Cerebras is now out of stealthmode (with VC money behind them), the startup released their flagship product this week, the Cerebras Wafer-Scale Engine (WSE). The WSE claims to perform AI calculations between 100 to 1,000 times faster than other alternative products. The size of the WSE is also something to behold, as its larger than an iPad and is built out of more than 400,000 cores and 18 gigabytes of memory is on the chip. It’s also 56x larger than the largest GPU ever made. Everything is on one piece of silicon. Cerebras is developing next-gen computing chips built for deep learning applications. Think of them in the same playing field as NVIDIA and Graphcore (one of my favorite companies at the moment). This reminds me of Moore’s Law and is exciting.
Tools of the week:
- Out of Deepmind: Kapitan – a python based templated configuration management for Kubernetes, Terraform and more. As a tool, you can manage complex deployments using jsonnet, kadet (alpha) and jinja2.
- Linear algebra with everyone’s most favorite YouTube channel 3BLUE1BROWN
- Cypress – testing for anything that runs in your browser – $ npm install cypress
Reads of the week:
- AI in financial services: Is finance ready for AutoML?
- A ProPublica article this week about how Jeff Bezos and Eric Schmidt have seduced the Pentagon, with one whistleblower raising flags.
Lastly:
- Boston FinTech Week is happening the 9th – 12th of September. Schedule here.
- Google Research Football, an environment for reinforcement learning (very cool work) & Git.
- A social media ‘win’; Twitter will stop accepting advertising dollars from state-backed media outlets.
A Blog 