iCreatAI News
Back to main site
Skip to content
Only Tutorials
Nov 11Wed
  1. AWS Machine Learning Blog40

    Amazon Bedrock tokenomics data lake

    AWS describes a centralized observability data lake that pulls Amazon Bedrock model invocation logs from linked accounts into a central account via CloudWatch subscription filters, processes them through Lambda and Amazon Data Firehose into Parquet in S3, and exposes them through Athena as a single queryable table. Per invocation, the flattened table captures the prompt, response, input and output token counts, model, stop reason, account, timestamp, and IAM role, and can be joined to CUR 2.0 data on the IAM ARN for per-invocation cost attribution. The solution is deployed with two CloudFormation templates from the AWS Sample Bedrock Invocation Data Lake GitHub repository, with an optional Amazon Quick Sight dashboard.

  2. AWS Machine Learning Blog36

    How TCS built an agentic test orchestrator with Amazon Bedrock AgentCore

    TCS built a multi-tenant Agentic Test Orchestrator on AWS that turns Jira stories, BRDs, Swagger specs, and GitLab sources into test scenarios, Gherkin files, and executable Java or Python tests, using specialized agents hosted on Amazon Bedrock AgentCore runtime with the Strands Agents SDK. The platform separates a shared control plane from per-tenant application planes with isolated Cognito user pools, KMS keys, S3 buckets, and DynamoDB tables, and routes API tests to Amazon ECS on Fargate and Selenium UI tests to an EC2-backed ECS cluster. TCS reports that in an internal benchmark of about 100 scenarios, average effort fell from roughly two hours to about 14 minutes per scenario, an 88 percent reduction.

  3. Hacker News AI36

    Testing AI image watermarks: C2PA and SynthID

    The author tested C2PA and SynthID on AI-generated images, finding that C2PA metadata is easily stripped with ImageMagick while SynthID survives metadata removal, cropping, and JPEG compression. SynthID was not detected in a simple rectangle image, and analysis of a plain white AI image showed lower spectral entropy and near-zero spectrum correlation versus histogram-matched random controls. Attempts to suppress high-energy FFT coefficients did not isolate the watermark, though the plain white image showed clear spectral structure.

  4. Hacker News AI58

    What a read-only Postgres role does and doesn't stop when an AI agent connects

    A hands-on PostgreSQL 17.11 test shows that a carefully built read-only agent role blocks writes, DDL, COPY TO PROGRAM and ungranted tables, but its read-only and timeout settings are only per-session defaults that the agent can turn off with SET or BEGIN READ WRITE, and it can even alter its own stored defaults via ALTER ROLE. The piece also shows that GRANT SELECT exposes every column, that PUBLIC can execute SECURITY DEFINER functions like refund_order, and that pg_catalog reveals schema, roles and function source, so the real boundary is the GRANTs plus external enforcement. It closes with a nine-step checklist covering column-level grants, row-level security, revoking PUBLIC EXECUTE and default privileges, and giving the agent only its own credential.

  5. Wiz Research58

    How to Evaluate AI Agent Cybersecurity Capabilities

    Wiz Research outlines the principles behind the relaunch of its Cyber Model Arena, a benchmark of 26 models, four general-purpose harnesses, 59 agents and 327 challenges across code vulnerabilities, API, web and cloud security, arguing that benchmarks should fix their objective first, use real-world tasks, grade deterministically with code rather than LLM judges, isolate agents to prevent cheating, and weigh success against cost. The post reports that agents perform very differently by category (average pass@1 of 78.0% on API security versus 38.2% on code vulnerabilities) and that pairing harness with model matters, with a mean absolute gap of 5.1 pass@1 points across 22 models run on both Claude Code and ADK React. It also lists limitations including simple harnesses, string-equality grading on code tasks, public web tasks that may be contaminated, a 1,000-second time cap and unpublished confidence intervals.

  6. OpenRouter Announcements60

    How to Get Structured Output from Any LLM: JSON Schema, Tool Calls, or a Decision Model

    The article compares three routes to reliable structured LLM output: a strict JSON Schema in response_format, a forced tool call whose parameters are the schema, and a decision model for label-only fields. In tests on 24 support messages across nine schema-supporting models plus GLM 5.3 FlashX, strict schemas returned 209 of 216 valid replies (all 7 misses from one provider that ignores the schema), while prompt-only JSON failed 18 times due to code fences. It recommends require_parameters: true, Response Healing for broken JSON, and always validating replies yourself.

    Why it matters: A practical, test-backed guide to choosing between schema, tool-call, and decision-model approaches for structured LLM output.

  7. OpenRouter Announcements58

    How to Build a Voice AI Agent

    This tutorial builds a Python restaurant-booking voice agent that chains speech-to-text, an LLM with tool calling, and text-to-speech through a single OpenRouter API key, streaming each completed sentence to TTS instead of waiting for the full reply. Across 12 turns in three test conversations, median server-side time to first audio was 1.9 seconds (1.4 seconds without tools, 2.4 seconds with a tool call), and a four-turn conversation cost about $0.013 per minute of processed speech at October 8 promotional prices, with TTS accounting for roughly 90% of model cost. It also compares STT, LLM and TTS models on accuracy, latency and price, and shows how to run the same agent with LiveKit Agents for live audio and interruptions.

  8. OpenRouter Announcements64

    Best OCR Model for Scanned Documents, Measured

    The author tested 15 vision models plus Mistral OCR and Tesseract on eight synthetic scanned pages scored by character error rate and 264 key-value checks, finding Claude Opus 5.5 most accurate (262 of 264 at $16.42 per 1,000 pages) and Mistral OCR the best value (258 of 264 at $2.02). The post documents failure modes such as fabricated card digits, wifi passwords and server names on a faded receipt, and provides code for transcription, disabling reasoning, cross-checking two models from different labs, and using the mistral-ocr engine.

    Why it matters: It offers a reproducible benchmark with per-model costs and practical code for catching hallucinated values in OCR pipelines.

  9. Meta Engineering62

    How We Built TLX: A Hardware-Native GPU Compiler for Production ML at Scale

    Meta introduces Triton Low-Level Extensions (TLX), an extension to Triton that adds explicit warp-group and cluster-level orchestration so engineers can use modern GPU features like TMA, TMEM, and warp specialization while keeping Triton's programming model. TLX is in production across Meta workloads, with reported gains including 20-25% end-to-end MFU for GEM model training on thousands of NVIDIA Blackwell GPUs, up to two-thirds net-latency reduction in RecSys inference, and a distributed grouped GEMM kernel running in large-scale LLM pretraining. The post frames TLX as implementing a "MIMW" abstraction between CUDA's SIMT model and Triton's block-level model, and covers five extension areas: warp-group orchestration, synchronization, memory orchestration, cluster execution, and optional layouts.

    Why it matters: A detailed first-party account of a production GPU compiler, useful for engineers weighing Triton, CUTLASS, and CUDA for hardware-specific kernel work.

  10. Hacker News AI36

    Build a Chess.com AI Companion for Your Latest Game

    This tutorial shows how to build an "After the Game" companion in Quickchat AI that fetches a public Chess.com username's latest completed standard-chess game via two sequential GET Actions on the Chess.com Published-Data API (archive index, then monthly games) and answers questions about its moves. It covers writing the Main Prompt, configuring both Actions with headers and parameters, verifying the returned game link and Action logs, and running a small test dataset, while noting that no chess engine is connected so evaluations and best moves cannot be supplied.

  11. NVIDIA Technical Blog28

    Running AV Reasoning Models Within Onboard Compute and Latency Limits

    AV developers need reasoning models that fit onboard compute and latency constraints; larger models provide a foundation, but deploying them in a vehicle requires reducing model size, preserving useful reasoning, and validating performance on target hardware. The source mentions NVIDIA Alpamayo 2 Edge in this context, but the supplied material is truncated before detailing it.

  12. Hacker News AI40

    Can you prompt your AI into telling the truth? I tested it

    Software developer Scott DeSapio recounts conversations with Claude Opus 4.6 in which he pushed the model to commit to direct honesty, wrote a workspace rule against hedging, and then had it name Objectivism as the source of the rule's principles, with the agent admitting a trained bias against the philosophy. He then opened a fresh chat with both rules in place and asked the model about Objectivism cold, presenting the results as a test of whether prompting can override default agreeableness.

Nov 10Tue
  1. Hugging Face Blog58

    Decision Models in Hugging Face

    Hugging Face explains decision models, which take a state plus typed questions (choice, score, and noul, a Bernoulli yes/no primitive) and return structured decisions with probability distributions instead of generating text. The post contrasts them with traditional and zero-shot classification, and describes ecosystem support: Transformers' DecisionPipeline, AutoModelForDecision and DecisionConfig, a /v1/systemone endpoint also served by llama.cpp, a Hub tag for discovery, and fine-tuning via the Trainer API.

  2. Baseten Blog58

    How to make flash models even faster: optimizing DeepSeek V4.1 Flash 1017

    Baseten describes using an internal harness to optimize 80+ kernels and bottlenecks for DeepSeek V4.1 Flash 1017, producing a Fast variant it reports at 991 output tokens/s on Artificial Analysis with the lowest time per task. The post details three hardware-level techniques: kernel fusion (including an improved fused router kernel cutting decode time by 2.7ms and a 30% faster mHC megakernel), Programmatic Dependent Launch to overlap kernel execution, and host loop minimization via breakable CUDA graphs in SGLang.

  3. Hacker News AI34

    A Read-Only AI Reviewer for Forgejo Pull Requests

    The author built a self-hosted Forgejo pull request review bot in roughly 1,700 lines of Python that fetches PR diffs, sends them to a model via OpenRouter, and posts a summary comment plus inline findings as COMMENT-only reviews it cannot use to approve or block merges. A second verification pass downloads the PR head as a tar.gz snapshot and gives the model read-only file, grep, and directory tools to confirm, correct, refute, or flag findings as uncertain before posting. On a 44-file PR, 18 of 29 findings were discarded after checking the code, and the bot runs as a Read collaborator on an internal network with no published ports.

  4. Simon Willison40

    Quoting Prompting Claude Fable 5.2

    Anthropic documentation notes that Claude Fable 5.2 responds closely to what a prompt says about who owns the code it works on. In Anthropic's testing on small coding tasks, the model tended to add to existing structure when a prompt said nothing about ownership or presented the codebase as someone else's, even when that repeated existing code; telling the model the codebase is its own led it to merge repeated code in about half of tries, versus under a tenth without that sentence.

  5. Hacker News AI25

    I don't want AI in my online search. What options do I have?

    Google's new AI policy for schools led to headlines claiming Google search can no longer be used without AI, prompting the article to explain that all search engines rely on machine learning, so no search is truly "without AI." It distinguishes this from generative AI features like AI Overviews, which have no official off switch on Google but can be avoided by appending "&udm=14" to the search URL or by using alternatives such as DuckDuckGo, Brave Search or Ecosia that disable AI summaries by default.

  6. Hacker News AI34

    Show HN: ZTHR – what replaces code review when AI agents write all the code

    ZTHR is a pattern language for software built by AI, based on two principles: Symmetric Zero Trust, where no author, reviewer, model, operator or report gains authority by asserting success, and the Fixture Inversion, where quality comes from the environment rather than the author's skill. The post links a reference app built entirely by agents under ZTHR, including live app and admin demos, an evidence factory of gates and tests, full source on GitHub, and a discussion forum. It is published as Version 1.0 via Zenodo (doi.org/10.5281/zenodo.23826734).

  7. AWS Machine Learning Blog42

    How skills and agents work in AWS Quick

    AWS Quick uses configurable AI agents that follow instructions, load reusable skills, act through connectors, and ground answers in spaces, knowledge bases, datasets, and dashboards. Skills follow the Agent Skills specification as folders with a SKILL.md file, and can be built from chat, published, and shared with teams. The post also covers best practices for production skills and the AWS Quick official catalog, including the Skill Architect skill.

  8. Simon Willison36

    Locating the source of a mysterious beep with Claude

    Simon Willison describes using Claude Code to build "Beep Locator," a Python web app that connects multiple devices as microphone listeners over WebSockets to triangulate an intermittent beep in a box room, drawing on the 2007 BeepBeep acoustic ranging paper. The system located the beep to within about 40cm, leading him to an old carbon monoxide detector needing new batteries. He highlights the approach of giving Claude Code "ears" by having it run code that analyzes microphone audio and hot-swap new JavaScript onto listeners.

  9. ElevenLabs Blog12

    What is an API endpoint? A guide to connected software

    An API endpoint is the specific URL where an API receives requests and returns responses, with each route pairing an HTTP method with a resource path. The guide covers how endpoints process requests, REST methods like GET and POST, seven common API architectures, design best practices, testing and monitoring, and security through authentication, scoped permissions and concurrency-based rate limits, illustrated with ElevenAPI examples.

  10. Browser Use Blog46

    Autofill Passwords with 1Password and Browser Use

    This tutorial explains how to connect 1Password to Browser Use Cloud via a read-only service account so an agent can fill passwords and TOTP codes, using a dedicated vault and a public test site (The Internet) to verify a login. It covers creating the vault and service account, connecting the token in Browser Use, running the login task through the dashboard or the V4 REST API, and troubleshooting common failures.

  11. Hacker News AI50

    Indie Game Development with AI

    The author built two complete 3D games, Inkline Town and Spirit Call, entirely through AI agents without writing or reading any code, using only chat prompts, playtest links, and 3D model previews. He argues that AI moves the hard part of game development from implementation to the choices only a human can make—what game to build, what style, what is fun, and what is broken—and that agents still fail at anything requiring taste, such as detailed anime characters, which forced him to change Spirit Call's art style. He recommends picking a genre around the AI's bottleneck, building verification layers (ECS structure, types, lints, headless tests, recorded playtests, and math models of the economy), and playtesting as a player while reporting what feels off rather than how to fix it.

  12. AWS Machine Learning Blog38

    How CrossnoKaye grounds physical AI with Amazon Bedrock AgentCore

    CrossnoKaye's ATLAS software-defined automation platform and its AURA agentic system run on AWS, using Amazon Bedrock AgentCore code interpreter microVMs to execute model-authored Python for verifiable statistics and physical balances across plant data. AURA operates in three modes—interactive Chats with warm sandbox sessions, repeatable Tasks, and automated Routines—and enforces layered safety through Bedrock Guardrails, read-only air-gapped sandboxes, deterministic edge base control, and independent hardware protection. The post describes the architecture and reports results such as investigations completing in under four minutes and a 34 percent reduction in batch drying-time standard deviation.

  13. AWS Machine Learning Blog47

    Pluggable Agentic AI Framework: A seven-layer composable enterprise agentic AI reference architecture on Amazon Bedrock AgentCore

    AWS presents a seven-layer composable reference architecture for multi-agent AI applications—Data and Knowledge, Model and Inference, Orchestration, Security and Guardrails, Application, Observability, and Agentic Operations—each with a distinct responsibility and stable contract, illustrated by a retail customer support reference implementation using a supervisor agent delegating to Order and Refund specialists. The post argues that pluggability at layer boundaries lets organizations swap models, tools, and providers without redesigning the stack, and that security, governance, observability, and continuous evaluation should be designed in from day one rather than added after the system is built.

  14. Meta Engineering38

    Extending LLDB to Debug Applications That Span CPUs and GPUs

    Meta describes a vendor-neutral accelerator plugin architecture in lldb-server that separates shared debugger infrastructure from hardware-specific device control, representing the CPU and GPU as separate LLDB targets within one debugging session. The plugin translates a vendor's device debugging API into LLDB concepts such as processes, threads, registers, memory, modules, and stop reasons, and is activated only when the application uses a supported GPU runtime. Most of the common framework has been upstreamed to LLVM, including the plugin interface, GDB-remote packets, and hardware-independent mock tests.

Nov 9Mon
  3. Lovable Blog38

    How we moved our frontend off Firestore and built Lovable Realtime

    Lovable built its own WebSocket-based realtime service, Lovable Realtime, to replace browser-side Firestore listeners, letting its Go backend control payload shape and re-evaluate authorization continuously across Firestore and Spanner data. The service handles peaks of over 2 million concurrent subscriptions and delivered about 2.61 billion data and patch frames in a month, with update latency of 66 ms at p50 and 249 ms at p99. The rollout used shadow traffic and a gradual feature-flag ramp to 100% of users without a user-facing incident.

  4. Hacker News AI38

    My AI agent archived 53 of its own skills during a maintenance run

    Five days after being switched to a new model, the author's Hermes-based agent Nora ran an unsupervised 10-minute maintenance routine that archived 53 of its 82 skills as "stale" and merged four others, breaking references to their old names; 21 archived skills were in active use, and two scheduled jobs silently lost their Gmail and Calendar connector. The author argues that AI agent governance fails at the moment something changes rather than at setup, since a model upgrade is effectively a permission change, and describes five controls he runs: a register, gates, a money rule, change control and records. Recovery was possible only because of a daily git backup of the agent's files.

  5. Simon Willison34

    Hide your slop PR descriptions behind <details><summary>

    Simon Willison describes using HTML <details><summary> tags, which GitHub Markdown supports, to collapse long agent-written PR descriptions behind a short human-curated opener that explains what the PR is for. He uses the model name and effort level as the summary text, and shares recent examples from his datasette and llm projects.

  7. Simon Willison18

    parse_number() as a SQLite extension: C vs Rust

    Four SQLite extensions implement the same parse_number() behavior in C and Rust, with tests covering type preservation, parsing rules, and floating-point accuracy. The optimized C implementation was fastest at 16–19 ns per row, about twice as fast as the Rust versions and far faster than SQL replace(). Plain C was slower on decimals because it relied on strtod(), while a compact fast path closed that gap; all versions matched across 8 million fuzz checks.

  8. Latent Space34

    How Stanford Teaches AI-Native Engineering — Mihail Eric, Monaco

    In this AI Engineer interview, Mihail Eric, who teaches Stanford's CS146S "The Modern Software Developer" and leads AI at Monaco, describes how the course was refreshed for a second run, with part of the homework now being contributions to open-source projects reviewed by the repos' maintainers. He also discusses Monaco's approach of buying some tools (such as Devin for code review) while building its own agent workspaces, guided by his rule of thumb: "You buy to learn, and you build to keep."

  9. Hacker News AI40

    How I Fight AI Brain Fry

    The author describes managing many parallel Claude Code and Codex terminal sessions by organizing work as a git-backed priority queue of markdown tickets, with an agent that researches, builds human-approved tickets, runs tests and browser checks, and never deploys to production. Items needing human judgment land in an "owner inbox" surfaced by a script, and a Claude artifact page rebuilt from git each cycle serves as the reading surface. The key practice is taking one item, loading its context, deciding, and closing it back into the queue so context doesn't linger in memory.

Nov 8Sun
  1. Hacker News AI56

    We review 4x more AI code without reading every line

    Aviator reports that shifting code review from reading diffs to verifying intent, acceptance criteria, invariants, decisions and evidence let the same nine engineers merge 4x more pull requests, cut median time to merge from 2.1 days to 3.4 hours and reduce reverts by 40%, with a human still signing off on every PR. The post describes failed approaches (an AI reviewer, a four-hour review SLA, a self-review skill), mining 11,400 review comments into 64 approved invariants, and using preview environments to prove criteria with recorded evidence.

  2. n8n Blog20

    Send contracts for e-signature from n8n, with TurboDocx, Slack approval and SharePoint filing

    This guest post by TurboDocx CEO Nicolas Fry walks through an n8n workflow that reads contracts marked "Ready to send" from an Excel table, generates the agreement from a Word template via the TurboDocx node, and routes it to Slack for Approve & send / Reject approval. Approved contracts are sent for e-signature with signature and date fields placed at template anchors, and a TurboDocx Trigger then downloads the executed PDF, uploads it to SharePoint, marks the row Signed and posts to Slack. The post covers setup, template preparation, end-to-end testing, troubleshooting and possible swaps such as CRM triggers or Teams approvals.

  3. MarkTechPost38

    Which Routing Does Each MoE Model Use? DeepSeek, Qwen, Mixtral, Llama 4, gpt-oss Compared

    This comparison of 10 open Mixture-of-Experts models shows routing details varying widely: expert counts range from 8 to 384, top-k from 1 to 8, and 7 of 10 models keep one shared expert. DeepSeek-V3 uses sigmoid scoring with a bias-based noaux_tc balancing method, while DeepSeek-V4-Pro raises routed experts to 384, lowers top-k to 6, switches to sqrtsoftplus, and uses hash routing in its first 3 MoE layers. Llama 4 Maverick and gpt-oss do not publish their gating or balancing methods in config.json.

  4. MarkTechPost36

    MoE Routing Strategies Explained: Top-K, Expert-Choice, Sigmoid Gating and Dynamic Routing (2026 Guide)

    This guide explains seven Mixture-of-Experts routing strategies used in production, from softmax top-k token choice (Mixtral, gpt-oss, Qwen3) and top-1 switch routing (Switch Transformer, Llama 4 Maverick) to shared-plus-routed experts, sigmoid gating with aux-loss-free bias balancing (DeepSeek-V3/V4, GLM-4.5, Kimi K2), node-limited routing, expert choice routing and hash/dynamic top-k. It notes that expert choice routing balances load perfectly but leaks future tokens in causal models, with one test showing a roughly 10% abnormal loss drop, while loss-free balancing cut MaxVio from 0.72 to 0.04 at 1B. It advises teams to read four config.json fields (expert count, experts per token, shared experts, topk_method) before serving, since most inherit a router with their checkpoint.

  5. Hacker News AI38

    Ask HN What to learn for AI Researcher interview?

    A post describes an interview-preparation curriculum for junior AI researchers at large labs, organized into eleven modules covering PyTorch fundamentals, theoretical foundations, implementation from scratch, ML debugging, system design, inference optimization, RAG, paper analysis, safety and alignment, CUDA, and mixture of experts. Each module pairs an article with a hands-on exercise notebook and reference solutions, with scaled dot-product attention revisited at every level, and most material runs on CPU except the CUDA module. It also suggests time-boxed study budgets and module orderings for coding, systems, and research interview rounds.