But is autodiff combined with a blackbox jit a real solution? The jit either works for your new model or it does not. If it does not, you can do pretty much nothing about it, other than ping jax authors or get your own hands dirty with jax internal code. Why is noone working on a usable low-level framework, where I can implement QRNN or more complicated stuff without relying on a black-box jit? Jax could have chosen to be this, but instead is a fancy solution to a non-problem.
You only need a few seconds to read the title, and you can cut away 150 out of those 200 with that. If it's a paper simply not in your area of research, you don't need to bother. You can then read the abstract (30 seconds) for the rest, and finally you'll pick 5-10 papers to read. You'd skim them with 5-10 mins per paper, and if you like what you see in 1-2 papers, you'll give it a more through read. So this is 10 mins + 30 mins + 2 hours + buffer, which is at best 3-4 hours.
Of course this requires a strong filter at the first stage. You can also set aside time to investigate specific topics outside your work, but you won't do that as often as you keep up with your own literature.
So in the case of AI/ML, the fashionable new thing is unsupervised learning, self-supervised learning, reinforcement learning, GANS, etc. The played out problems are image classification, speech recognition, etc.
It's in the nature of social networks that new social networks don't look like old ones. Youtube is definitely a social network, and so is Twitch, Discord, TikTok, etc. Youtube has the additional advantage of having persistent count, so I would count on Youtube lasting much longer than any other social network whose popularity changes generation to generation.
People were meme-ing the goose game hard much before its launch, and much before Dunkey covered it, but that probably helped sustain its exponential growth.
More than streamer-friendly, it's Reddit- and Twitter-friendly. Any small clip of the game is automatically funny and share-able. The bird's simple design makes it easy for fan artists to make more memes.
POs were a bit too limited for me. The sequencer isn't powerful enough to make anything more complex than basic chiptunes.
Not to diss on PO. I was super hyped about it as well, and would have liked to know about its limitations before buying it. Anyone of the numerous phone apps would be a cheaper and better replacement for a PO.
What got me started in music was Garageband. It's free (for macos users) and quite simple and intuitive. It's not as fully-featured as something like Ableton, but quite capable for making professional-level songs.
Alec Holowka was a game dev who committed suicide recently after allegations of sexual assault by ex-girlfriend Zoe Quinn. He was the developer behind Night in the Woods. The NitW studio said they've cut ties with him after Quinn's allegations.
Small ads on the web are still more targeted than newspaper ads, since web ads appear on specific websites which are of interest to the user. If I visit a motorcycling forum, I should get motorcycling ads, and they can be effective even if they're small.
What you're saying and what's in the link seems to be consistent with the advice of refusing to talk to the police at all, and not to talk partially or to answer some questions and not answer other questions, or answering but changing the statement later.
It indicates that the model is learning a good likelihood function, and that it has still not trained to convergence which makes the samples have lower likelihood scores than real text. That's all that's needed to make it a good recognizer.