๐Ÿ“ข Last Week in AI(LWAI) - March 4th to March 10th

Author

Kurian Benoy

Published

March 12, 2024

๐Ÿ—“๏ธ Monday:

The Government of Indiaโ€™s Ministry of Electronics and Information Technology (MeiTY) made an announcement regarding the deployment of ML models in production. While it was initially interpreted as a requirement for all, they clarified that it applies mainly to large organizations and not startups or Indie Hackers. This topic is still in the early stages of discussion, and we can expect further regulation in the future.

Tweet Link:

Meme about getting permission for open source models hosted in Huggingface from Govt of India

๐Ÿ—“๏ธ Tuesday:

The AI4Bharat team released IndicVoices, a dataset containing natural and spontaneous speech, including read, extempore, and conversational audio, contributed by 16,237 individuals. This dataset is a significant step towards advancing AI in the Indian context.

Paper Link:

Blogpost Link:

๐Ÿ—“๏ธ Wednesday:

1๏ธโƒฃ Telugu LLM Labs introduced Indic Gemma 7B/2B Instruction tuned LLM model - Naverasa for 9 Indic languages. Developed by Ravi Theja Desetty, a developer advocate at LlamaIndex, this model is a game-changer for Indic language processing.

Models Link:

Github Repo Link:

Blogpost Link:

Meme about Indie hacker building Indic LLM Navaresa which supports 10 languages

2๏ธโƒฃ Surya which is a document OCR toolkit that does accurate OCR in 90+ languages and Line-level text detection in any language came with awesome benchmarking results. They benchmarked against Google OCR and results looks really competitive which beats them even in few languages. More power to Open source.

Github Repo Link:

Tweet Link:

Benchmarking score of Surya vs google cloud OCR

๐Ÿ—“๏ธ Thursday: Two Kerala-origin open-source developers, elopers, and Jithin James , hit a home run with their YC-funded company, ragas_io. Y Combinator announced that ragas_io is an open-source evaluation and testing infrastructure for LLM applications.

Tweet Link:

Shahul ES and Jithin James posing in front of Y Combinator building

๐Ÿ—“๏ธ Friday: Answer.AI, a rising star in the startup scene, announced a collaboration with Tim Dettmers, Huggingface, and Mobius Labs. Together, they are releasing a groundbreaking project that enables efficient training of very large models (70b) on consumer gaming GPUs using FSDP/QLoRA.

Blog Link:

๐Ÿ—“๏ธ Saturday:

1๏ธโƒฃ Simon Wilson, renowned LLM expert and co-creator of Django framework, shared an insightful blog post confirming that the GPT-4 barrier has finally been broken.

Blog Link:

2๏ธโƒฃ Santhosh Thottingal delivered an intriguing talk on digital Malayalam typography. While not directly related to AI, font designing is a complex art that may be even more intricate than LLMs.

Video Link:

Thatโ€™s Kurian reporting about last week. Hope you enjoyed it. Let me know if I should continue doing this :P.