Indian AI Startups Rival Global Giants 🌟

PLUS: 7B Model Outperforms ChatGPT and Grok, LMs Can Teach Themselves to Use Tools

Today’s top AI Highlights:

1. Indian AI Startups Launch 3 New AI Chatbots for Indian Market

2. 7B Model Outperforms ChatGPT and Grok

3. Toolformer: Language Models Can Teach Themselves to Use Tools

4. OpenAI Releases a Preparedness Framework to Protect against “Catastrophic Risk”

& so much more!

Read time: 3 mins

Latest Developments 🌍

Embracing Regional Diversity and Technology 🗺️

Indian startups are carving a unique niche in the AI sector, successfully competing with global players by focusing on regional needs and linguistic diversity. These companies are not just adapting existing AI technologies, but are also innovating to cater to the specific requirements of the Indian market. The recent three announcements demonstrate their efforts and strong position:

1. Kissan AI has launched Dhenu 1.0, an agricultural generative AI chatbot for personalized crop advice to Indian farmers. Tailored for India's diverse landscape and trained on bilingual in English and Hindi, high-quality, domain-specific conversations, the model ensures inclusivity and accessibility.

2. Krutrim Si Designs, led by Ola, announced 'Krutrim AI', an AI chatbot that comes in two versions: the base model and Krutrim Pro. The name means 'artificial' in Sanskrit. Krutrim understands 20 Indian languages and is capable of generating content in ten, including Hindi, Kannada, and Marathi. Trained on over 2 trillion tokens, it surpasses GPT-4 in Indic language support and outperforms similar open-source LLMs in industry-standard benchmarks.

3. CoRover.ai, in collaboration with Google Cloud, has announced the launch of BharatGPT, a new generative AI chatbot. It supports over 14 Indian languages and offers capabilities for text, voice, and video interactions. BharatGPT integrates advanced features like Aadhar-based KYC authentication, sentiment analysis, and word embedding techniques.

World’s Best Opensource 7B Model 🏆

OneAI has released OpenChat-3.5-1210 is an upgrade to OpenChat-3.5, with a particular emphasis on enhancing coding performance. The model shows a near 15-point increase on the HumanEval benchmark, while also maintaining or improving performance on other benchmarks. Touted as the "World's Best Open Source 7B LLM", OpenChat-3.5-1210 surpasses ChatGPT and xAI’s Grok.

Enhancing AI with Autonomous Tool Usage 👷‍♂️

LMs are capable of impressive zero-shot and few-shot results on various tasks but have inherent limitations. They can't access up-to-date information, tend to produce factual inaccuracies, struggle with low-resource languages, lack precise mathematical abilities, and are unaware of time progression. Addressing the issue, researchers at Meta have developed a new model known as Toolformer, which enables these LMs to autonomously learn and use external tools.

Key Highlights:

1. Toolformer is designed for self-supervised learning, enabling it to autonomously decide when and how to use a variety of tools such as calculators, search engines, and translation systems. This approach significantly reduces the need for human annotations and helps overcome the above challenges.

2. Leveraging a 6.7B parameter GPT-J model, Toolformer demonstrates a notable improvement in zero-shot performance, exceeding larger models like GPT-3 in various tasks. This enhancement is achieved without compromising its fundamental language modeling capabilities.

3. Toolformer's architecture allows it to leverage a range of tools, for instance, it uses Atlas for QA, a basic arithmetic calculator, a Wikipedia search engine for comprehensive information retrieval, a machine translation system capable of handling 200 languages, and a calendar API for temporal context.

Safety Driven by Science and Grounded in Facts 💡

OpenAI has recently introduced its Preparedness Framework, a comprehensive strategy aimed at managing and mitigating risks associated with advanced AI models. This initiative involves meticulous evaluation and continuous updating of risk "scorecards" for assessing risks at each stage of model development, particularly at every 2x effective compute increase. The aim is to ensure that only those with acceptable post-mitigation risk levels are developed or deployed.

The framework emphasizes real-world testing, rigorous capability evaluations, and data-driven predictions to proactively identify and address potential safety concerns.

In addition to technical assessments, the framework establishes a dedicated Preparedness Team responsible for overseeing safety evaluations and synthesizing reports.

---

Tools of the Trade ⚒️

1. XMind AI: An advanced mind mapping tool leveraging AI for idea generation, brainstorming, and project planning, offering features like real-time collaboration, auto-generated slides, and cloud-based data access.

2. Rightsify: Generate unique, copyright-cleared music using AI. The generated music can be used for a wide range of purposes, like background music for content creators, soundtracks for TV, film, and gaming, music in hospitality and entertainment venues, and even for personal productivity.

3. Palet: Palet incorporates AI into the website design and coding process, eliminating the need for manual code insertion from ChatGPT. It boasts UI for the GSAP animation library, allowing users to create animations without coding. Additionally, it supports switching between UI and code editing seamlessly, providing flexibility and ease of customization.

4. Vexa Search: Search images and get detailed information, just like Google Lens. Uses Gemini API.
   It’ll be exciting to see what can be built upon this stack.

😍 Enjoying so far, TWEET NOW to share with your friends!

---

Hot Takes 🔥

1. More and more, it seems like inference and even model architecture is becoming commoditized, and the next war is going to be for proprietary datasets. ~ Charlie Guo

2. In AI, the ratio of attention on hypothetical, future, forms of harm to actual, current, realized forms of harm seems out of whack.
   Many of the hypothetical forms of harm, like AI "taking over", are based on highly questionable hypotheses about what technology that does not currently exist might do. ~ Andrew Ng

---

Meme of the Day 🤡

That’s all for today!

See you tomorrow with more such AI-filled content. Don’t forget to subscribe and give your feedback below 👇

Real-time AI Updates 🚨

⚡️ Follow me on Twitter @Saboo_Shubham for lightning-fast AI updates and never miss what’s trending!!

PS: I curate this AI newsletter every day for FREE, your support is what keeps me going. If you find value in what you read, share it with your friends by clicking the share button below!