Bilingual

Own Your Weights
15 days ago
- API dependence poses risks: model behavior, pricing, and availability can change unexpectedly.
- Open weights enable control: running models on personal hardware avoids external dependencies.
- Balance strategy: rent frontier models for cutting-edge needs, own baseline models for stability.
- Sovereignty stack includes controlling data, weights, and infrastructure to maintain autonomy.
- Establish local fallbacks for critical workflows to ensure resilience and self-reliance.
Inkling: Our Open-Weights Model
8 days ago
- The company has released a new AI model called Inkling, which is a Mixture-of-Experts transformer with 975B total parameters and 41B active parameters.
- Inkling supports a context window of up to 1 million tokens, was pretrained on 45 trillion tokens of multimodal data, and is available for fine-tuning via the Tinker platform.
- Key features of Inkling include multimodal reasoning (text, images, audio), controllable thinking effort for balancing performance and efficiency, and strong performance across reasoning, agentic coding, and safety benchmarks.
- Alongside Inkling, a preview of Inkling-Small (276B total parameters, 12B active) is shared, offering similar capabilities with lower cost and latency for certain workloads.
- The model's development involved large-scale RL training (over 30M rollouts), which improved reasoning and led to more concise reasoning styles.
- Inkling is designed as a broad foundation model for customization, emphasizing calibration, instruction following, and resistance to censorship (epistemics).
- Safety evaluations show Inkling has strong safeguards, performing well on benchmarks like FORTRESS and StrongREJECT.
- Inkling is available through various platforms and partners, including TogetherAI, Hugging Face, and others, with full weights released for customization.
American AI is locked down and proprietary. It's losing
3 days ago
- US AI strategy relies on proprietary, locked-down models, while China adopts an open-weights approach.
- AI models lack strong moats; switching between them is easy via APIs, making enterprise services the real differentiator.
- China's open models exploit US export controls by turning compute disadvantages into distribution advantages and commoditizing the AI layer.
- The performance gap between US frontier models and open models is closing, with Chinese models claiming comparable capabilities at lower costs.
- US companies prioritize short-term profits and face misaligned incentives, risking economic fallout if AI spending declines due to ecosystem weaknesses.
- Open technologies foster permissionless innovation and ecosystem benefits, contrasting with locked-down practices that limit global adoption.
Kimi K3 first open model in ErrataBench Top 10
a day ago
- Kimi K3 is the first open weights model to score over 90% on the ErrataBench benchmark.
- It outperforms Opus 4.8 in both performance and price, approaching 5.6 Terra-level performance.
- For local models with sufficient hardware, Kimi K3 is the best choice for English writing and editing tasks.
- ErrataBench evaluates models' ability to detect errors in large text samples and apply valid fixes using tools.
- This milestone marks the first time an open model has entered the top 10 of the benchmark, previously dominated by proprietary models.
- Note: Kimi K3 is described as a 'soon-to-be-open' model.

Hasty Briefsbeta

#open weights

Own Your Weights

Inkling: Our Open-Weights Model

American AI is locked down and proprietary. It's losing

Kimi K3 first open model in ErrataBench Top 10