New Year’s day, after the breakfast, I am scrolling through as usual my list of freshly published articles and papers on the most recent Machine-Learning development in order to stay on the frontline of the AI evolution. Among all the inspiring experiments, one article has caught my eyes: With 99 lines of code, you can have the magic of snow and ice like Princess Aish in Frozen. It’s an article about a new programming languageTaichi designed to provide native support for sparse data structures, such as graphics.
In this very first Edition of our AI Frontline column, we’d like to present you in more detail this productive programming language Taichi and share with you our views on the Top AI Trends in 2020.
Advancements in AI mean that computers can be teached on certain creativities. The “generative model” from Deep-Learning technology enables AI to create brand new things, even though for the moment it still has many limitations. Nonetheless, the whole creative industry from film to news to advertising and marketing will escalate the use of such AI techniques to test out new ideas and accelerate prototypes.
The implementation of AI on entertainment creatives is going to increase. In 2019, we saw Robert De Niro de-aged in Martin Scorsese’s The Irishman, with the assistance of AI. Using AI in creating new special visual effects is likely to become increasingly common.
In the past, even some special effects in movies and animations are as short as a second, they require a high-performance computer to perform calculations for weeks. For example, in “Frozen” there’s no real person, but the budget of production is as high as 150 million US dollars. Every second burnt thousands of dollar.
It is unimaginable for most people to use normal computers to make CG effects. That’s why Taichi, a new CG special effects programming language by a recent MIT Ph.D. from China, is revolutionary. A simple physical scene can be rendered in an ordinary PC in just a few minutes. It is 188 times faster than TensorFlow (Deep-Learning framework of Google), 13.4 times faster than PyTorch (Deep-Learning framework from Facebook), and the code length is only one tenth of other low-level methods.
GPT-2 is a large pre-trained language model that can generate human-written like paragraphs of text. Since its debut with a much lighter model in the February 2019, GPT-2 has garnered immediate attention and broad discussions in the NLP (Natual Language Processing) research community. On one side, NLP practitioners are fascinated by its state-of-the-art performance; on the other side, serious concerns are raised regarding its potential misuse, for example maliciously generated fake news. With the latter fears, GPT-2 was originally deemed as “The AI That’s Too Dangerous to Release”, and OpenAI has withheld its codes and weights until november.
OpenAI states its decision to finally release the full version was influenced by following reasons:
- Humans find GPT-2 outputs convincing
- No strong evidence of misuse so far
- OpenAI has developed its own detection model to combat the malicious uses. Its counter-model has reached about 95% effectiveness.
The advance of technology is inevitable. Will the AI take the jobs of journalists, artists or other professionals? It’s a question that many start to wonder. None of us can certainly say with great confidence that the probability is zero. For now, most of the AI techniques are applied as tools that interact with humans so to better inspire humans creativity.
The development of AI is creating new opportunities to improve our daily life. The ability to properly leverage AI can truly empower business transformation. At the same time, the AI research community should take actions to deal with ethical controversies surrounding the applications and to promote AI For Good objectives. Responsibile AI practices should serve humanity, rather than the other way around.
Join us at TECHENGINES.AI and keep posted with the lastest news of AI evolution!