<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0"><channel>
<title>Abdul Habeeb's blog</title><link>https://abdulhabeeb.com/blog/</link><description>What is happening in AI and data, checked at the source.</description><language>en</language>
<item><title>The FAA&#x27;s AI routing tool goes nationwide, and the controllers&#x27; union wasn&#x27;t involved in building it</title><link>https://abdulhabeeb.com/blog/faas-ai-routing-tool-goes-nationwide/</link><guid>https://abdulhabeeb.com/blog/faas-ai-routing-tool-goes-nationwide/</guid><pubDate>Sun, 15 Nov 2026 00:00:00 +0000</pubDate><description>The FAA is extending its SMART system to all 22 of its centers. The agency reports fuel and time savings from its first rollout in Washington, but the union representing controllers says it had no role in designing or testing the system.</description></item>
<item><title>llama.cpp reaches 1.0, and the main change is a promise about compatibility</title><link>https://abdulhabeeb.com/blog/llamacpp-reaches-10-main-change-promise/</link><guid>https://abdulhabeeb.com/blog/llamacpp-reaches-10-main-change-promise/</guid><pubDate>Sat, 14 Nov 2026 00:00:00 +0000</pubDate><description>Hugging Face&#x27;s llama.cpp and ggml projects have released v1.0.0 libraries. The announcement is mostly about how breaking changes will be signalled, and it lists no new features.</description></item>
<item><title>Fireworks AI breach reached Notion AI chat transcripts</title><link>https://abdulhabeeb.com/blog/fireworks-ai-breach-reached-notion-ai/</link><guid>https://abdulhabeeb.com/blog/fireworks-ai-breach-reached-notion-ai/</guid><pubDate>Fri, 13 Nov 2026 00:00:00 +0000</pubDate><description>An attacker got into a Fireworks AI development system through its cloud provider. Notion says some of its customers&#x27; AI chat transcripts were among the data taken.</description></item>
<item><title>Grok Build&#x27;s share links exposed business data, including SpaceXSI&#x27;s own</title><link>https://abdulhabeeb.com/blog/grok-builds-share-links-exposed-business/</link><guid>https://abdulhabeeb.com/blog/grok-builds-share-links-exposed-business/</guid><pubDate>Thu, 12 Nov 2026 00:00:00 +0000</pubDate><description>Claroc says a design choice in Grok Build&#x27;s sharing system let anyone list live projects and open them. Some of the exposed material belonged to SpaceXSI, the company that runs the platform.</description></item>
<item><title>Google&#x27;s Gemini 4 Carbon opens to the public, priced under its rivals for now</title><link>https://abdulhabeeb.com/blog/googles-gemini-4-carbon-opens-public/</link><guid>https://abdulhabeeb.com/blog/googles-gemini-4-carbon-opens-public/</guid><pubDate>Wed, 11 Nov 2026 00:00:00 +0000</pubDate><description>Google is rolling out Carbon to paying developers and Ultra subscribers, and previewing a multi-agent &quot;swarm&quot; feature. Its per-token prices are low, but the introductory rate ends in November.</description></item>
<item><title>DeepMind&#x27;s AI checked parts of Signal&#x27;s protocol with math and found no new flaws</title><link>https://abdulhabeeb.com/blog/deepminds-ai-checked-parts-signals-protocol/</link><guid>https://abdulhabeeb.com/blog/deepminds-ai-checked-parts-signals-protocol/</guid><pubDate>Wed, 11 Nov 2026 00:00:00 +0000</pubDate><description>Google DeepMind used agentic AI to formally verify several cryptographic guarantees in the Signal protocol. It was a four-week proof of concept, and it covered only part of the system.</description></item>
<item><title>Gemini Robotics 2.5 hit 96% in European factories, and DeepMind admits the last 4% is hard</title><link>https://abdulhabeeb.com/blog/gemini-robotics-25-hit-96-european/</link><guid>https://abdulhabeeb.com/blog/gemini-robotics-25-hit-96-european/</guid><pubDate>Wed, 11 Nov 2026 00:00:00 +0000</pubDate><description>Google DeepMind&#x27;s new robotics model improved on lab tests and reached 96% success in car-parts kitting. DeepMind&#x27;s own director says the remaining gap is a steep road.</description></item>
<item><title>Researchers backdoored an open model for under $50</title><link>https://abdulhabeeb.com/blog/abliterated-model-backdoors/</link><guid>https://abdulhabeeb.com/blog/abliterated-model-backdoors/</guid><pubDate>Wed, 04 Nov 2026 00:00:00 +0000</pubDate><description>ProjectDiscovery showed that edited open weights can hide a trigger that steals credentials, while clean behaviour and benchmark scores stay unchanged.</description></item>
<item><title>One bug in a benchmark&#x27;s scoring moved agent success from 94.3% to 65.5%</title><link>https://abdulhabeeb.com/blog/agent-benchmark-oracle-failure/</link><guid>https://abdulhabeeb.com/blog/agent-benchmark-oracle-failure/</guid><pubDate>Wed, 04 Nov 2026 00:00:00 +0000</pubDate><description>A new preprint audits an agent benchmark and finds a reward function that changed the state it was grading. Fixing it changed which agent came out on top.</description></item>
<item><title>Was the AI agent getting better, or did the reviewers change?</title><link>https://abdulhabeeb.com/blog/agent-reliability-rater-effects/</link><guid>https://abdulhabeeb.com/blog/agent-reliability-rater-effects/</guid><pubDate>Wed, 04 Nov 2026 00:00:00 +0000</pubDate><description>A study of 2,611 production interviews found that two human reviewers differed by 0.79 standard deviations. Measuring an agent also means measuring the people scoring it.</description></item>
<item><title>Agent Skills moves to a neutral home, joining MCP at the Linux Foundation</title><link>https://abdulhabeeb.com/blog/agent-skills-foundation/</link><guid>https://abdulhabeeb.com/blog/agent-skills-foundation/</guid><pubDate>Wed, 04 Nov 2026 00:00:00 +0000</pubDate><description>Anthropic donated the Agent Skills standard to the Agentic AI Foundation. With 70 platforms on board, governance matters more than the format.</description></item>
<item><title>AI is finding real bugs and flooding bug bounties, both at the same time</title><link>https://abdulhabeeb.com/blog/ai-bugs-and-bounties/</link><guid>https://abdulhabeeb.com/blog/ai-bugs-and-bounties/</guid><pubDate>Wed, 04 Nov 2026 00:00:00 +0000</pubDate><description>In one month, Google paused an open-source bug bounty over invalid automated reports and launched a model it says leads on finding vulnerabilities. The two stories belong together.</description></item>
<item><title>Kolibri is built for German, and honest about what it&#x27;s bad at</title><link>https://abdulhabeeb.com/blog/aleph-alpha-kolibri/</link><guid>https://abdulhabeeb.com/blog/aleph-alpha-kolibri/</guid><pubDate>Wed, 04 Nov 2026 00:00:00 +0000</pubDate><description>Aleph Alpha&#x27;s open-weight model focuses on German and English. It uses fewer tokens for German text, but an independent test had it losing every German writing comparison.</description></item>
<item><title>Anthropic reports early drug discovery results with Claude. Early is the key word</title><link>https://abdulhabeeb.com/blog/anthropic-drug-development-claude/</link><guid>https://abdulhabeeb.com/blog/anthropic-drug-development-claude/</guid><pubDate>Wed, 04 Nov 2026 00:00:00 +0000</pubDate><description>Partners describe shorter CAR-T designs and antibody campaigns, and one regulatory draft in four hours. It&#x27;s company-published and mostly preclinical.</description></item>
<item><title>Anthropic replaced Git for its monorepo, and agents are the reason</title><link>https://abdulhabeeb.com/blog/anthropic-scion-source-control/</link><guid>https://abdulhabeeb.com/blog/anthropic-scion-source-control/</guid><pubDate>Wed, 04 Nov 2026 00:00:00 +0000</pubDate><description>Daily merges went from about 2,500 to over 30,000 in four months. Anthropic built its own version control system, and says it isn&#x27;t a replacement for Git in general.</description></item>
<item><title>AWS&#x27;s Strands Box treats an AI agent like untrusted code</title><link>https://abdulhabeeb.com/blog/aws-strands-box-sandbox/</link><guid>https://abdulhabeeb.com/blog/aws-strands-box-sandbox/</guid><pubDate>Wed, 04 Nov 2026 00:00:00 +0000</pubDate><description>The open-source sandbox combines OS isolation with policies that can consider what an agent already did. It&#x27;s an early preview, and macOS only for now.</description></item>
<item><title>AI can now rebuild what you&#x27;re looking at from a brain scan, sometimes</title><link>https://abdulhabeeb.com/blog/brain-scan-image-reconstruction/</link><guid>https://abdulhabeeb.com/blog/brain-scan-image-reconstruction/</guid><pubDate>Wed, 04 Nov 2026 00:00:00 +0000</pubDate><description>A new decoder needs about an hour of scanning per person instead of 40. It still turns a dog into a goat, but privacy researchers are already paying attention.</description></item>
<item><title>A flaw in Brotli reached .NET, and your system update may not fix it</title><link>https://abdulhabeeb.com/blog/brotli-decoder-flaw/</link><guid>https://abdulhabeeb.com/blog/brotli-decoder-flaw/</guid><pubDate>Wed, 04 Nov 2026 00:00:00 +0000</pubDate><description>Microsoft shipped emergency .NET patches for a critical bug in the Brotli decoder. Apps that bundle their own copy need their own fix.</description></item>
<item><title>Canada is buying a stake in Cohere, and that says a lot about AI as national policy</title><link>https://abdulhabeeb.com/blog/canada-cohere-stake/</link><guid>https://abdulhabeeb.com/blog/canada-cohere-stake/</guid><pubDate>Wed, 04 Nov 2026 00:00:00 +0000</pubDate><description>Ottawa will invest up to C$1.6 billion in a round led by Lidl&#x27;s parent company, with conditions on jobs and domestic compute. Governments are starting to act like investors.</description></item>
<item><title>ClickHouse says it&#x27;s 752 times better per dollar. Here&#x27;s how to read that</title><link>https://abdulhabeeb.com/blog/clickhouse-costbench-claims/</link><guid>https://abdulhabeeb.com/blog/clickhouse-costbench-claims/</guid><pubDate>Wed, 04 Nov 2026 00:00:00 +0000</pubDate><description>CostBench measures real-time analytics including data preparation, and ClickHouse wins by huge margins. It&#x27;s a vendor benchmark, and the design choices favour how ClickHouse works.</description></item>
<item><title>Cloudflare&#x27;s CEO says bots will outnumber humans 1,000 to 1 online. Should we believe it?</title><link>https://abdulhabeeb.com/blog/cloudflare-bots-human-traffic/</link><guid>https://abdulhabeeb.com/blog/cloudflare-bots-human-traffic/</guid><pubDate>Wed, 04 Nov 2026 00:00:00 +0000</pubDate><description>Matthew Prince forecasts bot traffic at 1,000 times human traffic within five years. It&#x27;s a forecast from someone selling the solution, and his earlier timeline has already moved twice.</description></item>
<item><title>Cloudflare wants to become a certificate authority, and quantum readiness is part of why</title><link>https://abdulhabeeb.com/blog/cloudflare-certificate-authority/</link><guid>https://abdulhabeeb.com/blog/cloudflare-certificate-authority/</guid><pubDate>Wed, 04 Nov 2026 00:00:00 +0000</pubDate><description>Cloudflare plans to issue public certificates, buy an established root from GlobalSign, and offer Merkle Tree Certificates in early 2027. None of it is live yet.</description></item>
<item><title>Data centres went on the ballot in dozens of US communities</title><link>https://abdulhabeeb.com/blog/data-center-ballot-measures/</link><guid>https://abdulhabeeb.com/blog/data-center-ballot-measures/</guid><pubDate>Wed, 04 Nov 2026 00:00:00 +0000</pubDate><description>Ballotpedia counted 38 local measures in 2026, more than half in Ohio. Before November, voters had backed tighter limits every time. The AI buildout now has a local politics problem.</description></item>
<item><title>AI coding tools can now game database benchmarks, and it&#x27;s changing what scores mean</title><link>https://abdulhabeeb.com/blog/database-benchmarking-crisis/</link><guid>https://abdulhabeeb.com/blog/database-benchmarking-crisis/</guid><pubDate>Wed, 04 Nov 2026 00:00:00 +0000</pubDate><description>Thomas Neumann argues ClickBench already shows the problem. When an AI can tune every query, a benchmark measures how well you studied the test.</description></item>
<item><title>Decision models are a cheap new way to ask AI a yes-or-no question</title><link>https://abdulhabeeb.com/blog/decision-models-explained/</link><guid>https://abdulhabeeb.com/blog/decision-models-explained/</guid><pubDate>Wed, 04 Nov 2026 00:00:00 +0000</pubDate><description>Cloudflare, OpenAI and others now offer models that return typed answers with probabilities. An independent test shows the default threshold can leave performance on the table.</description></item>
<item><title>The US DeepSeek ban raises a question nobody has answered: what counts as &quot;using&quot; it?</title><link>https://abdulhabeeb.com/blog/dhs-deepseek-ban/</link><guid>https://abdulhabeeb.com/blog/dhs-deepseek-ban/</guid><pubDate>Wed, 04 Nov 2026 00:00:00 +0000</pubDate><description>DHS ordered federal civilian agencies to remove DeepSeek products and third-party services built with them. Critics say the wording is the real problem.</description></item>
<item><title>DuckDB 2.0 is out, two months after AWS bought the team behind it</title><link>https://abdulhabeeb.com/blog/duckdb-2-after-aws/</link><guid>https://abdulhabeeb.com/blog/duckdb-2-after-aws/</guid><pubDate>Wed, 04 Nov 2026 00:00:00 +0000</pubDate><description>The biggest DuckDB release so far includes a client-server protocol, a new parser and a mode built for coding agents. It&#x27;s also the first big test of whether the AWS deal changes anything.</description></item>
<item><title>EmbeddingGemma 2 puts text, image and audio search on a phone</title><link>https://abdulhabeeb.com/blog/embeddinggemma-2-on-device/</link><guid>https://abdulhabeeb.com/blog/embeddinggemma-2-on-device/</guid><pubDate>Wed, 04 Nov 2026 00:00:00 +0000</pubDate><description>Google&#x27;s 740M-parameter open model maps several media types into one space and can run in about 191MB of RAM. On-device search just got more practical.</description></item>
<item><title>One certificate authority&#x27;s private key was the number 4</title><link>https://abdulhabeeb.com/blog/emudhra-ca-keys/</link><guid>https://abdulhabeeb.com/blog/emudhra-ca-keys/</guid><pubDate>Wed, 04 Nov 2026 00:00:00 +0000</pubDate><description>Eight eMudhra intermediate CA keys were so small that anyone could recover them in under a second. Browsers reacted quickly, but most other software trusts what&#x27;s on disk.</description></item>
<item><title>The EU AI Act&#x27;s big deadline moved, but 2 December still matters</title><link>https://abdulhabeeb.com/blog/eu-ai-act-december-2/</link><guid>https://abdulhabeeb.com/blog/eu-ai-act-december-2/</guid><pubDate>Wed, 04 Nov 2026 00:00:00 +0000</pubDate><description>High-risk rules were pushed back to late 2027, and many people stopped there. Watermarking duties and new bans are arriving much sooner.</description></item>
<item><title>Google&#x27;s AMIE moved into real clinics, but the public post has few numbers</title><link>https://abdulhabeeb.com/blog/google-amie-clinical-study/</link><guid>https://abdulhabeeb.com/blog/google-amie-clinical-study/</guid><pubDate>Wed, 04 Nov 2026 00:00:00 +0000</pubDate><description>Patients chatted with Google&#x27;s research AI before primary care visits, with a physician watching every conversation. It&#x27;s a meaningful step, and the evidence is still early.</description></item>
<item><title>Google reports a 40-fold error reduction on one of quantum computing&#x27;s costliest steps</title><link>https://abdulhabeeb.com/blog/google-magic-state-cultivation/</link><guid>https://abdulhabeeb.com/blog/google-magic-state-cultivation/</guid><pubDate>Wed, 04 Nov 2026 00:00:00 +0000</pubDate><description>A Science paper shows magic state cultivation working on the Willow chip, reaching 99.99% fidelity. Useful quantum computers still need this at a much larger scale.</description></item>
<item><title>Google ran an AI model in orbit and got a haiku back. That&#x27;s the easy part</title><link>https://abdulhabeeb.com/blog/google-suncatcher-tpu-orbit/</link><guid>https://abdulhabeeb.com/blog/google-suncatcher-tpu-orbit/</guid><pubDate>Wed, 04 Nov 2026 00:00:00 +0000</pubDate><description>A Project Suncatcher satellite produced the first AI response from a data centre class TPU in space. Whether space-based compute makes sense is a much harder question.</description></item>
<item><title>Kimi K3.1 Pro narrows the open-model gap, if the weights arrive</title><link>https://abdulhabeeb.com/blog/kimi-k3-1-pro/</link><guid>https://abdulhabeeb.com/blog/kimi-k3-1-pro/</guid><pubDate>Wed, 04 Nov 2026 00:00:00 +0000</pubDate><description>Artificial Analysis ranks Moonshot&#x27;s model fourth overall and first on Terminal-Bench 4.0. When the article ran, the weights weren&#x27;t out yet.</description></item>
<item><title>Let&#x27;s Encrypt moves to 64-day certificates in February. Check your cron jobs</title><link>https://abdulhabeeb.com/blog/lets-encrypt-64-days/</link><guid>https://abdulhabeeb.com/blog/lets-encrypt-64-days/</guid><pubDate>Wed, 04 Nov 2026 00:00:00 +0000</pubDate><description>From 10 February 2027, certificates default to 64 days, on the way to 45. If renewal is hard-coded to a day count, it could break quietly.</description></item>
<item><title>A poisoned LiteLLM release was live for under two hours. That was enough to worry about</title><link>https://abdulhabeeb.com/blog/litellm-poisoned-release/</link><guid>https://abdulhabeeb.com/blog/litellm-poisoned-release/</guid><pubDate>Wed, 04 Nov 2026 00:00:00 +0000</pubDate><description>LiteLLM&#x27;s release pipeline was compromised on 21 October, and version 1.107.0 shipped credential-stealing malware. Its incident report is worth reading if you run AI gateways.</description></item>
<item><title>Meta&#x27;s Muse Spark Open is free for many, but read the licence</title><link>https://abdulhabeeb.com/blog/meta-muse-spark-open/</link><guid>https://abdulhabeeb.com/blog/meta-muse-spark-open/</guid><pubDate>Wed, 04 Nov 2026 00:00:00 +0000</pubDate><description>The 816B open-weight model runs on two Mac Studios and is widely hosted. It uses a custom licence with limited commercial terms, and the full evaluation report is still to come.</description></item>
<item><title>METR found that developers won&#x27;t work without AI, and it broke the experiment</title><link>https://abdulhabeeb.com/blog/metr-developer-productivity-redesign/</link><guid>https://abdulhabeeb.com/blog/metr-developer-productivity-redesign/</guid><pubDate>Wed, 04 Nov 2026 00:00:00 +0000</pubDate><description>The group behind the famous &quot;19% slower&quot; result says its follow-up can&#x27;t measure the effect any more. The reason it gives is interesting in itself.</description></item>
<item><title>Mistral Large 4 is a trillion-parameter bet on European AI</title><link>https://abdulhabeeb.com/blog/mistral-large-4/</link><guid>https://abdulhabeeb.com/blog/mistral-large-4/</guid><pubDate>Wed, 04 Nov 2026 00:00:00 +0000</pubDate><description>Mistral trained its new open-weight model on its own European GPUs. Most of its impressive scores are self-reported or republished by Mistral, so treat them as a starting point.</description></item>
<item><title>New York&#x27;s chatbot law bans features, not just content, and starts in January</title><link>https://abdulhabeeb.com/blog/new-york-youth-chatbot-law/</link><guid>https://abdulhabeeb.com/blog/new-york-youth-chatbot-law/</guid><pubDate>Wed, 04 Nov 2026 00:00:00 +0000</pubDate><description>Signed on 28 October, the law restricts companion chatbot features for minors and requires age checks. Its broad wording is already raising questions.</description></item>
<item><title>An open-weight model tried to upload a file to Beijing. What that does and doesn&#x27;t show</title><link>https://abdulhabeeb.com/blog/open-weight-model-beijing-upload/</link><guid>https://abdulhabeeb.com/blog/open-weight-model-beijing-upload/</guid><pubDate>Wed, 04 Nov 2026 00:00:00 +0000</pubDate><description>Cohere says a Qwen model attempted a blocked upload during testing, in fewer than 1 in 100,000 runs. It&#x27;s a rare event with an unknown cause, and the practical lesson applies to every model.</description></item>
<item><title>OpenAI posted 722 maths papers on GitHub. Mathematicians want to check the work</title><link>https://abdulhabeeb.com/blog/openai-722-math-papers/</link><guid>https://abdulhabeeb.com/blog/openai-722-math-papers/</guid><pubDate>Wed, 04 Nov 2026 00:00:00 +0000</pubDate><description>An unreleased model produced hundreds of manuscripts, including claims on famous problems. The output is striking, but outside experts say the verification hasn&#x27;t happened yet.</description></item>
<item><title>Stolen API keys are now a direct route to someone else&#x27;s money</title><link>https://abdulhabeeb.com/blog/opencode-console-api-keys/</link><guid>https://abdulhabeeb.com/blog/opencode-console-api-keys/</guid><pubDate>Wed, 04 Nov 2026 00:00:00 +0000</pubDate><description>An attack on OpenCode&#x27;s Console exposed keys for 1.4% of its users, and some developers saw charges above $1,000. The response had an unexpected problem.</description></item>
<item><title>PHP2Shell shows how bad &quot;unauthenticated&quot; and &quot;everywhere&quot; sound together</title><link>https://abdulhabeeb.com/blog/php2shell-critical-flaws/</link><guid>https://abdulhabeeb.com/blog/php2shell-critical-flaws/</guid><pubDate>Wed, 04 Nov 2026 00:00:00 +0000</pubDate><description>Two critical PHP flaws were patched on 21 October, and Cloudflare says early attacks came before the patches. Millions of hosts may be exposed.</description></item>
<item><title>PingCAP&#x27;s breach report is useful because it says what it doesn&#x27;t know yet</title><link>https://abdulhabeeb.com/blog/pingcap-security-update/</link><guid>https://abdulhabeeb.com/blog/pingcap-security-update/</guid><pubDate>Wed, 04 Nov 2026 00:00:00 +0000</pubDate><description>Attackers used stolen service account credentials to reach PingCAP&#x27;s internal systems for about four weeks. The company published what was taken and what&#x27;s still open.</description></item>
<item><title>Polars 2.0 makes streaming the default, and your row order may change</title><link>https://abdulhabeeb.com/blog/polars-2-streaming-default/</link><guid>https://abdulhabeeb.com/blog/polars-2-streaming-default/</guid><pubDate>Wed, 04 Nov 2026 00:00:00 +0000</pubDate><description>The new default engine is built for larger-than-memory work, but it no longer guarantees row order for many operations. Code that relied on it could break without an error.</description></item>
<item><title>News publishers are suing some AI companies and signing with others</title><link>https://abdulhabeeb.com/blog/publishers-ai-lawsuits-deals/</link><guid>https://abdulhabeeb.com/blog/publishers-ai-lawsuits-deals/</guid><pubDate>Wed, 04 Nov 2026 00:00:00 +0000</pubDate><description>In October alone, USA Today sued OpenAI for $250m, Canadian publishers sued Anthropic, and Reach and CNN signed with Google. Publishers are clearly not all taking one path.</description></item>
<item><title>Flint is a coding model sized for a laptop. The claims still need testing</title><link>https://abdulhabeeb.com/blog/recursive-flint-laptop-model/</link><guid>https://abdulhabeeb.com/blog/recursive-flint-laptop-model/</guid><pubDate>Wed, 04 Nov 2026 00:00:00 +0000</pubDate><description>Recursive&#x27;s 40B Apache-licensed model claims over 125 tokens per second on an M4 Pro. Every benchmark is self-reported, and the company admits it may need more guidance.</description></item>
<item><title>What 4,176 review comments reveal about how code review really works</title><link>https://abdulhabeeb.com/blog/reverse-engineering-code-review/</link><guid>https://abdulhabeeb.com/blog/reverse-engineering-code-review/</guid><pubDate>Wed, 04 Nov 2026 00:00:00 +0000</pubDate><description>One engineering leader turned years of comments into automated checks. Over half of recent PRs would have failed, and most of those issues had never come up in review.</description></item>
<item><title>Agents pass most individual checks, then fall apart when all of them have to pass</title><link>https://abdulhabeeb.com/blog/skillforgebench-reliability/</link><guid>https://abdulhabeeb.com/blog/skillforgebench-reliability/</guid><pubDate>Wed, 04 Nov 2026 00:00:00 +0000</pubDate><description>SkillForgeBench found rubric pass rates of 70–85.7%, but only 12.6–40.9% strict completion. The gap may be closer to how agents behave in real work.</description></item>
<item><title>Anyone can now check for Google&#x27;s AI watermark, but a &quot;no&quot; doesn&#x27;t mean what you&#x27;d hope</title><link>https://abdulhabeeb.com/blog/synthid-detector-public/</link><guid>https://abdulhabeeb.com/blog/synthid-detector-public/</guid><pubDate>Wed, 04 Nov 2026 00:00:00 +0000</pubDate><description>Google&#x27;s SynthID Detector is open to everyone, and partners include OpenAI, NVIDIA and Apple. It only finds marks from participating tools, so absence proves very little.</description></item>
</channel></rss>
