Week ending 31 October 2026 · Weekly briefing
The week's 5 most useful developments in AI, technology and Australian property. Each one answers the same questions: what changed, who it affects, what failed or is still missing, and what to check before you act on it.
Labels inside every story: reported by the provider, tested in our code lab, or not yet tested. Reply and tell us the AI decision you are weighing; reader questions set next week's worked examples.
Learned density model lowers DRO cost under distribution shiftA preprint centers distributionally robust optimization on a learned density model, lowering cost under distribution shift when data is scarce. Read the full story → |
Mesh-LLM lets you pool GPUs across machines for local AIMesh-LLM, a Rust project gaining stars on GitHub, lets machines share GPUs and memory to run large language models through one local API. Read the full story → |
AI safety fails in multi-turn chats, ShallowDefend flips the weaknessShallowDefend, an arXiv preprint, claims safety alignment is shallow and proposes an explicit decision step before every model response. Read the full story → |
OpenAI shares court briefs to argue fair use covers AI trainingOpenAI published excerpts from court briefs arguing fair use doctrine supports AI training, though no court has ruled on the question. Read the full story → |
NVIDIA Halloween demo runs full voice AI on one workstationNVIDIA's MonsterHouse runs a complete voice agent pipeline from speech recognition to facial animation on one workstation with two GPUs and no cloud. Read the full story → |
You're reading the weekly edition of Not A Tech Guy, an independent briefing on AI, technology and Australian property for people who have to make decisions about them. Found it useful? Forward it to one person who is weighing the same decision. If someone forwarded this, subscribe free → (weekly or daily, your choice).