pressure your governments and representatives. It is not like Discord wants to do this, but they see the writing on the wall with UK and US regulations in effect, EU pressure for chat controls, AU banning teams from social media, so on.
It is either this or get regulated even further.

Over the weekend I’ve exterminated with direct digital synthesis with aliasing. The outcome being an ESP-32 flicking a pin at 80MHz resulting in modulated transmissions at 400+ MHz.
There is prior art of doing this to transmit LoRa messages.

I’m going to have to do a write up on this.

I made this yesterday.
For some reason it caused some of my text fields to type text is reverse direction.
The cursor would move to the front of the field after each letter.
Confusion ensued 🤣

Let’s call it what it is

gist.github.com/Kubuxu/2769d...

There is only one way to proceed: write a reproducer and start bisecting

ICANN data matches one on Google’s official domains

Some more photos from Bangkok

Flying back from Bangkok. It was amazing to see everyone

rp2040 as the PIO was making generation of 64 coherent signals way easier (I still had to use multiple rp2040s).

CDMA - code division multiple access
Hadamard Transform - think Fourier Transform but in a world where we didn’t discover multiplication and division. Benefit, it is it is stupid fast as it uses only addition/subtraction but doesn’t translate to frequency domain nicely.

I was using rp2040 in this case which has a bit junky ADC but for 64 simultaneous coherent signals it can work with anything really. As long as you manage to keep system SNR/SINAD above 18dB.

Agreed, no custom hardware in this case. I was just wondering why CDMA is not used for cases where for example you have N same sensors on own bus. The answer is simple, simplicity and noise resistance.

I’ve been playing with CDMA based single wire comms on micro-controllers using Hadamard Transform. It seems doable using regular uC ADC to receive 64-128 coherent signals at the same time at few kbps.

Then I through again why? You can do better than that using i2c and polling every endpoint.

It is great to stumble onto Bluesky in the wild as I was showing edX to a friend of mine. @cs50.bsky.social

The AI is separate cost which they charge search credits for. It’s as optional as it gets.
The $5 plan is much better for recommending if their estimates are correct. Remember not everyone runs over 2000 searches per month (but I do).

Interesting pricing update from Kagi.
I very much like how transparent they are about pricing and costs. My first instinct was to minimise my bill. Their openness changed my mind.
https://blog.kagi.com/update-kagi-search-pricing

I’ve adopted this by using Glass. It’s a membership platform for photographers. In my opinion it leads to more aligned incentives between users and the developers.

My solution was to create a Bayesian model with Dirchlet prior where samples are distributed according to Multinomial distribution where the probability of Source label is set to zero and probabilities are re-normalised.

I haven’t worked with the Hidden variant of Markov model but it seems like a close fit although it will be a bit weird as emission probabilities would have to be tied state transition probabilities.

My solution was to create Bayesian model of the process.

The process of transforming is what we can observe but we are interested in figuring out the underlying weights/probabilities.

Each kind of a token has weight assigned which defines its probability to appear if they were drawn randomly.

A token transforms into another kind of token with a different letter when you push a button but it cannot transform into the same token.

Statistics question/exploration
Attached are samples of observed transitions. There is one true vector defining independent probability of each state but during sampling transition to the same state is disallowed.
I have a solution but it seems overkill

Both of USB-C ports in my X1 Carbon are failing 😕
I was looking for a new notebook either way but I will try to get this one repaired, it is too good to turn into e-waste.

So being able to say: given the data we have and the prior assumptions we’ve made, the parameter lies within range [X,Y] with 89% subjective probability is just what I’m looking for. [3/3]

In most cases where I reach for statistics I don’t want to know if given parameter value will be within a set range if the experiment is repeated N number of times.
I want to know what are the possible parameter values given the data I have. [2/3]

I've been learning about Bayesian data analysis recently. The ability to treat a parameter of a model as a probability distribution is something I've been missing since the first time I tried applying statistics in the real world. [1/?]

It’s a rare but such a good feeling when you start a brainstorm session with a X problem at hand only to, after all, leave convinced that there is no problem at all by considering the underlying requirements.

/me on Monday mornings or any mornings for that matter

s/gig/finger
IDK how I messed it up that badly, I guess that is what I get for posting at 7am 😂

@paul.bsky.social ignore the post about the long press, when I was long pressing I had to be moving my finger slightly sideways.