It should be noted that social distancing is an ancient practice even before 1500. For example before 1400 years back, the Persian Islamic scholar Muhammad al-Bukhari recorded the following 7th-century hadith : “If you hear of an outbreak of plague in a land, do not enter it; but if the plague breaks out in a place while you are in it, do not leave that place.”
Results are spectacular, but as always and especially with OpenAI announcements one should be very cautious (lessons learned from GPT3). I hope that the model is not doing advanced patchwork/stitching of training images. I think that this kind of test should be included when measuring the performance of any generative model. Unfortunately the paper is not yet released, the model is not available and no online demo has been published.
Recently a team of researchers discovered that in some cases advanced generative text models copy a large chunk of training data-set text ...
As other pointed out, its a fairly common practice even for western democratic government, for example French DGSE (CIA equivalent) assassination operations are called "opération Homo" and some of them are well known.
Correct me if I'm wrong, I think that unlike some other markets (China and to some extent other Asian countries / Russia / etc) it's surprising how dependent the EU is on US corporations for digital services. As said the EU anti-monopolies chief Margrethe Vestager: "In the past few weeks we have all been fascinated by what is possible digitally. But the coronavirus showed how dependent we are on US corporations, and that was a wake-up call,".
Throughout human history, you can find many discoveries that were made before understanding why they work in the first place: we know exactly how the neural networks work, but we don’t know why they are so effective, maybe because of the lack of a theoretical understanding of deep learning and complex neural networks.
For this particular case, BERT is the result of incremental enhancements of existing architectures and training procedures. Fundamentally, BERT is a sequence prediction algorithm, and historically, the sequence prediction models were based on complex recurrent or convolutional neural networks. Experiences showed that the best performing models were those having an attention mechanism (the concept of directing the focus on some words or sentences). Some researchers proposed a new simple network architecture, based solely on attention mechanisms, without complex recurrence or convolutions! and they showed that some of these models achieve state-of-the-art performances while being more parallelizable and requiring significantly less time to train.
Not a word about Secure Enclave in iPhone, it should be explicitly stated that the discussion is about Android/Linux Phones only. Otherwise, the conclusion that the most secure phone on the market today is the Pixel with CopperheadOS is not accurate (in my first read I interpreted it as it is the most secure phone not the most secure Android).
If you add occasional pixels to skew the average then probably it will be noticeable in the original image. But the interpolation scheme that uses only the four corners while ignoring the rest can be easily fooled. You can blend an entire lower resolution image in the four corner.
Love the idea. I think that a pure collaborative filtering algorithm has some limitations (cold start, etc). Maybe a hybrid CF + content-based approach will be more adapted for essays?
The idea of automagic conversion of legacy code is great and the market is huge. But then Watson failed to deliver in the past and I doubt that this time it will live up to its claim.
I use this extensively for mixing/recording video streams, although I think that Muaz library is great, but browsers support for WebRTC recording is far from being production ready.
Some issues off the top of my head :
- Non seekable videos in Chrome
- Video/audio lags and de-synchronization
- Loss of audio streams in Chrome
In 2000 using VB, you can build a GUI, make a working Windows application using a minimum amount of code. Also the documentation (msdn) and the community was really nice.
In software engineering, languages and tooling the progress is really slow. RAD tools existed for decades, OOP and functional programming paradigms remained largely the same for a very long period of time ... incremental enhancements but not really a breakthrough.
Very interesting conversation, the only thing that I wasn't expecting is the Egyptian pyramid analogy. Maybe I misunderstood something, but I think that Egyptian pyramid has a great structural integrity ?
"Most software today is very much like an Egyptian pyramid with millions of bricks piled on top of each other, with no structural integrity, but just done by brute force and thousands of slaves."
In our team, We serve PyTorch models in production using libtorch. A C++ library for loading models. You can easily call the C++ code in Go if you wrap it in a C interface.
Very nice results that challenge the conventional wisdom. DNN can work with symbolic data and calculations. I think that, it says a lot about our poor theoretical understanding of deep neural networks despite all the practical advances.
SDL is fine for cross platform development. Personally, I have used it for some commercial games in the past, but I think it's not comparable to Unity or unreal engine. SDL is not a game engine, it provides only lower level features (audio, opengl context initialization, joystick support, etc).
Even for things like in-game menus and text rasterisation you are on your own.
I think that Von Neumann raise three important issues, two of them are largely solved, and one of them is not yet solved.
- Hardware failures
- Complexity theory
- CS and Formal logic is combinatorial rather than analytical, we can't use the powerful tools of mathematical analysis to solve those combinatorial problems.
This reminds me of a remark of
Paul Erdős about the Collatz conjecture: "Mathematics may not be ready for such problems."
EDIT: text format
Nowadays it's more valuable to create custom nodes to enrich the node-based editor library. Sometimes we write shaders from scratch if we need to implement some original lighting, materials, etc
I agree with you, however in almost all commercial games that I worked on, we do write our own shaders. For simple shaders, artists use directly a graphical tool included in unity for shaders authoring without writing any code, for complex ones we write our own code.
Softimage|XSI : 3D animation, modeling and rendering. It was used in the film, video game and advertising industries as a tool to generate digital characters, environments and visual effects.
It has a unique user interface with some pretty awesome features.
Autodesk acquired the Softimage brand and 3D animation assets from Avid for approximately $35 million just to kill it some years later.
Edit: add info about Autodesk
Page Rank is an incredibly elegant algorithm.
But in practice I suppose that the actual Google search algorithm has little resemblance with the mathematically elegant PR algorithm, because, it has to implement many custom tweaking, tuning, DMCA black-lists, right to be forgotten list, etc ...
Most modern neural networks implementations are parallelized. And that is why we can run them extremely well on the GPU. For example Volta GPUs delivers 5X increase in deep learning training compared to prior generation NVIDIA Pascal architecture.
This is why I was asking for clarification about the hardware claims.
> hardware progress is slowing down quickly)
I would be interested to know more about this. I haven't heard yet that the progress of GPUs cores for example is declining quickly ...
PyTorch for example is often called a deep learning framework. But at its core it is a hardware accelerated linear algebra and tensors library with the support of auto differentiation.
So it is well suited for implementing numeric algorithms, classic machine learning algorithms and differentiable programs.
DL is really a subset of differentiable programs, but it happens to be a very interesting one.
Face ID sensor will projects 30000 infrared dots onto the user's face, and then it reads the projected pattern. The pattern is sent to a deep neural network in order to confirm a match with the phone owner's face.