LIVE PULSE
4.0 Anthropic CEO Amodei calls for slower AI development and shared safety rules — 11 src2.2 Agility Robotics unveils Digit 5 humanoid for warehouses and factories — 2 src2.0 Apple ships rebuilt Siri with Google Gemini, but not in the EU — 2 src1.8 Siri AI in macOS 27 Golden Gate: FAQ, Germany availability, privacy questions — 2 src1.4 Sam Altman says OpenAI will not go public in 2026, citing AI safety concerns — 5 src1.1 OpenAI contractors review real ChatGPT conversations to rate responses, report says — 2 src1.1 Anthropic data retention policy prompts firms to limit Claude use for sensitive work — 1 src1.1 Study examines issue bias in LLMs used as writing assistants before Swedish 2026 election — 1 src1.1 Study Audits Misalignment in Multi-Modal World Models — 1 src1.1 Retrieval-Grounded Reasoning Approach Proposed for Universal Multimodal Embeddings — 1 src4.0 Anthropic CEO Amodei calls for slower AI development and shared safety rules — 11 src2.2 Agility Robotics unveils Digit 5 humanoid for warehouses and factories — 2 src2.0 Apple ships rebuilt Siri with Google Gemini, but not in the EU — 2 src1.8 Siri AI in macOS 27 Golden Gate: FAQ, Germany availability, privacy questions — 2 src1.4 Sam Altman says OpenAI will not go public in 2026, citing AI safety concerns — 5 src1.1 OpenAI contractors review real ChatGPT conversations to rate responses, report says — 2 src1.1 Anthropic data retention policy prompts firms to limit Claude use for sensitive work — 1 src1.1 Study examines issue bias in LLMs used as writing assistants before Swedish 2026 election — 1 src1.1 Study Audits Misalignment in Multi-Modal World Models — 1 src1.1 Retrieval-Grounded Reasoning Approach Proposed for Universal Multimodal Embeddings — 1 src
HEATPULSEAI MAGAZINES
FLIP · FOLLOW · SAVE
Log inSign up

model quantization

topic3 events
papersSEP 10 04:00 UTC

Bit-flip attacks on vision-language-action models: vulnerability depends on action-decoding design

Researchers report the first bit-flip attack against a vision-language-action model, showing that Rowhammer-style faults can corrupt INT8 weights in deployed quantized systems. Using only a few gradient-selected bit flips, they drive the policy's closed-loop success rate down to near zero. The paper finds that a VLA model's action-decoding architecture largely determines how exposed it is to such faults.

security/safety/robotics/quantization
papersSEP 10 04:00 UTC

Divergence-based approach proposed to evaluate fidelity loss in quantized LLMs

A new arXiv paper argues that zero-shot task accuracy is an inadequate yardstick for quantized large language models, because it relies only on argmax predictions and hides changes in output distributions. The authors introduce a divergence-based method for measuring how much behavioral fidelity is lost when models undergo aggressive post-training compression for memory-constrained edge devices.

quantization/evaluation/llm/edge-devices
papersSEP 10 04:00 UTC

DiffLUT-Net Trains FPGA Lookup-Table Networks End-to-End with Learnable Connectivity

Researchers have introduced DiffLUT-Net, a framework that trains neural networks made of FPGA lookup tables directly through differentiable methods rather than converting pretrained quantized models. The approach also learns the connectivity structure of the LUT network, aiming to make hardware-efficient inference on FPGAs more effective. The work is available as a paper on arXiv.

fpga/hardware/research