A website that sorts the recent Peloton bike rides by difficulty.
https://pelohard.com/
I love my Peloton and it was really annoying finding a ride with a difficulty over a certain threshold that was recent. I'm an embedded developer by day and did this with a lot of help from ChatGPT and Claude. Hoping to add some more features to it as time goes on.
Fold the sides of a 50lb horse feed bag down, fill with dirt and add seed potatoes is literally all I did. We go through lots of bags, so I have many empty ones. I have a problem with weeds (thistles) and oregano taking over everything in the garden area, so this was my lazy man's way of avoiding weeding.
I also have a giant compost pile consisting of mostly wood shavings and horse manure that I planted about a dozen potatoes into as an experiment, along with a couple dozen carrots. So far they seem to be doing really well.
I'm very impressed! Using Gpt-4o and Gemini, I've rarely had success when asking the AI models to create a PlantUML flowchart or state machine representation of any moderate complexity. I think this is due to some confusing API docs for PlantUML. Claude 3.5 Sonnet totally knocked it out of the park when I asked for 4-5 different diagrams and did all of them flawlessly. I haven't gone through the output in great detail to see if its correct, but at first glance they are pretty close. The fact that all the diagrams were able to be rendered is an achievement.
As the article mentions, this is also used by authors. I went to a book signing for Mark Z. Danielewski, the author of House of Leaves and he mentioned that for his books, he always ends up throwing out the beginning of the book.
I understand the skepticism around AI in ed tech, and I think people have the right to be skeptical of this being portrayed as a "cure all". Saying that it is of no value because nobody will pay for it and the kids wont learn from it because they're hungry does not capture the whole picture. I was never a great student because I had trouble focusing in class, but if I had this to guide me through my homework, I believe I would have been a much better student. Looking at math homework and having no idea how to even start and no resources to help is very different than looking at the problem and working with AI to help you understand how to attack the problem. Sure, you can't turn every single kid into a math wiz, but I think there is a real possibility that this will help almost every kid become better at math than they would have. Assuming this is low cost, I think many parents would be very willing to pay for it.
If you have ever worked with children you will realize vast majority of them lack the willpower to learn new things and seek out answers. A good teacher can provide the social accountability and guide them, but it's not something you can put on auto pilot with AI.
Children look to adults as role models for what to learn and why. We already know that children respond better to role models who have similar ethnicities and backgrounds to them, let alone being the same species.
AI cannot and will never be able to provide this motivation to learn, which is what kids actually need, because it is not human.
> AI cannot and will never be able to provide this motivation to learn, which is what kids actually need, because it is not human.
We hear all the time about digital addiction, gamification, FB, Tiktok, etc. Addiction is arguably more difficult to achieve than motivation. AI will be able to motivate people just fine
Using modern digital gamification to reinforce learning outcomes, a-la TikTok , will probably produce a generation of absolutely brilliant, weird psychopaths.
At the same time, I think asking a person or a parent requires much more will power than asking a computer (for whatever reason -- but part of it is just that a teacher might not have as much time as a child needs/wants). I do agree the social accountability of a good one-on-one teacher is the most ideal -- for me I got that from my parents/siblings. But lots of folks don't have access to that, and school systems don't have the resources to supply that, so maybe AI might be a good middle ground.
I am about a month away from finishing my teaching degree (math). My experience so far AND the best research is extremely clear on the following:
Kids learn best by working in small groups with other kids.
These groups need to be gently guided by adults, but they should mostly be left to do a lot of independent exploration and discussion amongst themselves. The teacher is there to prod discussion in productive directions, provide feedback, answer questions, give hints and encouragement where needed. Admittedly, AI could do certain part of the teachers job, but it can never replace a peer group.
My understanding is that the earlier rolling code systems are easily defeated and I think this can be done (possibly with stock firmware) using a Flipper Zero.
Prior to that, garage doors had a set of DIP switches (16, or 32, I can't remember). You matched the switch configuration on your opener with the switch configuration on the controller. And as you might imagine, in a typical suburban area about 80% of the garage doors are set to all zeros.
Because the range of the devices was "lucky if you can open the door from the bottom of your driveway", most people didn't notice this. Of course that meant you could open a large number of garage doors by sending the "0" signal for each manufacturer with enough wattage.
Compatible models are made by reverse engineering each individual model's rolling code implementation (in the early days) and making an accessory that had the necessary seed value or other component to allow it to be "paired" with a compatible door head unit. Considering it wasn't uncommon for the higher-end models to charge $150 for an accessory remote, manufacturers had a bit of incentive to roll their own slightly incompatible implementations.
This is from memory and minimal memory at that, but -- late 90s or early 00s, I think, "HomeLink" was created, which basically allowed car manufacturers to integrate a door opener into the car. If you bought a higher-end model, your visor might have the buttons in it. I believe licensing allowed third-parties to easily create fully compatible accessories at that point (pay a fee, get the patent license/datasheets sort of arrangement).
[0] Genie thinks they were first in 1995 but I seem to recall we had a rolling code door installed as early as 1993.
There were also so-called "learning codes" that overlapped substantially with the introduction of rolling codes. Think fixed-code remotes with a random preprogrammed code instead of dip switches. Chamberlain's "billion code" is representative. They have to be paired with the opener receiver in the same way as rolling code remotes.
The largest garage door manufacture in the US uses the Security+ and Security+ 2.0 algorithms that are rolling, but can be fairly trivially decoded to gain the serial number and rolling value of a remote. [0] This is how the flipper zero decodes remotes for playback later.
They use rolling frequencies at least our 2021(?) garage opener does, to retry in the presence of narrow band noise from fcc violating devices. We seem to have neighbors blasting rf in the band that our previous builder installed 2014 model used, because it rarely worked unless super close.
At my first real Engineering job, I used this paper to both understand and implement a PID control system. Then by using the Ziegler Nichols method of tuning, the system was controlled way better than anyone had expected. Since then I have used it a handful of times with minor tweaks and each time it feels like magic when things work well.
ChatGPT 3.5 is free, but GPT4 is not. I can see a lot of people using something as good or better than GPT4 that is supported by ads and no monthly fee.
I love my Peloton and it was really annoying finding a ride with a difficulty over a certain threshold that was recent. I'm an embedded developer by day and did this with a lot of help from ChatGPT and Claude. Hoping to add some more features to it as time goes on.
reply