This week in AI: Existential threat from AI, Nvidia announces AI supercomputer, UK wants global AI watchdog and more

Nvidia announces new generative AI supercomputer

Nvidia says DGX GH200 has nearly 500x more memory than the previous generation Nvidia DGX A100, which was introduced in 2020.

• The DGX GH200 uses Nvidia's NVLink technology to combine 256 GH200 superchips, which each house an ARM-based Grace CPU, and H100 Tensor Core GPU, that allow for performance up to 1 exaflop (equivalent to one quintillion floating point operations per second) with a whopping 144 Terabytes of shared memory.


• The NVLink Switch System allows all GPUs in a DGX GH200 to work together as one, providing 48 times more bandwidth than the previous generation.

AI poses 'extinction' risk, say experts

The latest statement, housed on the website of US-based non-profit Center for AI Safety, gave no detail of the potential existential threat posed by AI.

• A one-line statement signed by dozens of specialists, including Sam Altman whose firm OpenAI created the ChatGPT bot, said tackling the risks from AI should be "a global priority alongside other societal-scale risks such as pandemics and nuclear war".


• Dozens of academics and specialists from companies including Google and Microsoft signed the latest letter, which comes two months after billionaire Elon Musk and others called for a pause in the development of such technology until it could be shown to be safe.

MIT researchers develop small-scale language model more efficient than larger ones

The self-learning language model can learn from its own predictions eliminating the use of annotated training data.

• Researchers at MIT's Computer Science and Artificial Intelligence Laboratory (CSAIL) have developed a small, self-learning language model that surpasses large-scale language models.


• CSAIL's algorithm for the model called Simple Pseudo-Label Editing (SimPLE) allows it to learn from its own predictions, eliminating the use of annotated training data.