Back to all articles
•Technology

Thinking Machines Lab Debuts Customizable AI Model Inkling

View original source

Thinking Machines Lab, founded by former OpenAI CTO Mira Murati, unveiled its first AI model, Inkling, on Wednesday. Unlike comparable models from OpenAI, Anthropic, or Google, Inkling features an open-weight design, allowing external developers to modify it. Inkling leverages a mixture-of-experts system with 975 billion parameters, but it efficiently uses only 41 billion for any task. It was trained on 45 trillion tokens across text, image, audio, and video, although outputs are currently text-based.

  • Inkling aims to provide organizations with a platform that offers customizable AI solutions, contrasting with general-purpose models like ChatGPT.
  • Thinking Machines promotes Inkling by emphasizing its ability to deliver calibrated responses and the flexibility of adjusting 'thinking effort' to balance speed and precision.
  • The introduction of Inkling serves as a proof of concept for the startup, aligning with their vision that adaptable AI will outperform fixed models.

Challenges and considerations include:

  • Customers are responsible for ensuring the safety of their AI customizations using the Tinker platform.
  • Financial and operational questions remain, especially concerning how Thinking Machines will cover significant costs without the subscription model used by competitors.
  • Despite hurdles, Thinking Machines emphasizes its rapid progress, achieving significant milestones in nine months, much faster than rivals like OpenAI.
  • Concerns about AI training methods, including potential use of outputs from competitors’ models, known as 'distillation.'

Industry reactions:

  • Companies like Microsoft and Hugging Face support Thinking Machines' approach, asserting that custom models prevent double costs and protect proprietary knowledge.
  • An experiment with Bridgewater Associates reportedly outperformed proprietary models and demonstrated substantial cost savings.

The article also briefly mentions the cultural approach within Thinking Machines, focusing on continuity over individual personalities. Despite several co-founder departures, the company maintains a strong stance on sustaining its innovative trajectory.