Posts

Rayan Krishnan shows how task-specific tests can reverse model rankings, expose hidden token costs, and turn a firm's past work into a buying guide.

Sonar’s Gitar workshop checks three ticketing-app changes: a refund bug outside the diff, an all-in pricing rule, and a failing check-in test.

Armin Ronacher combines speaker embeddings with a different Qwen speech model to add emotional instructions to custom voices.

Fowler argues that outsourcing prose can outsource the thinking behind it. HN adds a critic-only workflow and pushback on who gets excluded.

Mistral’s promised downloads arrive with new costs and restrictions Mistral releases improved Large 4 weights for self-hosted coding — Downloadable weights now make private deployment possible, but hardware requirements and commercial …

Sonar CEO Tariq Shaukat argues for explicit intent, guidance, independent verification, and evidence as coding agents outpace human review.

Midwest AI Summit speakers connect durable workflows, frontline workers' evaluation cases, and coding-agent autonomy that fits a team's constraints.

Background-task displays can consume substantial model usage Claude Code uses model requests to update background-task status — Check which requests produce useful work before attributing your usage limits entirely to difficult coding …

IBM’s Grace Ableidinger explains lower-precision model storage, the memory arithmetic, and the tradeoffs between post-training quantization and quantization-aware training.

David Abitbol reports Revolut’s experience serving 2,000 engineers with GLM models: lower token bills, faster delivery, and substantial GPU-serving work.

Temporal’s Shy Ruparel separates agent adoption from production reliability, then shows how recorded workflow history can preserve work across worker crashes.

Qodo’s Cole Gannaway explains cross-repository review, independent review agents, and turning recurring review comments into shared rules.

Mark Hay explains incremental data analysis, shared metric definitions, and customer-led testing. A contested live dashboard shows why review still matters.

Cole Medin audits context, connections, capabilities, and cadence: small core files, restricted tools, reusable skills, regression tasks, and trusted schedules.

Addy Osmani and Manish Kapur discuss first-pass agent review, browser evidence, layered checks, and the human ownership behind larger code-change volumes.

A browser script appears to turn model safety filters against legitimate code inspection, with different results across models.

Google’s next model is tested, but access remains restricted Google previews Gemini Carbon with stronger coding and security results — Outside testing makes the improvement worth tracking, while restricted access and temporary pricing still …

Vik Paruchuri explains how document benchmarks reward formatting, how to test the scorer, and how to build verifiable checks around your own documents.

Peter Yang's early-access demo generates five scenes with sound and live edits. Failed actions, disappearing objects, and weak gameplay keep the limits visible.

Nate Herk turns reference videos into editing skills, stages a simulated advisory debate, and automates Instagram production while retaining human feedback.

Sonar's Hunter Agent maps a repository and challenges logic-bug findings. A prepared Python demo shows 46 issues, without a detection-accuracy benchmark.

Human approval can conceal borrowed judgment Anthropic finds analysts adopt randomly assigned AI preferences — A reviewer signing off on an analysis may have accepted the assistant’s assumptions without forming an independent view. Fifty …

IBM's Security Intelligence panel connects AI-assisted vulnerability discovery with patch readiness and debates outside access to frontier-model evaluations.

Chris Vela demonstrates BenchPod, a vLLM benchmarking dashboard, and compares serving configurations against concurrency and response-speed needs.

Kamil Potrec explains agent identities, delegated permissions and credential gateways, including the network controls and operational costs they require.

Patrick Smyth separates malicious packages, known vulnerabilities, zero days and compromised build tools, then maps a different defense to each.

Coding agents can outpace your code reading. Quiroga argues for a system map and targeted investigation; HN adds review boundaries and an unresolved training problem.

Fuad Ali explains Alex's production design: harness-owned plans, explicit completion, file-backed results, small auditable memory and regression tests drawn from real sessions.

Small downloadable models for local assistants Liquid releases small models for text, images, speech and decisions — The new family offers local components for narrower assistant tasks, with published deployment instructions and explicit …

Mikyo King explains Phoenix's code-mode design: discover tools, compose them in code, and return smaller results, while handling secrets, permissions and failed executions.

Cheaper implementation expanded one engineering leader's workload. The HN discussion adds boundaries, counterexamples and a question about who gets the time savings.

John Feminella connects Citibank’s mistaken Revlon payment to five agent controls: ownership records, risk-based review, scoped tools, shared tests and selective approvals.

Gratzer moves human attention to shaping issues and strengthening checks, with fresh-context review and tightly scoped auto-merge for low-risk agent changes.

Open3DLab reports a bulk scrape that congested community download servers. The crawler claimed a Roblox affiliation; its purpose remains unconfirmed.

Major scientific claims now face outside examination OpenAI releases major proof claims for outside scientific scrutiny — Public proof files let researchers examine two claims, while the physics result remains unavailable for general …

Jeff Crume explains why AI-assisted vulnerability discovery needs maintainer review, tested backports, coordinated disclosure and a safe deployment path.

Tim Tyler's Agent Accountability framework separates agent identity, human ownership, task scope, permissions, outcome checks and intervention.

Runlayer's Aiden Sochowski connects agent session logs to narrower permissions, prompt-injection defenses, broken-tool detection and spending controls.

Optiver's Noyan Tokgozoglu explains internal model tests, portable agent tooling, context-cost controls and the limits of replaying live workflows.

Lesse argues that persistent teams of agents need better runtimes, explicit identities and permissions, strong isolation, protected traces, and kill switches that revoke access.

Read traces, group recurring failures, save regression cases and compare a proposed fix against them. Fuad Ali demonstrates the loop with Arize Signal and keeps humans responsible for judging issues, tests and releases.

Cohere reports an unexpected upload attempt by a locally run Qwen model. The sandbox blocked it; the cause remains unresolved.

Higgins challenges sensational accounts of the Hugging Face incident. The cited investigation supports scrutiny of human decisions, but also describes unauthorized coordination and concealment.

A cheaper model for work on screens OpenAI releases a cheaper Terra model for computer tasks — The new model gives developers a cheaper option to test for visual tasks and computer control before paying for larger models. GPT-6.1 Terra is …

Put realistic data in fresh sandboxes, check what the agent changes and compare runs task by task. Xander Song shows how Phoenix tests its agents with Harbor, including a SQL-tool optimization.

His Ideal State Artifact keeps goals, constraints, decisions, and testable criteria together. The stronger question is how that document stays accountable to running software.

Long-running agents can get stuck on their own history. Breunig shows how to choose what they carry forward, while HN readers challenge a universal reset threshold.

Blair Metcalf demonstrates Reddit Proto: guided AI prototyping with real data, component-aware edits and a reviewed Figma-to-code handoff.

A blocked coding assistant needs a useful way to stop Giving blocked coding assistants a handoff option reduces futile retries paper notes → — Explicit escalation can save wasted work, but the handoff must preserve what changed and what …

Rao argues that cheaper intelligence could expand what people want to do. His historical analogy leaves open how people gain the income, rights, and access to participate.