• The 79
  • Posts
  • This new AI startup is worth $32B with no product

This new AI startup is worth $32B with no product

Welcome back everyone! Here’s what you need to know about AI today:

👉 ByteDance releases a new reasoning AI model

👉 Ilya Sutskever’s AI startup reaches $32b in valuation with no products

👉 OpenAI may require verified ID for its API use in the future

and many more!

đź“§ Did someone forward you this email? Subscribe here for free to get the latest AI news everyday!

Read time: 4.9 minutes

BYTEDANCE

ByteDance enters AI reasoning race with a new powerful model

Source: ByteDance’s GitHub | The darker blue dashed bar represents the performance of Seed-Thinking-v1.5

What’s going on: ByteDance, TikTok’s parent company, has introduced Seed-Thinking-v1.5, a new large language model designed to compete in the growing field of reasoning-focused AI. Unlike standard models that prioritize quick responses, Seed-Thinking-v1.5 emphasizes structured reasoning, using a chain-of-thought approach to break down complex problems before answering. The model leverages a Mixture-of-Experts architecture, activating only 20 billion of its 200 billion parameters at a time, which boosts efficiency without sacrificing performance.

What does it mean: Every huge tech company is releasing new reasoning AI models these days. What sets ByteDance apart is their training approach which relies on custom reinforcement learning frameworks, VAPO and DAPO, stabilizing long reasoning processes. Two verification tools, Seed-Verifier for checking mathematical equivalence and Seed-Thinking-Verifier for step-by-step judgment, enhance the model’s reliability by reducing errors and reward manipulation. Built on ByteDance’s HybridFlow system with Ray clusters, their training setup minimizes GPU downtime effectively.

More details:

  • While not yet available for public use, its technical paper reveals strong results across benchmarks, positioning it as a rival to models like OpenAI’s o3-mini-high and Google’s Gemini 2.5 Pro.

  • The model performs well on challenging tasks, scoring 86.7% on the AIME 2024 math exam, 55% on Codeforces programming challenges, and 77.3% on the GPQA science benchmark. It also shows versatility, with an 8% higher win rate over DeepSeek R1 in non-reasoning tasks.

  • Interested in the technical details? Read their paper.

SSI

Safe Superintelligence reaches $32B valuation with no products

Source: CTech | Ilya Sutskever

What’s going on: Safe Superintelligence (SSI), founded by former OpenAI chief scientist Ilya Sutskever, has secured $2 billion in new funding, valuing the AI startup at $32 billion. This round follows a previous $1 billion raise, which shows intense investor enthusiasm for AI ventures despite SSI lacking a real public product.

What does it mean: Sutskever, who left OpenAI after a failed attempt to remove CEO Sam Altman, started SSI with Daniel Gross, ex-Apple AI lead, and Daniel Levy, a former OpenAI researcher. The company’s sole focus is developing a “safe superintelligence” designed to exceed human capabilities while prioritizing safety.

More details:

  • SSI operates with a lean team of about 20 employees across Palo Alto and Tel Aviv, using Google Cloud’s TPU chips for its research.

  • Investors, including Alphabet, Nvidia, Andreessen Horowitz, and Lightspeed Venture Partners, are betting on Sutskever’s track record in advancing AI breakthroughs at OpenAI.

  • No technical details or models have been released, and SSI remains secretive about its progress.

🆔 OpenAI has plans to implement an ID verification process for organizations accessing advanced future AI models via its API to enhance security and prevent misuse, requiring government-issued IDs and limiting verification to one organization every 90 days.

📊 Meta's unmodified Llama-4 Maverick AI model ranked significantly lower than competitors like OpenAI's GPT-4o, Anthropic's Claude 3.5 Sonnet, and Google's Gemini 1.5 Pro on the LM Arena benchmark after it was revealed that Meta had initially used an optimized, experimental version to achieve a higher score.

âš– Former OpenAI employees have filed an amicus brief supporting Elon Musk's lawsuit against OpenAI, arguing that the company's transition to a for-profit structure would violate its original mission to prioritize the safety and broad benefits of AI over financial interests.

📱 ChatGPT became the most downloaded non-game app worldwide in March, with a 28% increase in installs from February, totaling 46 million downloads, surpassing Instagram and TikTok.

AI + Digital marketing

Recommend a marketing funnel for a freelance graphic designer targeting small businesses, using email, social media, and a portfolio site, with specific actions for each stage.

Gemini 2.0 Flash’s answer

Thank you for staying with us like always! If you are not subscribed, subscribe here for free to get more of these emails in your inbox! Cheers!