Make the agent’s backend easier to inspect
Latent Space interviewed Supabase’s founders about agent-friendly backends. Local runtime and compute features discussed include alpha releases.
Was this useful to you?
Browse news and guides for your profession or interests.
Latent Space interviewed Supabase’s founders about agent-friendly backends. Local runtime and compute features discussed include alpha releases.
Was this useful to you?
Google announced Carbon, initially for selected cyber defenders. Broader access remains planned.
Was this useful to you?
OpenAI introduced GPT-6.1 Terra for visual reasoning and computer-based tasks. OpenAI reports stronger coding and fewer factual errors; its evaluations do not establish error-free work.
Was this useful to you?
A review of 28 studies reports that agricultural TinyML deployments largely rely on ESP32, STM32, and ATMega microcontrollers. Some plant-disease detection systems report 90–97% accuracy and latency as low as 7.6 ms, but memory and energy metrics are often missing. The review says models typically run inference on-device in the field while training remains in the cloud. Quantization appears in over half of deployments, whereas structured pruning and neural architecture search are rare.
Was this useful to you?
Prism ML’s ternary-weight Ternary-Bonsai-2-2B reduces the language model to about 0.52 GB, versus roughly 3.8 GB at FP16. It retains a 262K-token context window and image input, with the vision component loaded separately. Weights take values of −1, 0, or +1; additional scale metadata raises the actual storage cost. Image input requires an optional 0.36 GB Q8_0 mmproj file.
Was this useful to you?
OpenGVLab released the 0.9-billion-parameter InternVL2-1B vision-language model under the MIT license. It reportedly scores 754 on OCRBench, above PaliGemma-3B at 614 and Mini-InternVL-2B at 654. The model combines an InternViT-300M vision encoder with Qwen2-0.5B-Instruct and has an 8K context window. It targets tasks including OCR, document and chart analysis, and video question answering.
Was this useful to you?
Crusoe confirmed it sold an unfinished data center project in southeastern Wyoming to Google. Google said it plans to invest $17 billion to develop the campus; the purchase price was not disclosed. Crusoe announced the project last year with Tallgrass, with plans to eventually scale it to 10 gigawatts. Google is taking over development; the report says it had raised concerns about costs and the construction timeline under Crusoe’s management.
Was this useful to you?
Hugging Face described native TPU support through the TorchTPU backend. The integration brings Transformers, Diffusers and Kernels into PyTorch workflows using TPU devices.
Was this useful to you?