AI Engineering Signal #62
GLM-5.2 tops Artificial Analysis open-weights intelligence index
Signals
GLM-5.2 tops Artificial Analysis open-weights intelligence index
benchmark against Claude Opus 4.6 coding tasks before next sprint; local deployment now viable.
Web
AMD silently removed memory encryption from consumer Ryzen CPUs
audit AGESA firmware versions before running sensitive model weights on Ryzen local inference hardware.
Web
US holds off blacklisting DeepSeek amid broader China-firm review
DeepSeek-V3 and R1 procurement stays legally ambiguous; build a fallback routing path now.
Reuters
llama.cpp gains model management and download via API
local serving stacks can automate model lifecycle without external tooling; update deployment scripts.
ChatGPT image generator bypassed to produce violent content
red-team your multimodal endpoints; assume prompt-injection vectors exist in any exposed image pipeline.
Web
Tesco migrating 40k workloads off VMware
factor Broadcom exit costs into any infrastructure model still priced on legacy VMware licensing.
Web
GOP bill proposes federal oversight of data center grid access
GPU cluster site selection now carries federal regulatory risk; flag for infrastructure planning.
Web
The Take
Open-weights models are closing the gap on frontier APIs faster than most teams have updated their build-vs-buy assumptions. Meanwhile the infrastructure layer underneath those deployments is getting less stable — AMD's silent security regression, Broadcom's licensing pressure, and incoming federal grid oversight mean the infrastructure audit backlog is growing faster than the model capability backlog.
Subscribe
Related Signals