Loading…
Browse by type
Explainer reads
Explainer posts on Yomu, summarised newest first.
- 1,318 posts
- 20 companies
The Good Stuff
Paid picks we use ourselves. Links carry a referral code — a subscription through one earns us a commission at no extra cost to you.
Explainer posts
From one switch to a control panel: meet `dataCollection`
Sentry is replacing the JavaScript SDK boolean option sendDefaultPii with dataCollection, a category-level control for automatically collected data. In version 11, dataCollection becomes the default, sendDefaultPii is removed, and several categories—including user information, cookies, HTTP bodies, GenAI inputs and outputs, database query data, and stack-frame variables—are collected by default. The post distinguishes PII, which is collected by default, from sensitive credentials, whose values in collected headers are filtered automatically; manually attached data remains unaffected. Developers can preserve v10’s restrictive baseline or select true, false, allow lists, and deny lists for supported fields, while beforeSend and beforeSendSpan handle additional filtering. Because v11 also enables span streaming by default, span drops move to ignoreSpans, and existing withStreamedSpan wrappers should be removed.
Fast, fault-tolerant PyTorch training on AI Runtime
At scale, training jobs face frequent GPU interruptions, while remote-storage input pipelines can leave accelerators idle. Distributed Checkpoint (DCP) has each rank write a shard plus a .metadata file, allowing parallel saves and reloads onto a different number of GPUs; DCP is presented as useful for DDP as well as sharded training. Asynchronous saves stage data locally and upload in the background; cited measurements report 1.8x savings for a 2.8B-parameter DDP model and 58x for a 20B-parameter FSPD model on 32 H100s, though the comparison excludes torch.save network storage time. UCVolumeDataset and the accompanying DataLoader cache files on local NVMe and prefetch disjoint rank/worker slices while computation runs. Correct recovery also requires checkpointing data position plus seeds and RNG state so resumed sampling remains consistent.
Why 78% of customers try to bypass your AI agent
Twilio research cited in the post finds that 78% of consumers have tried to bypass an AI agent to reach a human, while 63% want easy escalation and only 47% of brands provide it. The reported causes are unclear AI disclosure, discomfort with high-stakes tasks, and prior failures: 90% of tested consumers misidentify text-based AI, 80% report a negative AI experience, and failed attempts add 81 seconds before escalation. Consumers are most comfortable with scheduling and reservations, returns and exchanges, and concert seat selection, but pull back on medical inquiries, payments, and loan applications. The proposed operating model combines routine automation with clear disclosure, obvious escalation, persistent customer memory, and real-time detection of sentiment and escalation risk. It argues that trust, rather than wholesale replacement, should determine how humans and AI divide customer-service work.
AI didn't write our SDK. It changed how we built it.
Stripe's extensibility SDK was built with AI involved across prototyping, architecture, implementation, and refinement, but the main change was to the development loop rather than code-generation speed. For a custom-object design, one engineer produced a working prototype, demo, and documentation site in roughly two days instead of an estimated two to three weeks, allowing the team to evaluate boundaries and reprioritize the feature for the next major launch. The team then used reviewed, detailed specifications and persistent domain guidance to delegate bounded work, including about twenty ESLint rules covering authoring constraints such as sandbox boundaries, imports, and type safety. AI also helped keep scripts, workflows, SDK generation, runtime enforcement, tooling, documentation, examples, and tests aligned, while human judgment remained responsible for abstractions and trade-offs. The post concludes that AI's leverage came from making more artifacts concrete earlier, enabling earlier feedback and a tighter deliverable.
LLMs Are Not (Consistently) Bayesian: Quantifying Internal (In)consistencies of LLMs’ Probabilistic Beliefs
LLMs are deployed in medicine, science, and law, where observed evidence may not determine a single correct answer and systems must represent and update uncertain beliefs. The work treats LLMs as information processing rules and measures an information processing gap, defined as deviation from Bayes updates, to quantify internal consistency in probabilistic belief updates. Extensive experiments compare multiple ways of incorporating evidence: some yield nearly Bayesian updates, while others rely on learned heuristics. Although exact Bayesian updating is optimal information processing, the non-Bayesian heuristic updates often achieve better downstream task performance, suggesting that the LLMs’ probabilistic models of the world are misspecified. The measure also serves as a diagnostic for issues in LLM-powered inferential systems.
So, is ClickHouse winning the observability wars?
The article examines whether ClickHouse is “winning” observability by separating the storage-and-query question from the broader work of collection, schema design, correlation, visualization, alerting, and investigation. It argues that ClickHouse fits telemetry because its columnar layout stores fields separately, enabling compression and allowing analytical queries to read only selected columns across millions or billions of events; ordering keys, sparse primary indexes, and data-skipping indexes further reduce work. The discussion covers both rich, wide events and conventional logs, traces, and metrics, and presents ClickStack as a combination of ClickHouse, an OpenTelemetry Collector, and a UI that preserves familiar workflows. The conclusion is qualified: ClickHouse has become the de facto storage and query engine for many teams and a strong default for large telemetry workloads, but it has not won observability as a whole, while full-text indexes, PromQL support, and broader product layers remain important.
One Write Path, Multiple Entries: How Rippling uses Unity Catalog-managed Apache Iceberg™
Rippling describes an architecture in which Databricks writes some tables while Trino, Snowflake, and other lakehouse engines read them, making catalog ownership a design decision rather than a metadata detail. Its rule is that the catalog follows the writer: AWS-native workloads use Glue, Databricks workloads use Unity Catalog, and Snowflake-produced shared data uses Horizon Catalog with Snowflake-managed Iceberg tables. For selected Databricks outputs, Unity Catalog-managed Apache Iceberg stores table data and metadata in Rippling-owned S3, exposes access through the Iceberg REST Catalog API or federation, and uses credential vending for scoped storage access. This lets downstream engines read Parquet directly without duplicate materialization or producer-side query execution. Databricks also provides Predictive Optimization for maintenance, while the trade-off is continued Unity Catalog dependence for metadata, permissions, lineage, and lifecycle.
How scientists are using Claude to accelerate research and discovery
Claude is being used by scientific research groups as a collaborator across experiment planning, data analysis, and discovery, rather than only for literature reviews or coding. The post describes Biomni, a Stanford platform that gives a Claude-powered agent access to hundreds of biological tools and databases; in an early trial, it completed a genome-wide association study analysis in 20 minutes, while reported validation cases included wearable data and 336,000-cell gene-activity data. Other systems encode expert workflows, including MozzareLLM for interpreting gene-knockout clusters, while a Lundberg lab project uses molecular relationship maps to propose genes for under-studied biological processes. These systems remain imperfect and require guardrails, expert-defined skills, or prospective testing, but the reported examples suggest AI can reduce research bottlenecks and support new approaches to scientific investigation.
What QSR reports miss about the decisions matter the most
Traditional QSR reporting can show that a limited-time offer (LTO) missed systemwide expectations, but not whether the cause was weak demand, lower franchise participation, ingredient shortages, inconsistent execution, slower service, or menu cannibalization. The post presents Lovelytics’ QSR Executive Performance Control Tower, an AI blueprint/solution built on Databricks that connects corporate performance, supply chain and operations, customer and digital, and financial signals. It is designed to distinguish demand from operational and participation issues, link changes to likely drivers and financial impact, and give corporate and franchise leaders a daily briefing on what changed, why it matters, and where to focus. The stated advantage is faster, more actionable decision-making based on a governed, reusable intelligence layer rather than disconnected reports.
Object Storage + WAL: Lakebase Postgres for the agentic era
The post presents Lakebase Postgres as a design for agent workloads that need isolated copies, restores, replicas, historical queries, and many short-lived databases without repeatedly moving large datasets. It contrasts conventional data-centric OLTP with a transaction-centric model in which PostgreSQL’s write-ahead log (WAL) is the source of truth, with LSNs making database history addressable. Lakebase splits compute, which runs standard Postgres and uses RAM and local NVMe caches, from storage, where safekeepers replicate WAL through quorum acknowledgements, a pageserver materializes pages, and object storage retains immutable history. Commits wait for replicated WAL acknowledgement, not page uploads, while reads request pages at an LSN. The design supports point-in-time queries, cheap branch or read-only compute, scale-to-zero suspension, and shared transactional and analytical access, while still relying on RAM and NVMe for performance.
Beyond the Cloud: Building an Offline
A digital-first economy assumes constant high-speed internet, leaving mobile wallet apps unusable when connectivity drops, such as on subways or in rural areas. A demo wallet solution combines MongoDB Atlas as the cloud source of truth with ObjectBox as an on-device operational store that mirrors the same schema. The ObjectBox Sync Server continuously bridges the two, enabling local reads and writes that sync automatically when connectivity returns. For offline sends, the wallet checks the requested amount against the last-synced balance to prevent overspending, ensuring safe local operations. The architecture treats connectivity as an enhancement rather than a requirement, keeping the wallet responsive in airplane mode with full data access.
Decoding cosmic signals with deep learning and Keras
Astroparticle physics must reconstruct cosmic particles from indirect measurements because high-energy particles create extensive air showers before reaching distributed detectors. Modern observatories such as Pierre Auger, IceCube, and CTAO record spatially distributed sensor data and nanosecond-resolution waveforms, producing image-like spatiotemporal structures at enormous scale. The post explains how deep learning can improve instrument sensitivity, reveal hidden patterns, search for anomalies, and support event reconstruction across cosmic-ray, neutrino, and gamma-ray observations. It also frames fast prototyping of physics models in Keras as a practical entry point, while noting that more data and efficient scaling can sometimes outperform physics-inspired architectures. The resulting tension is between predictive performance from machine-learned black boxes and the physical insight, causal understanding, and interpretability scientists seek.
Understanding and addressing AI harms
An evolving framework addresses AI harms beyond catastrophic risks, covering concerns such as biological threats, child safety, disinformation, and fraud. It complements the Responsible Scaling Policy by examining physical, psychological, economic, societal, and individual-autonomy impacts, then assessing likelihood, scale, affected populations, duration, causality, technology contribution, and mitigation feasibility. Policies and practices include a Usage Policy, pre- and post-launch evaluations, red teaming, adversarial testing, misuse detection, and enforcement ranging from prompt modifications to account blocking. Examples include computer-use safeguards for banking platforms and communications tools, plus hierarchical summarization intended to detect harms while maintaining privacy standards, alongside model-response evaluations balancing helpfulness with harmlessness. For Claude 3.7 Sonnet, encouraging safe, helpful responses to ambiguous prompts produced a reported 45% reduction in unnecessary refusals while maintaining safeguards against truly harmful content; the broader approach remains under development.
Challenges in red teaming AI systems
AI red teaming is presented as adversarial testing intended to identify vulnerabilities and risks in AI systems, but inconsistent practices make safety comparisons difficult. The post surveys domain-specific expert testing, policy vulnerability testing, frontier-threat assessments, multilingual and multicultural work, language-model-assisted attacks, automated red teaming, multimodal testing, and crowdsourced or community-based approaches. It describes an iterative process in which experts articulate threat models and probe systems qualitatively, then standardize inputs and use language models to generate many variations for more quantitative automated evaluation. In automated red teaming, a red team model generates attacks and a blue team model is fine-tuned on those outputs to improve robustness. The conclusion calls for technical standards, independent testing capacity, professional services, third-party access, and links between red teaming and model release conditions.
Frontier model security
Frontier AI models and their research are presented as strategic assets requiring protections beyond ordinary commercial technology because theft or misuse could affect economic and national security. The post recommends treating advanced AI as akin to critical infrastructure, with public-private cooperation and possible procurement or regulatory requirements. Its central control is multi-party authorization: no individual retains persistent access to production-critical environments, and time-limited access requires a coworker’s approval and business justification. It also argues that NIST’s SSDF and SLSA can be adapted to model development and coupled software, creating a chain of custody that supports provenance for deployed systems. Anthropic says it is implementing these controls and other cybersecurity practices through an iterative process, while acknowledging protections must evolve as model capabilities scale.
How Ubuntu Is Using Rust to Rebuild Core System Tools
Canonical is selectively adopting Rust to rewrite core system utilities across Ubuntu releases to improve memory safety, security, and long-term resilience. Starting in Ubuntu 26.04 LTS, uutils coreutils shipped as the default implementation targeting complete GNU compatibility, alongside sudo-rs, which introduced intentional behavior changes like asterisk display on password entry. Canonical is also funding ntpd-rs to consolidate NTP, NTS, and PTP into a single utility, while collaborating on UPKI to bring certificate revocation to system utilities. To manage Rust packaging at distribution scale, Canonical vendors dependencies per package rather than creating individual Debian packages for crates and embeds software bills of materials into binaries using cargo auditable. Future plans include exploring Rust implementations of compression libraries like bzip2-rs, zlib-rs, and zstd-rs.
How AI observability works with MLflow
Traditional application monitoring treats AI workflows as black boxes, leaving engineering teams unable to determine whether incorrect responses result from model hallucinations or erroneous upstream tool data. MLflow provides AI observability by capturing distributed traces containing model calls, prompts, retrieved context, execution latencies, and tool interactions. In an example mortgage lending application, trace inspection revealed that an AI assistant accurately summarized raw JSON data returned by an internal pipeline tool rather than hallucinating conflicting metrics. Beyond latency attribution and tool debugging, MLflow is compatible with OpenTelemetry and provides automatic tracing integrations for frameworks such as LangChain and Claude Code. Capturing these traces enables teams to evaluate subsequent application versions, detect model provider shifts, and identify tool regressions before reaching users.
How to evaluate session replay software: a developer's guide
Session replay software varies fundamentally based on architecture, distinguishing tools designed for product analytics from those built for developer debugging. Recording approaches split into DOM snapshot and mutation capture, which optimizes for compression and structured debugging, versus pixel capture, which favors visual fidelity for complex animations. Privacy depends heavily on masking location, where client-side masking redacts data before network transmission, whereas server-side scrubbing acts after transmission. For engineering teams, integration depth connects replays natively to traces, console errors, and backend issues rather than requiring timestamp lookups across separate products. Performance costs across CPU, memory, and network require benchmarking on target applications alongside strategies like error-based sampling.
8 top conversational AI platforms in 2026
Published in August 2026, this guide compares eight conversational AI platforms amid rapid market consolidation and acquisitions including NiCE’s roughly $955 million purchase of Cognigy. It defines the category as software for building, deploying, and operating agents across voice and messaging, then distinguishes vendor-managed services that own reasoning from infrastructure providers that carry conversations while teams own the reasoning. The evaluation criteria include model flexibility, voice and messaging parity, deployment control, cross-channel context, human handoff, pricing, and integration depth. Twilio is presented as an infrastructure layer for teams choosing their own models, with ConversationRelay, Agent Connect, Conversation Orchestrator, Conversation Memory, and Flex supporting channels, continuity, and escalation; other entries include Google Conversational Agents, Kore.ai, and NiCE Cognigy. The guide argues that portable models and conversation data will matter as the conversation layer separates from agent reasoning.
Leveraging Databricks to Support FISC Security Guidelines
The FISC Security Guidelines define control objectives for Japanese financial institutions, while the post addresses the challenge of translating them into auditable controls on Databricks data and AI workloads. The Databricks Data + AI Platform, Unity Catalog, and optional Enhanced Security Monitoring and Compliance Security Profile features are presented for identity and network controls, encryption, fine-grained access, audit logs, lineage, and hardened runtimes. For AI workloads, Unity Catalog’s Model Registry links training data, code versions, and evaluation metrics to model versions, while a separate framework maps 97 AI-specific risks to 73 controls. The FISC Customer Capabilities Mapping Matrix connects guideline units to platform capabilities and customer responsibilities, including configuration, SIEM integration, backup, recovery, and evidence; the Security Analysis Tool is suggested for finding workspace gaps, and the material is identified as educational guidance requiring expert review.
Top topics
- Architecture830
- Performance812
- LLMs648
- Machine Learning617
- Open Source561
- AI Agents542
- AI454
- Developer Experience446
Weekly digest
The best engineering reads, once a week.