It's a painful review of the state of the modern web. Hard to believe that every time when I try to open the JIRA my browser downloads the 50 Mb of JSs.
The "sailplane problem of modern Image Generation AI". Right after appearing of model Image Generator AI have tried to generate the photorealistic image of the sailplane. But that was impossible 2 years ago and still impossible today (but lot better). That last attempts examples (DALL-E and MidJourney)
How to do things if you're not that smart and don't have any talent
This is a blog post aimed at people who want to do important work or make meaningful contributions to work, but feel they aren’t that smart and don’t have any talent.
Looks like me!
Despite decades of advances in document rendering technology, most of the world's documents are stuck in the 1990s due to the limitations of PDF. Yet, modern document formats like HTML have yet to provide a competitive alternative to PDF. This post explores what prevents HTML documents from being portable, and I propose a way forward based on the EPUB format. To demonstrate my ideas, this post is presented using a prototype EPUB reading system
During my work in LitRes (Russia's largest ebook store) we had EPUB as an internal format for storing books in MyBook.RU. The main problem for us was validation: there's just too much things that you can do inside the EPUB, and the validation process was difficult. As a result we came to almost the same decision: we just cut off all the "suspicious" elements from the files and have our "Light EPUB" format.
Interesting picture. What I see here:
If you've ever experienced Facebook showing you an ad for something that you were talking about out-loud about moments earlier, you've already dismissed everything I just said. You have personally experienced anecdotal evidence which overrides all of my arguments here.
Here's a Reply All podcast episode from Novemember 2017 that explores this issue: 109 Is Facebook Spying on You?. Their conclusion: Facebook are not spying through your microphone. But if someone already believes that there is no argument that can possibly convince them otherwise.
I've experienced this effect myself - over the past few years I've tried talking people out of this, as part of my own personal fascination with how sticky this conspiracy theory is.
From my point of view: the only model of development for the future of AI is open source. Especially because of trust issues. People are really afraid of AI, and without a deep understanding of how it works, it's impossible to bring it into society.
Microsolf Phi-2 Language Model - it's an old and new way of doing language models. Instead of using a lot of random data, we can use a small subset of high quality data and use that to train the model. Just like in the "good old days" of classical machine learning.
As a result, we have a small and effective model that can be used on mobile devices.
The idea is that in the near future, this will be the main method for the creation of LMs: