Time to BLOOM 🌸
BigScience Language Open-science Open-access (aka BLOOM) – a large-scale transformer-based language model with 176 billion parameters – has been on the news for some time now. This new open-source language model is unique not because it is architecturally different from GPT-3 (175B) but because it didn’t come from the big tech, which for years had a monopoly on the research and development of large-scale language models ( LLMs ).
Started as a research workshop a year ago, BigScience is an open collaboration bootstrapped by HuggingFace, Grand Equipement National de Calcul Intensif ( GENCI ), IDRIS and Montreal AI Ethics Institute, among others. The team consists of over 1,000 researchers from across the world who have been working together on creating open-source large-scale language models like BLOOM on the 28 petaflops Jean Zay (IDRIS) supercomputer located near Paris, France, to counter big tech’s stronghold on large language models.
BLOOM is not the first open-source language model of such size. Meta, Google, and others have recently been open-sourcing a few models. Unfortunately, those are not the best these companies can offer.
The most talked-about language model, GPT-3, is still not open-source. Even after several open-source language models have been mushrooming in the space, claiming to be comparable with GPT-3.
The question is, will OpenAI open-source GPT-3 now?
Currently, GPT-3 is available through OpenAI’s API. The API is quite expensive. While the high costs have limited GPT-3’s mainstream adoption, many researchers and professionals are using it to build prototypes and commercial applications.
Also, its partnership with Microsoft, where the tech giant acquired the exclusive license of GPT-3 as part of its larger AI at scale initiative, could be in the way, even if the company plans to open-source GPT-3. But, it is highly unlikely as OpenAI is looking to monetise its technology.
However, its GPT-2 is open source.
Top stories of the week >>
Can AI develop a personality?
A couple of weeks back, Google suspended an engineer, Blake Lemoine, for claiming AI has become sentient. The incident triggered a debate on whether AI has become conscious already. Meanwhile, we went down a different rabbit hole and sounded out a few technologists to understand if AI is capable of developing a personality.
Read the complete story here.
Why Quickbooks did not work out in India
Intuit recently decided to pull the plug on its QuickBooks products from India. Read the complete story to understand why it discontinued its services nearly after a decade of operation.
How does YOLOv6 compare against YOLOv5?
Computer vision is one of the most buzzing fields of AI. Major companies are dedicating massive amounts of resources to launch the next big thing in this field. Last week, Chinese company Meituan released YOLO v6. It is not part of the official YOLO series but was named since the authors of this architecture were heavily inspired by the original one-stage YOLO.
How is it different from YOLOv5? Click here to know.
People & Technology >>
Amitabh Kant: The legacy of NITI Aayog’s outgoing CEO
And it’s a wrap after over six years. Amitabh Kant’s long tenure (after three extensions) at NITI Aayog has finally ended. Parameswaran Iyer has now been appointed as the Chief Executive Officer (CEO) of NITI Aayog for two years.
Check out Kant’s successful stint here.
When Narayana Murthy endorsed tech for the Indian cricket team
Javagal Srinath has been one of the proprietors of promoting technology in cricket. Recalling his early days at Machinecon 2022, Srinath said that Narayana Murthy – who also happened to be his neighbour – helped endorse the tech they had built two decades ago.
Read the complete story here.
AIM Videos >>
Techies on the streets of Bengaluru
Here’s the first episode of OYE TECHIE, a voice pop series that captures the essence of techies in Namma Bengaluru. Check out the video below:
Startups, Apps & Tools >>
The company is working on creating new generative image models similar to OpenAI’s DALLE-2.
An AI model that can draw images from any prompt.
An AI content generator that delivers premium results in seconds.
Research & Papers >>
Deepnash: Mastering the Game of Stratego with Model-Free Multiagent Reinforcement Learning
DeepMind researchers recently introduced Deepnash, an autonomous agent capable of learning to play Stratego from scratch, up to a human expert level. Stratego is a strategy board game that AI has not yet mastered. Check out the research paper for more details.
A Path Towards Autonomous Machine Intelligence
Yann LeCun shared his vision in this latest paper, which distils much of his thinking over the last five to ten years about promising directions in AI, and discussed his plans to inspire future work in the field over the next decade.
Check out the paper here.
Events in Focus >>
How to achieve real-time AI inference on your CPU
Intel, in collaboration with Analytics India Magazine, is organising this webinar for the attendees to understand how to optimise a deep learning neural network model and achieve fast AI inference with a CPU.
Click here to register.
Top News of the Week