TED AI, Strawberry Project, LLM Ganga-1B

In today's edition of Superintelligence, among other development sin the world of AI, we cover AI edtech success in the US, Defense AI startup's solution for european security, and Japanese AI chip market expansion.

The AI World Today

  • Multilingual AI-Enhanced TED

  • OpenAI's Strawberry Tech Unveiled

  • Chinese AI Apps Succeed In US

  • Japanese AI Chip Market Expansion

  • Defense AI Startup’s $490M Funding

  • LLM Ganga-1B Hindi AI Launched

    +

  • AI Spotlight

  • Heads Up

  • Save the Date

  • Good Reads

TED Talks Overcoming Language Barriers with AI

Screenshot: TED/Youtube

TED is launching a pilot program to overcome language barriers by integrating generative AI for voice-cloning and lip-syncing. In collaboration with Panjaya.ai and TED’s global Translators community, this initiative will deliver TED Talks in Brazilian Portuguese, French, German, Italian, and Spanish languages while preserving the speaker's unique voice and style. This new approach enhances inclusivity and aims to provide a seamless viewing experience, marking a significant advancement in multilingual content delivery. This initiative builds on the success of TED’s subtitling program, combining human and artificial intelligence to make ideas accessible to a global audience.

OpenAI Develops Strawberry for Advanced AI Reasoning

Image: OpenAI

OpenAI is developing a new AI technology called "Strawberry," aiming to enhance its models' reasoning capabilities. The project, detailed in internal documents, focuses on enabling AI to autonomously navigate the internet and perform "deep research." Strawberry involves specialized post-training methods to improve AI performance. This initiative is part of OpenAI's strategy to achieve advanced reasoning abilities, which is crucial for creating AI with human-like intelligence. The project signifies a step towards advanced AI systems capable of long-term planning and complex problem-solving, positioning OpenAI at the forefront of AI innovation.

Chinese EdTech AI Apps Gain Popularity in US

AI Generated Illustration

Chinese AI education apps Question.AI and Gauth have gained traction in the US amidst fierce competition in China. With over 200 large language models developed domestically, more Chinese developers are targeting Western markets. Question.AI and Gauth, which use generative AI to assist with homework, are among the top educational apps in the US. Question.AI launched in mid-2023, and Gauth started in 2020 as a math solver. Both offer instant solutions and explanations. Their success highlights the advanced technology developed in China and the growing influence of Chinese AI in the global education sector.

SoftBank Buys Graphcore in Major AI Deal

Image: Graphcore

SoftBank has acquired British AI chip startup Graphcore for over $600 million, significantly below Graphcore’s 2020 peak valuation of $2.8 billion. This move reflects SoftBank owner Masayoshi Son's ambition to position the company as a key player in AI technology. The acquisition aims to replicate the success of SoftBank's previous investment in Arm, another British chip company, which has tripled in value since going public last year. Breaking into the AI chip industry will be challenging, given Nvidia's dominance, controlling about 80% of the market.

German Startup Secures $490M for NATO Protection

Image: Axios

German defense-tech startup Helsing has raised €450 million in venture capital, valuing the company at €4.95 billion. The funding round, led by General Catalyst, also included Elad Gil, Accel, Saab, Lightspeed, Plural, and Greenoaks. This makes Helsing one of Europe’s most valuable startups, as the company focuses on enhancing NATO's Eastern flank amid increased defense investment following Russia’s invasion of Ukraine. Helsing develops software to boost the capabilities of weapons systems like drones and jet fighters. The company also plans to spend €70 million on Baltic defense projects over the next three years and has opened a new entity in Estonia.

IIT Gandhinagar Debuts Ganga-1B for Hindi Language

Image: Generated/Superintelligence

IIT Gandhinagar's Lingo Research Group has launched Ganga-1B, a large language model (LLM) designed for Hindi, surpassing existing models in performance. Part of Project Unity, it was trained on diverse Hindi public-domain data. Ganga-1B's specifications include 16 layers, 32 attention heads, and a vocabulary size of 30,000. It achieved over 600 downloads within 48 hours of release. The model aims to enhance AI capabilities in Indian languages, with future expansions planned for Tamil, Telugu, Marathi, Gujarati, and Urdu, supporting e-governance and education sectors.

AI Spotlight

Advances in Mathematical Reasoning and AI

Image: Illustration/Superintelligence

On July 11, at least five publications were published that highlight significant advancements in mathematical reasoning and AI.

"Skywork-Math" explores data scaling laws for enhancing the mathematical reasoning capabilities of large language models, making them more efficient at solving complex problems.

"MAVIS" presents mathematical visual instruction tuning, which leverages visual data to improve the accuracy of mathematical models, thus enabling better performance in visual and numerical tasks.

"Is Your Model Really A Good Math Reasoner?" provides a comprehensive checklist for evaluating the mathematical reasoning abilities of AI models, ensuring robust assessment methods.

"Self-Recognition in Language Models" investigates self-awareness in AI, which can influence how models approach and solve mathematical problems, contributing to more intuitive AI reasoning.

"Scaling Up Personalized Aesthetic Assessment" examines task vector customization for aesthetic evaluations, which, while seemingly unrelated, uses mathematical precision in evaluating visual aesthetics, demonstrating the interdisciplinary nature of mathematical AI.

Why They Matter: These papers advance AI's mathematical reasoning, visual understanding, self-recognition, and task-specific customization, enhancing overall AI performance and accuracy in complex, interdisciplinary tasks. This research could revolutionize fields requiring precise mathematical computation, such as engineering, economics, and scientific research.

Heads Up

AI Service: Meet Rufus, Amazon’s Generative AI-Powered Shopping Assistant

AI Invest : Canada Invests $15m in AI Health Care Initiative

AI Funding: Estonian-American AI startup Pactum Raises USD20 Million

AI Partnership: SAP Business AI and CleaVision partner for advanced ROP diagnosis

AI Funding: U.S. military to Award $3 Billion Contract for AI-Driven Intelligence

AI Legal: Canadian Federal Court Asked to Decide If AI Can Be Author in New Copyright Case

AI Defence: US Marines Release New AI Strategy

Save the Date

July 25 -26: Nairobi Hosts International Conference on Artificial General Intelligence and Consciousness

Good reads

June 25: GEN AI: Too Much Spend, Too Little Benefit?