Sunday Snaps #07

Meta's Code Generation Tool, Snapchat's Multilingual AI ChatBot, Meta's Self-designed Silicon chips and Sam Altman's Call for Regulating AI

May 21, 2023

Meta' CodeCompose: Yet Another Code Generation Tool?

Meta has announced the development of CodeCompose, a generative AI tool for coding similar to GitHub's Copilot. Currently used internally by Meta's teams, CodeCompose provides code suggestions for Python and other languages in IDEs like VS Code. The largest CodeCompose model has 6.7 billion parameters and is fine-tuned on Meta's first-party code, incorporating internal libraries and frameworks. It offers suggestions such as annotations, import statements, and even entire chunks of code, utilizing the surrounding code and comments for better accuracy. Meta claims a high acceptance rate of over 20% among its employees.

While CodeCompose's training data includes public code with permissive licenses and StackOverflow content, it's unclear if it was inadvertently trained on copyrighted code. Meta emphasizes that developers have the freedom to disregard CodeCompose's suggestions, and security was a key consideration during the model's creation. However, concerns have been raised regarding generative coding tools introducing security vulnerabilities in developed applications. Despite the controversies surrounding code-generating AI, Meta is optimistic about CodeCompose's progress and intends to keep its development in-house.

An AI ChatBot That Talks in Hindi as well

Snapchat’s AI ChatBot is one of the first in the industry to have the capability of talking in more than just the standard language of the industry i.e. English.

Although it’s also powered by GPT models, the replies from ‘My AI’ feel a lot more human than ChatGPT or Bing AI.

“My AI can recommend birthday gift ideas for your BFF, plan a hiking trip for a long weekend, suggest a recipe for dinner, or even write a haiku about cheese for your cheddar-obsessed pal,” the company wrote in a blog post. “Make My AI your own by giving it a name and customizing the wallpaper for your Chat.”

After Apple, Amazon, Google and Microsoft, Meta Now Builds Its Own AI Chips

Meta has introduced its first custom-designed AI chip called MTIA (Meta Training and Inference Accelerator) for optimized AI workloads.

The ASIC chip is built on TSMC's 7nm process, providing high-performance capabilities such as 102.4 TOPS at INT8 precision and 51.2 TFLOPS at FP16 precision, while operating at a 25W TDP. By developing its own chip, Meta aims to maximize architectural efficiency, reduce power consumption, and lower costs. The in-house chip enables performance analysis, fine-tuning, and efficient deployment of models, ultimately accelerating software development cycles and improving user experience.

Meta has also developed a compiler technology that runs under the PyTorch environment to enhance the chip's performance and power efficiency compared to GPUs.

Meta's initiative aligns with a trend among tech giants, including Microsoft, Google, Apple, and Amazon, who are also investing in developing in-house AI chips. These companies are leveraging customized hardware solutions to optimize their workloads and enhance AI processing capabilities. For example, Google has its TPUs, Apple has been working on the M1 and M2 chips, and Amazon is focused on Trainium and Inferentia processors.

With their in-house designs, these companies aim to have greater control over the architecture and align it with their future workload requirements.

CEO of OpenAI calls for US to regulate artificial intelligence

Sam Altman, CEO of OpenAI, testified before a US Senate committee, advocating for the regulation of artificial intelligence (AI). Altman called for the establishment of a new agency to license AI companies and expressed concerns about the potential dangers of AI technology, including misinformation during elections.

He acknowledged the impact of AI on jobs but expressed optimism about the quality of future jobs. Altman suggested a combination of licensing, testing requirements, and independent audits for AI companies to ensure responsible development and deployment of AI models.

Both Republican and Democrat senators showed bipartisan support for regulating the AI industry, recognizing the need to maximize the benefits while minimizing the risks. However, the fast pace of technological advancements raised concerns about the agency's ability to keep up with the rapidly evolving field of AI.

Altman's testimony highlighted the importance of government collaboration with the industry to prevent negative consequences and shape the future of AI technology.

Check out the previous issues of Sunday Snaps:

Sunday Snaps 🚀

Sunday Snaps #06

Harshit RV

May 14, 2023

This was the week of Google I/O which shed light on lot of new and innovative products Google has been working on since the last few years including Bard, PaLM 2, a more robust Firebase hosting among other things. And well, Google is putting AI in basically everything, including search

Read full story