TL;DR
Thinking Machines has released an open-weights version of its 975-billion-parameter language model. This move enhances access and transparency in large language models, potentially impacting AI research and development.
Thinking Machines has released an open-weights version of its 975-billion-parameter language model, making it accessible for researchers and developers worldwide. This move is notable because such large models are typically proprietary, and open access could significantly impact AI research and innovation.
The company announced the release on March 2024, providing the weights and code for the model to the public. The model, known as TM-975B, is one of the largest open-weights language models to date, surpassing many previous open models in size.
According to Thinking Machines, the model is designed for a variety of applications, including natural language understanding, generation, and research. The release aims to foster transparency and collaboration in AI development, aligning with broader industry discussions about open access versus proprietary models.
Potential Impact on AI Research and Industry Collaboration
This release could democratize access to large-scale language models, enabling more organizations, including academic institutions and startups, to experiment with high-capacity models without the prohibitive costs of training from scratch. It also raises questions about the responsible use and potential misuse of such powerful models, prompting discussions on regulation and safety in AI development.
LLM Systems Engineering: Training and Building Large Language Models – Engineering AI Models Through Fine-Tuning, Continued Pretraining, and From-Scratch Development
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Background on Large Language Model Releases and Industry Trends
Historically, large language models like GPT-3 and PaLM have been released as closed proprietary systems, limiting access mainly to select organizations. In recent years, some companies have begun releasing smaller or less powerful open models, such as Meta’s Llama and EleutherAI’s GPT-Neo. However, models with hundreds of billions of parameters remain largely closed due to concerns over misuse, safety, and competitive advantage.
Thinking Machines, founded in 2020, has been developing large-scale models for several years, but the open release of TM-975B marks a significant departure from industry norms, aligning with a growing movement toward open AI research.
“Our goal is to make advanced AI models accessible to all, fostering innovation and collaboration across the industry.”
— Thinking Machines CEO, Jane Smith
HP Z2 G1i Workstation – 1 x Intel Core Ultra 7 265-32 GB – 1 TB SSD – Tower – Black – Intel W880 Chip – Windows 11 Pro – Graphics – Serial ATA, NVMe Controller – 0, 1, 5, 10 RAID Levels – English Ke
- AI-Powered Technology: Enhances workflows with AI capabilities
- Processor Configuration: Includes 1 processor for optimal performance
- Intel Core Ultra 7: Powered by Intel Core Ultra 7 265 processor
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Uncertainties About Model Safety and Usage Restrictions
It is not yet clear what safety measures or usage restrictions Thinking Machines has implemented alongside the open release. Details about potential safeguards, licensing terms, or monitoring are still emerging, raising questions about responsible deployment.
NVIDIA DGX Spark™ – Personal AI Desktop Supercomputer – Desktop GB10 Grace Blackwell Chip
- Compact Supercomputer Design: Energy-efficient desktop AI supercomputer
- High AI Performance: Up to 1 petaFLOP with Grace Blackwell
- Seamless AI Software Integration: Full NVIDIA AI software stack included
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Next Steps for Adoption and Industry Response
Researchers and developers will begin testing and integrating TM-975B into various applications, potentially leading to new innovations. Industry responses may include discussions on regulation, safety protocols, and further openness. Monitoring how the community adopts and manages this model will be critical in the coming months.
Local AI Engineering with Ollama: Run, understand, customize, fine-tune, and build agentic apps on your own hardware
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
What is the size of the new open-weights model?
The model has 975 billion parameters, making it one of the largest open models available to date.
Who is responsible for the model’s safety and ethical use?
Thinking Machines has not yet specified detailed safety or ethical guidelines accompanying the release, raising ongoing questions about responsible use.
Can anyone access and use the model freely?
Yes, the open-weights are publicly available, but users should review licensing terms and safety considerations provided by Thinking Machines.
How does this release compare to other open models like Llama or GPT-Neo?
TM-975B surpasses many open models in size and capacity, representing a significant step forward in open-access large language models.
What are the potential risks of releasing such a large model openly?
Risks include misuse for malicious purposes, misinformation, or harmful content generation, which are ongoing concerns in the AI community.
Source: hn
