I love this kind of logging, I used an app called Daylio and before that some other website. Great for the days you feel down to look back at how many positive ones there are. And collecting the data often leads to insights about what makes for good days. Good luck with the app :)
Since I already have the flask app set up I could inject the items into the page with a render template before serving it. Or write directly I guess since I already have the html snippet I want to insert! Good suggestion. The current architecture is partly just an artefact of the order I developed things :)
'Deep learning from the Foundations' was the previous version of this new course, but both built on their respective part 1s. So part 1 is 'practical deep learning for coders', very good intro to DL for getting things done that starts with big wins using pretrained models and then digs into them enough that you know how to make them better. Part 2 is for those wanting to go all the way to implementing the ideas 'from scratch', reading papers and diving into advanced topics.
Playing with p5 is great fun and easy to get started with, so that might be a good starting point.
If you're interested in using deep learning and pushing more towards the generative AI side, I made a free course for that: https://github.com/johnowhitaker/aiaiart, and I'd be happy to chat through the landscape and help you get started :)
It depends where you live! I've lost laptops to guys with knives, guys with guns, mysterious cat burglars... hence my preference for easily replaceable ~$500 used Thinkpads.
A good run-down of the different algorithms: https://ruder.io/optimizing-gradient-descent/
I think people favour adaptive learning rate options like Adam in practice since they generally do seem to perform well, and are often less sensitive to initial conditions and the exact hyper-parameters used. There will always be people who like to test N optimizers with parameter sweeps to squeeze a tiny bit of extra performance out, but for the rest of us the default Adam or AdamW options and good, unobjectionable choices :)
Tried it out (sceptically).
Quite a fun experience - hallucinations in the 'visual artefacts' sense. I doubt it's comparable to a drug trip (maybe those with experience can comment) but it is a somewhat pleasing way to zone out and be mindful.
At least the way I use it, it's not taking much away from my problem solving. It's just that instead of having to type `particlesGeometry.setAttribute('position', new THREE.BufferAttribute(positions, 3))` I just write `//Add as an attribute` and then hit TAB, since Copilot is smart enough to see that I've just prepared some geometry and populated an array of positions (both operations also sped up by not having to type the obvious bits).
You're still having to think through the solutions (I'm not just typing '//make a cool particle sim') but no longer need to hit SO every few minutes for syntax examples when using a new library or something.
Fun discovery: some of my own pictures are up there around my house! I've documented hundreds of species in my garden using iNaturalist, with all photos CC0 in the hopes that they'll be useful to someone. Neat to see that some at least are getting found and used :)
Four or five figures I'd guess? I'm not clued up on costs/performance for TPU stuff to give a better estimate, but guessing at a week on a 256 TPU pod, call it $30k?
There is a dataset of 5 billion image-text pairs (laion-5b) scraped by various parties. This can then be filtered and used to train these models. Cost is a bit of an issue but there are orgs that have provided compute for open model training. And Imagen is nice because the text encoder part is already available and doesn't need more training, so it would just be the diffusion model components being trained. I'd guess we'll see a biggish training run starting in a few weeks.
I've watched my parents pensions get wiped out twice. IIRC the first time my dad withdrew a few decades worth of pension and bought a pair of boots.
It's been interesting to see alternatives like the one described in the article pop up. Another popular one is property - employees might club together and purchase a small flat every N months and each own some shares in it. Then there is the ever-favoured cash/gold under the mattress approach...
https://cleanup.pictures/ (not affiliated) worked really well on a number of them - especially those where the hole is simply an area of sky or grass. For many I can't tell where the hole was, despite it being almost a one-click solution.
Seconded - the past year or so has been a bit of a step-change in terms of what is possible in this space. See also: GauGAN2, the many CLIP-guided [diffusion/VQGAN/other] image synthesis pipelines in the AI art community.
Looks like you can get some really neat effects!
I remember using a similar 'frustrated internal reflectance' setup with perspex + IR LEDs + an IR webcam + a projector to make a large cheap multi-touch surface. You're limited by the frame rate of a camera+whatever processing needs to happen, which I imagine introduces a lot more lag than, say, an apple pencil. But it's a good cheap way to get a novel input device.
I enjoyed Sabine Hossenfelder's thoughts on this and other 'Theory of everything' projects (Do we need a Theory of Everything? https://www.youtube.com/watch?v=mdu9KvLxHFg). A little bit of a reality-check for those who get really invested in these ideas, but I, like her, think it's really cool that there are people like Wolfram et al thinking these interesting thoughts and trying to push Physics in new directions.
Not a fun situation to be in. We (Zimbabwe) took off something like 25 zeroes (3, 12 then 10). Dollarization was the only sensible option for anyone wanting to buy things with a currency that didn't devalue daily. Unfortunately a 2-currency system like this is ripe for exploitation, especially once the government starts trying to manipulate exchange rates.
Thanks for that article.