Building an open AI company - with Ce and Vipul of Together AI

TL;DR

Together AI, a platform for open and independent AI systems, focuses on optimizing training and inference processes for greater efficiency and better models.

Transcript

hey everyone welcome to the Len space podcast this is alesio partner and CTO residents at deible partners and I'm joined by my co-host swix founder of small AI hey and today we have we're together with together welcome to the studio guys thank you thanks for having us maybe you guys want to I don't know how you typically give self- intros but um do... Read More

Key Insights

• 🤗 Together AI emphasizes the importance of open-source and collaboration in the AI community.
• 💦 They are working towards a more accessible and equitable data ecosystem for training AI models.
• 🔨 The company recognizes the significance of fine-tuning and provides tools and infrastructure to enhance model performance.
• 🥰 There is ongoing research and exploration in areas such as state-of-the-art models, hybrid architectures, and efficient embeddings.

Questions & Answers

Q: What is the origin and current focus of Together AI?

Together AI started with a focus on building a platform for open source, independent, and user-owned AI systems. The company aims to provide an alternative to big Labs developer platforms and has a strong emphasis on decentralization.

Q: How does Together AI compare to Apple's closed and polished approach?

Together AI's co-founder, Alesio, previously worked at Apple and has experience with open-source and collaborative projects. While Apple focuses on providing an excellent user experience, Together AI aims to provide open and accessible developer platforms while maintaining a high level of polish.

Q: What role does data movement play in Together AI's approach?

Together AI recognizes that data movement is a crucial aspect of AI systems. They have built a decentralized cloud infrastructure that combines data centers worldwide, allowing for improved efficiency and accessibility.

Q: How does Together AI address the challenge of data availability for training AI models?

Together AI has developed the "Red Pajama" dataset, which aims to provide accessible data for training AI models. The dataset is openly available and can be used for reproducibility and experimentation. The company is also exploring ways to incorporate other data sources beyond internet data to improve the availability and diversity of training data.

Q: What are the key insights from Together AI's discussion?

• Together AI is focused on open and independent AI systems.
• The company's cloud infrastructure is decentralized and disaggregated, allowing for improved efficiency.
• Fine-tuning is a significant aspect of Together AI's business, with many customers utilizing their platform for training and improving models.
• The company is investing in research and development, particularly in the areas of state-of-the-art models and innovative approaches like hybrid architectures.
• Together AI believes there is still much room for improvement in embeddings and training processes, with a focus on finding optimal mixtures of different architectures and datasets.

Summary & Key Takeaways

• Together AI, formerly known as TogetherXYZ, is a platform for open and independent AI systems.

• The company aims to build a platform for everything else outside of big Labs' developer platforms, focusing on open and user-owned AI models.

• Together AI utilizes a decentralized and disaggregated approach to building its cloud infrastructure, combining data centers worldwide.