Agent (Muse) Compute Demand - by FD - Robonomics

kept by eddie sep 28

Estimating Meta’s Muse agent for 100 M DAU, the analysis finds ~0.1 GW CPU/VM power, ~1‑2 GW inference power, requiring ~$800 M CPUs, ~$2 B DRAM, and up to 3‑4 GW total.

A Guide to AI Inference Engineering - ByteByteGo Newsletter
LLM inference splits into compute-bound prefill and memory-bound decode, driving optimization techniques like batching, quantization, speculative decoding, and disaggregation.
also on #ai-inference, #developer-tools
Apple Silicon costs more than OpenRouter
Running local LLMs on Apple Silicon costs about 3x more per million tokens than using OpenRouter, and also runs slower.
also on #ai-inference, #developer-tools
Meta Muse is People
Meta tested a human‑in‑the‑loop concierge for its Muse AI assistant to handle phone calls, raising privacy and trust concerns despite the agent’s real‑world capabilities.
also on Muse, Meta, #ai-agents
Meet Noah Shinn, Whose Instinct AI Competes With Meta's Muse - Business Insider
23-year-old Noah Shinn's invite-only AI agent Instinct raised $250M at a $2.5B valuation and now competes with Meta's personal assistant Muse in the battle for practical AI assistants.
also on Muse, Meta, #ai-agents
Is AI Really Eating the World? - philippdubach.com
AI models are commoditizing, shifting value to applications and integration over model providers, despite massive infrastructure spending.
also on DeepSeek, Microsoft, Meta, #cloud-computing