Machine learning models
Source reports covering Machine learning models. Stories will appear when analysis and community reactions are ready.
5 reports · page 1 of 1
Connected topics are temporarily unavailable. You can still browse stories below.
Source reports
Beating GPT-5.6 Sol on retrieval with 100x cheaper open models
tool
5 min read
Castform utilizes Neon to efficiently transform raw data into usable formats, allowing agents to read, search, and mutate data at scale. A "good agent" requires strong capabilities in context to find relevant data and in model performance to determine search criteria.
Run Kimi K3 using 29 GB of RAM at 0.50 tok/s
tool
12 min read
WASTE is an embeddable C inference engine that allows the Kimi K3 model, which has 2.78 trillion parameters, to run on consumer laptops by streaming activated weights directly from NVMe storage. It operates without third-party runtime dependencies and manages memory by keeping the model trunk in memory while streaming selected experts from disk.
Self-hosting Kimi K3: 20% more hardware cost, 20% better task resolution
tool
25 min read
Kimi K3 requires an 8ÃB300 node with 288GB of HBM per GPU due to its 1.4TB weight, exceeding the memory capacity of the 8ÃB200 node. This configuration results in approximately 20% higher hardware costs while supporting 16 concurrent sessions.
Transformer Transformer: A Unified Model for Motion-Conditioned Robot Co-Design
research
7 min read
Transformer Transformer is a unified model that generates a complete robot design optimized for specific tasks based on manipulation demonstrations. A prototype designed for cloth flinging on an ALOHA2 bimanual platform achieved a 73% reduction in tracking error and a 30% decrease in maximum joint speed compared to the original design.
Small models also found the vulnerabilities that Mythos found
research
23 min read
Anthropic Mythos's showcase vulnerabilities were tested on small, inexpensive, open-weight models, revealing similar analysis results. AI cybersecurity capability varies significantly with model size, indicating that the security moat relies on the system architecture rather than the model itself.