Full archive
AI archive
Newsletter issues, market reads, model updates, product launches, policy shifts, and startup coverage from Superpower Daily.
LMCache Reworks the Cache Plumbing That Can Stall Long-Running Agents
The changes address a specific failure mode in long, concurrent agent sessions: reusable model state can fill a shared pool before any request makes progress. The reported gains are promising, but they come from targeted validations rather than a broad production benchmark.
Nvidia Dynamo’s Agentic Speedups Target the Work Around the Model
The reported gains come from changing how serving systems route, retain and stream state under concurrency. The next challenge is whether these AgentX-specific improvements hold across production mixes and unfinished scheduling work.
Anthropic Told Investors Its Revenue Run Rate Reached $65 Billion
The confidential investor update suggests a rapid expansion ahead of a potential public listing, though a run rate is an extrapolation rather than booked full-year revenue.
Cerebras Turns Up the Power on Its CS-4 Inference Rack
The new system aims to double CS-3 performance through faster clocks, denser racks and more cooling—not a new chip. Its 44 GB of SRAM per wafer remains a constraint as Cerebras pursues partnerships for split inference workloads.
TestMu’s New Agent Test Shows What It Can’t Verify
Agent Assurance is built around observable effects rather than an agent’s own account of its work. Its central unresolved test is whether that evidence-based scorecard predicts safer behavior in real systems.
AgentX Replays Claude Code Sessions to Test AI Serving Systems
The replay dataset is designed to measure infrastructure behind agentic coding workloads, not whether a model writes better code. Its value depends on whether synthetic traces preserve the traffic patterns operators need to serve.
Wispr’s $280 Million Bet Is That Voice Can Move Beyond Dictation
The company is putting new capital behind meeting notes, hardware partnerships and a speech model whose promised gains must hold up in noisy, real-world use.
Amazon Bought Rare Books. A Tracker Led to Its Las Vegas Scanning Site.
Amazon acknowledges buying books through commercial channels to improve its products and services. A tracked shipment connects one rare-book order to a site where employee posts describe cutting and scanning volumes, leaving the scale and purpose of the operation unresolved.
3M Expert Asked ChatGPT for a Zero-Fault Defense. Jurors Assigned 3M 30% Blame.
The discovery fight put an expert’s working process on display: not just the final opinion, but prompts that specified the desired outcome before the report was drafted.