Multistrategy funds are growing beyond their own payrolls
Separate accounts expand outside capacity while turning integration into an organizational-boundary problem.
Read the note →The topic collection
AI & Engineering.
Models, agents, formal methods and the systems behind them.
Switch topic
104 notes across 45 editions. Dates refer to original publication; prices, events and availability may have changed.
No notes match that search.
Try a different word or .
METR's audit asks whether the monitor receives the work
Eligibility, receipt and review timing are different control measures—and each can fail independently.
Read the note →Google makes Confidential G4 available, with attestation before sensitive data release
Confidential GPU instances narrow the infrastructure trust boundary without certifying the application above it.
Read the note →A cuFile call does not prove a GPU-direct read
A successful storage call can conceal CPU staging; route evidence belongs in benchmark provenance.
Read the note →SageMaker offers managed GPU operations inside a customer-written training loop
AWS puts distributed GPU operations behind APIs while the customer retains loop and reward logic.
Read the note →Meta's cash flow gives the AI spending argument a new number
A quarterly free-cash-flow reversal makes the AI capital cycle more concrete.
Read the note →Outside listening: an opinion conversation about AI finance
Reuters Breakingviews connects the neocloud selloff, credit stress and the wider AI investment cycle.
Read the note →A speech endpoint can stay healthy while every request fails
Binary WebSocket frames, readiness semantics and first-audio latency belong in one production acceptance test.
Read the note →G-Research examines a storage format built for the way models actually read
Lance separates layout from compression to serve shuffled samples, embeddings and reproducible model training.
Read the note →Correct final numbers can conceal the wrong agent process
A process-level evaluation finds hidden workflow deviations in 92.6% of apparently successful runs.
Read the note →The AI buildout has three prices: capital, power and usable revenue
Capital concentration, electricity constraints and falling token prices create distinct hurdles for the AI buildout.
Read the note →Two platform changes worth treating as production work
Longer cache lifetimes, service-tier restrictions and redirect-aware domain controls all need operating tests.
Read the note →The monitor is part of the system
Two OpenAI incidents turn control preservation into a concrete acceptance criterion.
Read the note →Microsoft proposes a control plane for the “agentic insider”
Project Perception and Agent 365 separate adaptive intelligence from deterministic authority.
Read the note →Citadel’s boundary: assistance without surrendering judgment
Ken Griffin describes an AI deployment while keeping investment judgment with people.
Read the note →Walleye builds across internal teams and outside managers
New senior hires sharpen questions about shared tooling, permissions and external-manager structure.
Read the note →Two concrete AI handoffs in investment research
Redding Ridge and Coatue show where task routing, provenance and human refinement become measurable.
Read the note →The lead: Jane Street's AI agents need identities, budgets and colleagues willing to say no
Jane Street's own account of permissions, compute incentives and review capacity.
Read the note →Balyasny moves an AI research leader into fundamental equities
An applied-AI research leader takes a fundamental-equities mandate.
Read the note →Two AI measurements worth testing inside a real firm
Self-assessment cost and COBOL migration parity deserve local tests.
Read the note →Agent attackers leave their own control panels open
An agent intrusion investigation shows how exposed consoles compound the risk.
Read the note →HRT found the traffic the application logs missed
Attributing the work behind a research-network bottleneck.
Read the note →Bridgewater’s Vault: an evaluation that investors could recognize
A direct operator talk on retrieval, filtering and moving research habits.
Read the note →Citadel Securities tests the green status slide
Project history becomes a testable institutional-memory system.
Read the note →Keep in the loop
Good reading. On your terms.
Add LEON to your RSS reader for new notes as they are published.