Roundup Here’s this week’s collection of AI-related news that we found interesting. Read on to find out more about a new chip coming to Microsoft Azure and how Twitter hopes to deal with deepfakes.
Graphcore ML chips coming to Microsoft Azure: Graphcore, a British AI hardware startup, is teaming up with Microsoft to bring its Intelligence Processing Unit chip to cloud users.
“The Graphcore IPU is unique in keeping the entire machine learning knowledge model inside the processor,” it said this week. “With 16 IPU processors, all connected with IPU-Link technology in a server, an IPU system will have over 100,000 completely independent programs, all working in parallel on the machine intelligence knowledge model.”
It’s not available to rent via Microsoft’s cloud just yet, however. If you’re interested you’ll have to sign up online, the service will be prioritized for those working in natural language processing – a strong point for the IPU.
Despite all the flashy claims of the IPU’s performance when it comes to training and inference for the large language model BERT, it’s not really clear how good the hardware really is. Graphcore hasn’t revealed the full specs, and it hasn’t submitted any results to ML Perf, an industry effort to benchmark AI hardware. There are also few comparisons with other types of chips for different types of more commonly used models like ResNet-50 in computer vision.
You can read more about the announcement here.
Here’s what it’s like to work for Amazon Mechanical Turk: Unfortunately current machine learning systems can’t learn from the messy real world yet, and require large training datasets that have been carefully preprocessed by teams of humans.
Services like Amazon Mechanical Turk, an online marketplace where workers – or turkers – bid to perform simple computational tasks for a small wage, provide a way for companies or researchers to outsource the laborious work of labeling data.