Matrix Multiplication Mayhem

The tech world has been on fire lately with Microsoft and Google's intense showdown. From?GPT-4 vs LaMDA?to?ChatGPT?vs?Bard, the heat has now reached the Bing vs Google Search battle. But what's brewing under the hood is a friendly yet fierce competition between?NVIDIA's CUDA-powered GPUs?and?Google's custom TPUs?over?matrix multiplication.

Solving for efficient matrix multiplication can significantly reduce the computing resources required for training and inferencing tasks. While other methods like quantisation and model shrinking have also proven to cut down on computing, they lack accuracy.?

Thanks to the?CUDA architecture, NVIDIA GPUs are well-suited to matrix multiplication tasks as they can effectively parallelise across multiple cores. Last year, NVIDIA?announced?a multi-year collaboration with Microsoft to build one of the most powerful AI supercomputers in the world. The cluster combined Microsoft Azure's advanced supercomputing infra with NVIDIA GPUs, networking and a full stack of AI softwares to help enterprises train, deploy and scale SOTA models like ChatGPT, GPT-3.5, and many more to come in the future.?

Google's?TPUs, on the other hand, which are built on custom-designed machine learning ASICs (application-specific integrated circuits) and optimised for TensorFlow, help the team go beyond matrix multiplication and speed up their machine learning models. TPUs are also the brains behind Google's Search, YouTube, and?LaMDA AI models. Recently, Google-backed?DeepMind, found a new method to discover better algorithms for matrix multiplication called?AlphaTensor, giving them an edge over NVIDIA.?

Read the full story here.

Web3 Alternatives?

Web2 applications are slowly calling it a day as data privacy takes centre stage. Many experts also believe that internet players' solutions are not foolproof. Plus, with a monopoly of players such as YouTube, Spotify, Facebook and others, it has become difficult for customers to explore newer avenues to earn, transact, and share data in a decentralised manner.

Read more here.?

Code Readability vs Performance?

Ask any developer, and they will tell you how important it is to have well-written codes explaining everything they are planning to accomplish, but most of the time, the opposite happens, and vice versa. Though there is no single thumb rule here, but, just like any other field of science and engineering, programming also has basic rules that govern good software engineering – it is called the SOLID, DRY, KISS.

Read to find out more.

Can Google's Bard Earn From Ads??

So far, Google Search has been making revenue from ads. According to industry analysts, Google's ad revenues in the US are projected to touch $75.9 billion in 2023, hitting?$81 billion?in 2024. But with the introduction of its experimental?conversational AI chatbot Bard, powered by LaMDA, alongside?Bing's resurrection, what would the future of ad revenue for Google Search look like??Read more here.?

MLDS 2023?

MLDS 2023 was truly a developers' paradise. Here's a sneak peek into India's biggest developers' conference: