Launch of Tinker
Thinking Machines Lab, a heavily funded AI startup co‑founded by a group of prominent former OpenAI researchers, announced its first commercial offering, a platform named Tinker. Mira Murati, the lab’s co‑founder and chief executive officer, described the tool as a way to empower researchers, developers, and even hobbyists to create custom frontier‑level AI models without the need to manage large GPU clusters.
How Tinker Works
Tinker provides an application‑programming interface (API) that lets users fine‑tune two open‑source models—Meta’s Llama and Alibaba’s Qwen. Users can submit a few lines of code to start a fine‑tuning run that employs either supervised learning, where labeled data guides the model, or reinforcement learning, where the model receives positive or negative feedback based on its outputs. The platform abstracts away the low‑level details of distributed training while still giving users full control over the data they feed the model and the algorithms they apply.
Early Adoption and Feedback
The service is currently free, though the company expects to introduce pricing later. Access is being granted to a vetted group of beta users. Eric Gan, a researcher at Redwood Research, highlighted Tinker’s reinforcement‑learning capabilities, noting that the tool makes it markedly easier to coax specialized abilities out of a model—such as writing backdoors in code—than building an RL pipeline from scratch. Robert Nishihara, chief executive officer of Anyscale, praised the platform for striking a balance between abstraction and tunability, calling it a “great API” that many will want to use.
Safety and Governance
Thinking Machines Lab acknowledges the risk that open‑source models can be modified for malicious purposes. The company currently vets all applicants for API access and plans to develop automated systems to guard against misuse. Murati emphasized that the lab’s broader mission is to reverse the trend of increasingly closed commercial AI models by providing a transparent, research‑oriented alternative.
Funding and Vision
In a previous funding round, the startup secured $2 billion in seed capital, valuing the company at $12 billion. This substantial backing reflects investor confidence in the team’s expertise—many of whom were instrumental in creating ChatGPT—and in the market potential of democratizing frontier AI capabilities. Murati, who previously served as OpenAI’s chief technology officer and briefly as its chief executive officer, said the lab’s aim is to make frontier AI accessible to as many smart people as possible, fostering broader research and innovation.
Impact on the AI Landscape
The launch of Tinker positions Thinking Machines Lab as a major player in the emerging ecosystem of AI model fine‑tuning tools. By lowering the technical barriers to customizing powerful language models, the platform could accelerate the development of specialized applications across domains such as mathematics, legal drafting, and medical question answering. At the same time, the lab’s commitment to safety, open research, and community‑driven development offers a counterpoint to the dominant model of closed, proprietary AI services.
Este artículo fue escrito con la asistencia de IA.
News Factory SEO te ayuda a automatizar contenido de noticias para tu sitio.