"Open source" in ML is a really bad description for what it is. "Free binary with a bit of metadata" would be more accurate. The code used to create deepseek is not open source, nor is the training datasets. 99% of "open source" models are this way. The only interesting part of the open sourcing is the architecture used to run the models, as it lends a lot of insight into the training process, and allows for derivatives via post-training
There's nothing magic about Soylent for weight loss. It's a simple equation of calories in and calories out. The advantages that Soylent offered me was convenience for counting said calories, convenience for meal prep, and being reasonably certain my body was getting a decent distribution of micronutrients
I did ~1.5 years of only Soylent, then transitioned into 2/3 meals per day being Soylent, which I've done for the last ~6-7yrs.
I'm the healthiest I've ever been, but it does require discipline, exercise and attention like anything else. Calories are calories and if you consume more than you burn, you'll poop a lot and gain weight. If you drink at a significant deficit (my 1.5years was at 1200kcal/day) you will poop once or twice a week and it will take a few months of your body getting used to it for it to be more than liquid.
As others have said though, it's a deceptively dehydrating liquid. You absolutely still need to drink water, and your water intake will largely dictate how much you pee.
It's a little deeper than that, a lot of advertising works on engagement -based heuristics. Today, most people would call it "AI" but it's fundamentally just a reinforcement learning network that trains itself constantly on user interactions. It's difficult-to-impossible to determine why input X is associated with output Y, but we can measure in aggregate how subtle changes propagate across engagement metrics.
It is absolutely truthful to say we don't know how a modern reinforcement learning network got to the state it's in today, because transactions on the network usually aren't journaled, just periodically snapshot for A/B testing.
To be clear, that's not an excuse for undesirable heuristic behavior. Somebody somewhere made the choice to do this, and they should be liable for the output of their code.
If your bootloader is unlocked, you can fastboot boot an image and that will only run in memory the one time, however, it will share a data partition (apps, preferences, etc) so it might not behave as well as a native install would have.
I think you've convinced me that it's a slightly more complicated problem than I initially gave it credit for; thank you for that!
I think you could solve for the disparate community theme problem by also requiring title match for mergers. You could probably also solve for it by having a 2-way merger whitelist on links. E.g community A and B both maintain lists of "similar" communities and then if A's list contains B and vice-versa they would merge.
Comment moderation I got nothing though. That's a tough one.
Lemmy needs some sort of built-in way to merge them. That'd be the best solution I think. Then you could just pick a list of relevant communities and it'd be pretty seamless
I don't know of any off the top of my head, but with a cheap digital caliper and tinkercad, I assume you'd be able to model one fairly trivially. You could friction-fit two halves around the cable, and secure it with some simple adhesive, or some kind of simple bolt/nut fastener mount if you wanted to get clever.
Depends on where you work and what their policies are. My work does have many strict policies on following licenses, protecting sensitive data, etc
My solution was to MIT license and open source everything I write. It follows all policies while still giving me the flexibility to fork/share the code with any other institutions that want to run something similar.
It also had the added benefit of forcing me to properly manage secrets, gitignores, etc
The canvas API needs specific access to hardware that isn't usually available via browser APIs. It's usually harder to get specific capability information from a user's GPU for example. The canvas API needs capability information to decide how to draw objects across differently capable hardware, and those extra data points make it that much easier to uniquely identify a user. The more data points you can collect, the more unique each visitor is.
Here's a good utility from the EFF to demonstrate the concept if you or anyone else is curious.
Just think, an extra long shirt can cover that hole, and we could embed a flexible display, wifi module, and a camera in the extra space. This could scan the faces of those around you, and display personalized ads! This is an excellent solution to the hole in your pants, and frankly, the only secure one.
"Open source" in ML is a really bad description for what it is. "Free binary with a bit of metadata" would be more accurate. The code used to create deepseek is not open source, nor is the training datasets. 99% of "open source" models are this way. The only interesting part of the open sourcing is the architecture used to run the models, as it lends a lot of insight into the training process, and allows for derivatives via post-training