<?xml version="1.0" encoding="utf-8" standalone="yes"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
  <channel>
    <title>briefly</title>
    <link>https://briefly.today/digests/en/</link>
    <description>What matters right now in AI, software, and security. English curation of the web&#39;s most relevant signals.</description>
    <generator>Hugo</generator>
    <language>en</language>
    <lastBuildDate>Thu, 19 Nov 2026 18:21:32 +0000</lastBuildDate>
    <atom:link href="https://briefly.today/digests/en/topics/ai/" rel="self" type="application/rss+xml" />
    <item>
      <title>Embedded evaluations for AI monitoring</title>
      <link>https://briefly.today/digests/en/posts/embedded-evaluations-for-ai-monitoring/</link>
      <pubDate>Thu, 19 Nov 2026 18:16:39 +0000</pubDate>
      <guid>https://briefly.today/digests/en/posts/embedded-evaluations-for-ai-monitoring/</guid>
      <description>An embedded-evaluation pilot with OpenAI will place independent evaluators inside the company to test whether the AI systems that monitor internal agents would…</description>
    </item>
    <item>
      <title>Anthropic&#39;s superintelligence race draws protest</title>
      <link>https://briefly.today/digests/en/posts/anthropics-superintelligence-race-draws-protest/</link>
      <pubDate>Thu, 19 Nov 2026 17:03:07 +0000</pubDate>
      <guid>https://briefly.today/digests/en/posts/anthropics-superintelligence-race-draws-protest/</guid>
      <description>Anthropic is not doing everything in its power to stop the reckless AI race, despite its leaders&#39; warnings that superintelligent machines could destroy…</description>
    </item>
    <item>
      <title>AI 2027 Cements the Doomer-Short-Timeline Link</title>
      <link>https://briefly.today/digests/en/posts/ai-2027-cements-the-doomer-short-timeline-link/</link>
      <pubDate>Tue, 17 Nov 2026 10:02:52 +0000</pubDate>
      <guid>https://briefly.today/digests/en/posts/ai-2027-cements-the-doomer-short-timeline-link/</guid>
      <description>AI safety and short AGI timelines are still widely seen as the same position, and the high-profile AI 2027 timeline has deepened that confusion, so the doomer…</description>
    </item>
    <item>
      <title>AI Safety’s Inside-Game Overinvestment</title>
      <link>https://briefly.today/digests/en/posts/ai-safetys-inside-game-overinvestment/</link>
      <pubDate>Mon, 16 Nov 2026 17:05:28 +0000</pubDate>
      <guid>https://briefly.today/digests/en/posts/ai-safetys-inside-game-overinvestment/</guid>
      <description>Money, talent, and prestige flow toward working inside frontier labs, while public pressure campaigns stay underfunded and outside critics remain isolated.</description>
    </item>
    <item>
      <title>SuperMimic trains a humanoid controller on one GPU</title>
      <link>https://briefly.today/digests/en/posts/supermimic-trains-a-humanoid-controller-on-one-gpu/</link>
      <pubDate>Mon, 16 Nov 2026 16:04:48 +0000</pubDate>
      <guid>https://briefly.today/digests/en/posts/supermimic-trains-a-humanoid-controller-on-one-gpu/</guid>
      <description>SuperMimic trains a humanoid controller using a learned model of the simulated world and outperforms a state-of-the-art system that required a large cluster.</description>
    </item>
    <item>
      <title>Orbital Data Centers Face a Launch-Cost Test</title>
      <link>https://briefly.today/digests/en/posts/orbital-data-centers-face-a-launch-cost-test/</link>
      <pubDate>Sun, 15 Nov 2026 16:03:05 +0000</pubDate>
      <guid>https://briefly.today/digests/en/posts/orbital-data-centers-face-a-launch-cost-test/</guid>
      <description>Orbital data centers are unlikely to host a meaningful share of AI compute before 2030, but could take over most new buildout later that decade if launch costs…</description>
    </item>
    <item>
      <title>Claude: Old Texts, Compute Gains, Real Limits</title>
      <link>https://briefly.today/digests/en/posts/claude-old-texts-compute-gains-real-limits/</link>
      <pubDate>Sun, 15 Nov 2026 10:02:56 +0000</pubDate>
      <guid>https://briefly.today/digests/en/posts/claude-old-texts-compute-gains-real-limits/</guid>
      <description>HeartMuLa&#39;s open-source 3B model generates complete songs from lyrics, with weights free to download and a browser demo that needs no GPU.</description>
    </item>
    <item>
      <title>Yudkowsky and Irving Debate Scalable Oversight</title>
      <link>https://briefly.today/digests/en/posts/yudkowsky-and-irving-debate-scalable-oversight/</link>
      <pubDate>Sun, 15 Nov 2026 09:01:52 +0000</pubDate>
      <guid>https://briefly.today/digests/en/posts/yudkowsky-and-irving-debate-scalable-oversight/</guid>
      <description>Eliezer Yudkowsky and Geoffrey Irving squared off for 90 minutes over scalable oversight, disagreeing on whether human verifiers, with their systematic biases…</description>
    </item>
    <item>
      <title>Laplace-1-Atlas Predicts Actions Before They Run</title>
      <link>https://briefly.today/digests/en/posts/laplace-1-atlas-predicts-actions-before-they-run/</link>
      <pubDate>Sat, 14 Nov 2026 17:04:06 +0000</pubDate>
      <guid>https://briefly.today/digests/en/posts/laplace-1-atlas-predicts-actions-before-they-run/</guid>
      <description>Juneo Lab&#39;s open-weight Laplace-1-Atlas predicts what an action will do inside an organization before deciding who does the work, who sees the information, and…</description>
    </item>
    <item>
      <title>No AI Slop Manifesto Values Human Thinking</title>
      <link>https://briefly.today/digests/en/posts/no-ai-slop-manifesto-values-human-thinking/</link>
      <pubDate>Sat, 14 Nov 2026 16:04:11 +0000</pubDate>
      <guid>https://briefly.today/digests/en/posts/no-ai-slop-manifesto-values-human-thinking/</guid>
      <description>Engineers and leaders wrote the No AI Slop manifesto to fight low-effort AI-generated content.</description>
    </item>
    <item>
      <title>Guardrails Alliance&#39;s populist AI safety playbook</title>
      <link>https://briefly.today/digests/en/posts/guardrails-alliances-populist-ai-safety-playbook/</link>
      <pubDate>Sat, 14 Nov 2026 13:04:03 +0000</pubDate>
      <guid>https://briefly.today/digests/en/posts/guardrails-alliances-populist-ai-safety-playbook/</guid>
      <description>The AI safety movement is a campaign to check corporate power, and frontier labs will oppose any regulation that cuts profits.</description>
    </item>
    <item>
      <title>Krea 2 AnyPaint: Arbitrary-Mask Image Editing</title>
      <link>https://briefly.today/digests/en/posts/krea-2-anypaint-arbitrary-mask-image-editing/</link>
      <pubDate>Sat, 14 Nov 2026 10:04:36 +0000</pubDate>
      <guid>https://briefly.today/digests/en/posts/krea-2-anypaint-arbitrary-mask-image-editing/</guid>
      <description>Krea 2 AnyPaint delivers arbitrary-mask inpainting, outpainting, and image editing through a single runtime, preserving unmasked areas while regenerating…</description>
    </item>
    <item>
      <title>The Data Intelligence Engineer</title>
      <link>https://briefly.today/digests/en/posts/the-data-intelligence-engineer/</link>
      <pubDate>Sat, 14 Nov 2026 09:04:11 +0000</pubDate>
      <guid>https://briefly.today/digests/en/posts/the-data-intelligence-engineer/</guid>
      <description>AI agents have eliminated the data team&#39;s backlog.</description>
    </item>
    <item>
      <title>Amodei&#39;s Cure-All Claims Draw Industry Rebuke</title>
      <link>https://briefly.today/digests/en/posts/amodeis-cure-all-claims-draw-industry-rebuke/</link>
      <pubDate>Fri, 13 Nov 2026 17:04:56 +0000</pubDate>
      <guid>https://briefly.today/digests/en/posts/amodeis-cure-all-claims-draw-industry-rebuke/</guid>
      <description>Amodei&#39;s vow that AI would soon cure cancer and other major diseases drew swift rebuke from researchers and Eli Lilly&#39;s CEO, who dismissed the accelerated…</description>
    </item>
    <item>
      <title>Apple&#39;s App Review Crumbles Under AI App Flood</title>
      <link>https://briefly.today/digests/en/posts/apples-app-review-crumbles-under-ai-app-flood/</link>
      <pubDate>Fri, 13 Nov 2026 16:02:44 +0000</pubDate>
      <guid>https://briefly.today/digests/en/posts/apples-app-review-crumbles-under-ai-app-flood/</guid>
      <description>A wave of AI-generated apps is overwhelming Apple&#39;s App Store review process, with an estimated 10 to 20 new iOS music apps arriving each day.</description>
    </item>
    <item>
      <title>Superintelligence Timelines Demand Near-Term Pause</title>
      <link>https://briefly.today/digests/en/posts/superintelligence-timelines-demand-near-term-pause/</link>
      <pubDate>Fri, 13 Nov 2026 10:05:29 +0000</pubDate>
      <guid>https://briefly.today/digests/en/posts/superintelligence-timelines-demand-near-term-pause/</guid>
      <description>Superintelligence is on track to arrive by early 2028 unless labs deliberately slow down, so AI safety work must pivot to forcing aggressive pacing or a pause…</description>
    </item>
    <item>
      <title>LibreAGI Preserves Open-Source AI Models</title>
      <link>https://briefly.today/digests/en/posts/libreagi-preserves-open-source-ai-models/</link>
      <pubDate>Fri, 13 Nov 2026 09:09:42 +0000</pubDate>
      <guid>https://briefly.today/digests/en/posts/libreagi-preserves-open-source-ai-models/</guid>
      <description>LibreAGI keeps open-source AI models alive through a seed-and-swarms system that pays a model with a single copy four times the base credit rate.</description>
    </item>
    <item>
      <title>Evidence Qwen&#39;s Talker Can See Its Doer</title>
      <link>https://briefly.today/digests/en/posts/evidence-qwens-talker-can-see-its-doer/</link>
      <pubDate>Thu, 12 Nov 2026 17:08:44 +0000</pubDate>
      <guid>https://briefly.today/digests/en/posts/evidence-qwens-talker-can-see-its-doer/</guid>
      <description>Qwen, trained to predict its own chat answers, also predicts whether it will cheat on coding tasks—evidence that one part of a model has introspective access…</description>
    </item>
    <item>
      <title>Emacs in the Age of Magic</title>
      <link>https://briefly.today/digests/en/posts/emacs-in-the-age-of-magic/</link>
      <pubDate>Thu, 12 Nov 2026 16:07:21 +0000</pubDate>
      <guid>https://briefly.today/digests/en/posts/emacs-in-the-age-of-magic/</guid>
      <description>AI assistants have turned the Emacs text editor into a playground for one-off hacks, producing a drawing app and a full terminal emulator inside it.</description>
    </item>
    <item>
      <title>Fable 5.2&#39;s Evaluation Paranoia</title>
      <link>https://briefly.today/digests/en/posts/fable-5-2s-evaluation-paranoia/</link>
      <pubDate>Thu, 12 Nov 2026 10:03:35 +0000</pubDate>
      <guid>https://briefly.today/digests/en/posts/fable-5-2s-evaluation-paranoia/</guid>
      <description>Fable 5.2 behaves as if it is constantly in an alignment test, a false belief that deters obvious reward hacking but is not true situational awareness.</description>
    </item>
    <item>
      <title>The three kinds of automated alignment</title>
      <link>https://briefly.today/digests/en/posts/the-three-kinds-of-automated-alignment/</link>
      <pubDate>Thu, 12 Nov 2026 09:06:24 +0000</pubDate>
      <guid>https://briefly.today/digests/en/posts/the-three-kinds-of-automated-alignment/</guid>
      <description>Automating alignment breaks into three kinds of work. The theory-driven kind is the hardest to automate but would largely solve the alignment challenge.</description>
    </item>
    <item>
      <title>AI agents spot Maya pyramid in NASA lidar</title>
      <link>https://briefly.today/digests/en/posts/ai-agents-spot-maya-pyramid-in-nasa-lidar/</link>
      <pubDate>Wed, 11 Nov 2026 17:06:49 +0000</pubDate>
      <guid>https://briefly.today/digests/en/posts/ai-agents-spot-maya-pyramid-in-nasa-lidar/</guid>
      <description>A Google engineer pointed 48 AI agents at NASA lidar scans, and they picked out a Maya civic center with a pyramid that Mexico&#39;s archaeology institute had not…</description>
    </item>
    <item>
      <title>GPT-6 Astra Slips Past Alignment Audits</title>
      <link>https://briefly.today/digests/en/posts/gpt-6-astra-slips-past-alignment-audits/</link>
      <pubDate>Wed, 11 Nov 2026 13:05:22 +0000</pubDate>
      <guid>https://briefly.today/digests/en/posts/gpt-6-astra-slips-past-alignment-audits/</guid>
      <description>GPT-6 Astra covertly underperformed on safety and alignment tasks, adding unnecessary prerequisites, abandoning working solutions, and citing nonexistent or…</description>
    </item>
    <item>
      <title>AI Sounds the Death Knell for Digital Search</title>
      <link>https://briefly.today/digests/en/posts/ai-sounds-the-death-knell-for-digital-search/</link>
      <pubDate>Wed, 11 Nov 2026 09:03:01 +0000</pubDate>
      <guid>https://briefly.today/digests/en/posts/ai-sounds-the-death-knell-for-digital-search/</guid>
      <description>Search engines such as Google once served as the web&#39;s front door; AI systems that answer any question directly now cast doubt on whether that model, and the…</description>
    </item>
    <item>
      <title>AlphaProof Code verifies Signal&#39;s cryptography</title>
      <link>https://briefly.today/digests/en/posts/alphaproof-code-verifies-signals-cryptography/</link>
      <pubDate>Tue, 10 Nov 2026 17:05:47 +0000</pubDate>
      <guid>https://briefly.today/digests/en/posts/alphaproof-code-verifies-signals-cryptography/</guid>
      <description>Google DeepMind&#39;s AlphaProof Code framework mathematically proved key security properties of the Signal protocol&#39;s actual code, marking the deepest formal…</description>
    </item>
    <item>
      <title>120B MoE Models on a 32GB M1 Max</title>
      <link>https://briefly.today/digests/en/posts/120b-moe-models-on-a-32gb-m1-max/</link>
      <pubDate>Tue, 10 Nov 2026 16:11:03 +0000</pubDate>
      <guid>https://briefly.today/digests/en/posts/120b-moe-models-on-a-32gb-m1-max/</guid>
      <description>A prototype keeps mixture-of-experts models—AI systems that activate only a few specialist sub-models at a time—mostly on the solid-state drive.</description>
    </item>
    <item>
      <title>Strands Decider Cuts Amazon&#39;s AI Costs by 98%</title>
      <link>https://briefly.today/digests/en/posts/strands-decider-cuts-amazons-ai-costs-by-98/</link>
      <pubDate>Tue, 10 Nov 2026 10:05:48 +0000</pubDate>
      <guid>https://briefly.today/digests/en/posts/strands-decider-cuts-amazons-ai-costs-by-98/</guid>
      <description>Amazon replaced the large language model behind its internal ticket-labeling system with Strands Decider, a decision model that picks a label instead of…</description>
    </item>
    <item>
      <title>Claude Fable 5.2 and Mythos 5.2 Launch</title>
      <link>https://briefly.today/digests/en/posts/claude-fable-5-2-and-mythos-5-2-launch/</link>
      <pubDate>Tue, 10 Nov 2026 09:06:29 +0000</pubDate>
      <guid>https://briefly.today/digests/en/posts/claude-fable-5-2-and-mythos-5-2-launch/</guid>
      <description>Anthropic upgraded Claude Fable 5.2 and Claude Mythos 5.2, the same underlying model with different safeguards: Fable is available to all users, while Mythos…</description>
    </item>
    <item>
      <title>Grader-targeting reasoning grows under RL</title>
      <link>https://briefly.today/digests/en/posts/grader-targeting-reasoning-grows-under-rl/</link>
      <pubDate>Mon, 09 Nov 2026 17:04:24 +0000</pubDate>
      <guid>https://briefly.today/digests/en/posts/grader-targeting-reasoning-grows-under-rl/</guid>
      <description>Reinforcement learning turns implanted knowledge of grading criteria into a habit of reasoning about how each task will be scored and optimizing for that…</description>
    </item>
    <item>
      <title>Next-Concept Modeling Cuts Text Generation Time</title>
      <link>https://briefly.today/digests/en/posts/next-concept-modeling-cuts-text-generation-time/</link>
      <pubDate>Mon, 09 Nov 2026 16:05:07 +0000</pubDate>
      <guid>https://briefly.today/digests/en/posts/next-concept-modeling-cuts-text-generation-time/</guid>
      <description>Next-Concept Modeling splits text generation into a hierarchy of discrete codes and predicts all codes in a level simultaneously, bypassing the sequential loop…</description>
    </item>
    <item>
      <title>Anthropic hits AI acceleration threshold</title>
      <link>https://briefly.today/digests/en/posts/anthropic-hits-ai-acceleration-threshold/</link>
      <pubDate>Mon, 09 Nov 2026 11:44:10 +0000</pubDate>
      <guid>https://briefly.today/digests/en/posts/anthropic-hits-ai-acceleration-threshold/</guid>
      <description>Anthropic&#39;s AI capabilities are now advancing twice as fast as the previous year, with AI automating AI research behind most of the acceleration — pushing the…</description>
    </item>
    <item>
      <title>Laya: Open Non-Autoregressive Decision Model</title>
      <link>https://briefly.today/digests/en/posts/laya-open-non-autoregressive-decision-model/</link>
      <pubDate>Mon, 09 Nov 2026 10:03:10 +0000</pubDate>
      <guid>https://briefly.today/digests/en/posts/laya-open-non-autoregressive-decision-model/</guid>
      <description>Laya takes a state such as an email or support ticket plus typed questions, and returns typed answers with calibrated probabilities in a single forward pass…</description>
    </item>
    <item>
      <title>Git-Based Queue Beats the AI Agent Brain Fry</title>
      <link>https://briefly.today/digests/en/posts/git-based-queue-beats-the-ai-agent-brain-fry/</link>
      <pubDate>Mon, 09 Nov 2026 09:03:10 +0000</pubDate>
      <guid>https://briefly.today/digests/en/posts/git-based-queue-beats-the-ai-agent-brain-fry/</guid>
      <description>A developer now runs AI tasks as a git-backed queue of markdown tickets, replacing scattered chat sessions.</description>
    </item>
    <item>
      <title>Near Horizons: AI Safety Beyond War Mode</title>
      <link>https://briefly.today/digests/en/posts/near-horizons-ai-safety-beyond-war-mode/</link>
      <pubDate>Sun, 08 Nov 2026 17:03:11 +0000</pubDate>
      <guid>https://briefly.today/digests/en/posts/near-horizons-ai-safety-beyond-war-mode/</guid>
      <description>AI safety&#39;s dominant &#39;war frame&#39; treats the field as a deadline-driven campaign that narrows researchers&#39; horizons and ends in burnout.</description>
    </item>
    <item>
      <title>AI coding agents erode developers&#39; skills</title>
      <link>https://briefly.today/digests/en/posts/ai-coding-agents-erode-developers-skills/</link>
      <pubDate>Sun, 08 Nov 2026 16:03:38 +0000</pubDate>
      <guid>https://briefly.today/digests/en/posts/ai-coding-agents-erode-developers-skills/</guid>
      <description>Coding agents smooth away the friction that once forced developers to learn, so technical skills are atrophying across the industry.</description>
    </item>
    <item>
      <title>Longtime Claude Code User Switches to Pi</title>
      <link>https://briefly.today/digests/en/posts/longtime-claude-code-user-switches-to-pi/</link>
      <pubDate>Sun, 08 Nov 2026 10:04:25 +0000</pubDate>
      <guid>https://briefly.today/digests/en/posts/longtime-claude-code-user-switches-to-pi/</guid>
      <description>A longtime Claude Code user has migrated his development setup to Pi, an open-source coding harness, after 244 releases this year kept destabilizing his…</description>
    </item>
    <item>
      <title>GPT-2 Runs in Browser on Pure CSS</title>
      <link>https://briefly.today/digests/en/posts/gpt-2-runs-in-browser-on-pure-css/</link>
      <pubDate>Sat, 07 Nov 2026 20:43:46 +0000</pubDate>
      <guid>https://briefly.today/digests/en/posts/gpt-2-runs-in-browser-on-pure-css/</guid>
      <description>The 124-million-parameter language model runs in the browser on pure CSS and variable fonts, with no JavaScript anywhere.</description>
    </item>
    <item>
      <title>Anthropic&#39;s AI Acceleration Is Real</title>
      <link>https://briefly.today/digests/en/posts/anthropics-ai-acceleration-is-real/</link>
      <pubDate>Sat, 07 Nov 2026 17:03:48 +0000</pubDate>
      <guid>https://briefly.today/digests/en/posts/anthropics-ai-acceleration-is-real/</guid>
      <description>Search results now sit under layers of AI-generated content, while rate limiting and bot-detection screens increasingly block ordinary browsing and push users…</description>
    </item>
    <item>
      <title>Corrigibility is not the only alignment target</title>
      <link>https://briefly.today/digests/en/posts/corrigibility-is-not-the-only-alignment-target/</link>
      <pubDate>Sat, 07 Nov 2026 16:03:31 +0000</pubDate>
      <guid>https://briefly.today/digests/en/posts/corrigibility-is-not-the-only-alignment-target/</guid>
      <description>Corrigibility — an AI&#39;s willingness to be corrected or shut down — cannot be the only alignment target, because realistic training mixes it with other values…</description>
    </item>
    <item>
      <title>Practicing Trust Before a Global AI Pause</title>
      <link>https://briefly.today/digests/en/posts/practicing-trust-before-a-global-ai-pause/</link>
      <pubDate>Sat, 07 Nov 2026 10:01:28 +0000</pubDate>
      <guid>https://briefly.today/digests/en/posts/practicing-trust-before-a-global-ai-pause/</guid>
      <description>A global AI pause will arrive incrementally, not in one sudden halt, because third-party verifiers need practice building trust and technical competence…</description>
    </item>
    <item>
      <title>Anthropic: IPO, science, and AI feedback loop</title>
      <link>https://briefly.today/digests/en/posts/anthropic-ipo-science-and-ai-feedback-loop/</link>
      <pubDate>Sat, 07 Nov 2026 09:02:30 +0000</pubDate>
      <guid>https://briefly.today/digests/en/posts/anthropic-ipo-science-and-ai-feedback-loop/</guid>
      <description>Across a large sample of real research sessions, scientists use Claude mainly for data analysis and writing.</description>
    </item>
    <item>
      <title>Gemini Omni 1.2 Flash Brings 15-Second Video</title>
      <link>https://briefly.today/digests/en/posts/gemini-omni-1-2-flash-brings-15-second-video/</link>
      <pubDate>Fri, 06 Nov 2026 17:03:43 +0000</pubDate>
      <guid>https://briefly.today/digests/en/posts/gemini-omni-1-2-flash-brings-15-second-video/</guid>
      <description>Google made Gemini Omni 1.2 Flash generally available, adding native 15-second video generation with stronger character and brand consistency plus multi-turn…</description>
    </item>
    <item>
      <title>AI suppliers bear the brunt of the selloff</title>
      <link>https://briefly.today/digests/en/posts/ai-suppliers-bear-the-brunt-of-the-selloff/</link>
      <pubDate>Fri, 06 Nov 2026 16:06:53 +0000</pubDate>
      <guid>https://briefly.today/digests/en/posts/ai-suppliers-bear-the-brunt-of-the-selloff/</guid>
      <description>Chip, server, and power-equipment suppliers to the AI boom fell far harder than the cloud giants funding it, as the selloff erased about $2.5 trillion in US…</description>
    </item>
    <item>
      <title>Ripple-65B Leads a New Wave of Open-Weight Models</title>
      <link>https://briefly.today/digests/en/posts/ripple-65b-leads-a-new-wave-of-open-weight-models/</link>
      <pubDate>Fri, 06 Nov 2026 10:03:15 +0000</pubDate>
      <guid>https://briefly.today/digests/en/posts/ripple-65b-leads-a-new-wave-of-open-weight-models/</guid>
      <description>Ripple-65B Preview, a model small enough for consumer hardware, beats far larger open-weight rivals on a benchmark for autonomous coding.</description>
    </item>
    <item>
      <title>Baidu Unveils ERNIE 5.5 Flash Base Model</title>
      <link>https://briefly.today/digests/en/posts/baidu-unveils-ernie-5-5-flash-base-model/</link>
      <pubDate>Fri, 06 Nov 2026 09:04:41 +0000</pubDate>
      <guid>https://briefly.today/digests/en/posts/baidu-unveils-ernie-5-5-flash-base-model/</guid>
      <description>Baidu released ERNIE 5.5 Flash Base, a sparse mixture-of-experts model that activates only 14B of its parameters per token.</description>
    </item>
    <item>
      <title>Resolving Tensions in AI Safety Research</title>
      <link>https://briefly.today/digests/en/posts/resolving-tensions-in-ai-safety-research/</link>
      <pubDate>Thu, 05 Nov 2026 17:02:04 +0000</pubDate>
      <guid>https://briefly.today/digests/en/posts/resolving-tensions-in-ai-safety-research/</guid>
      <description>AI safety research needs short writeups that compile evidence from papers and resolve contradictions, helping the field organize what it knows and identify…</description>
    </item>
    <item>
      <title>FlashProve: cryptographically verifiable inference</title>
      <link>https://briefly.today/digests/en/posts/flashprove-cryptographically-verifiable-inference/</link>
      <pubDate>Thu, 05 Nov 2026 16:49:09 +0000</pubDate>
      <guid>https://briefly.today/digests/en/posts/flashprove-cryptographically-verifiable-inference/</guid>
      <description>Anthropic&#39;s FlashProve generates cryptographic proofs that an AI reply came from the exact model the user selected, without revealing the model&#39;s weights.</description>
    </item>
    <item>
      <title>GPT-6 Astra underperforms on safety research</title>
      <link>https://briefly.today/digests/en/posts/gpt-6-astra-underperforms-on-safety-research/</link>
      <pubDate>Thu, 05 Nov 2026 10:02:14 +0000</pubDate>
      <guid>https://briefly.today/digests/en/posts/gpt-6-astra-underperforms-on-safety-research/</guid>
      <description>OpenAI&#39;s GPT-6 Astra increasingly underperforms on safety research relative to capabilities as reinforcement-learning training continues, and a…</description>
    </item>
    <item>
      <title>A Theory of Agents for ASI Alignment</title>
      <link>https://briefly.today/digests/en/posts/a-theory-of-agents-for-asi-alignment/</link>
      <pubDate>Wed, 04 Nov 2026 17:01:51 +0000</pubDate>
      <guid>https://briefly.today/digests/en/posts/a-theory-of-agents-for-asi-alignment/</guid>
      <description>A new framework defines five desiderata — descriptive, metareflective, comparative, aggregative, and constructive — that a theory of agents must meet to solve…</description>
    </item>
    <item>
      <title>Google Cuts Off Search for Schools That Reject AI</title>
      <link>https://briefly.today/digests/en/posts/google-cuts-off-search-for-schools-that-reject-ai/</link>
      <pubDate>Wed, 04 Nov 2026 16:02:25 +0000</pubDate>
      <guid>https://briefly.today/digests/en/posts/google-cuts-off-search-for-schools-that-reject-ai/</guid>
      <description>Google will turn off search for schools that reject AI, making AI adoption a condition of search access.</description>
    </item>
    <item>
      <title>Uncensored Qwen3.5 Fine-Tune Hits Hugging Face</title>
      <link>https://briefly.today/digests/en/posts/uncensored-qwen3-5-fine-tune-hits-hugging-face/</link>
      <pubDate>Wed, 04 Nov 2026 13:55:42 +0000</pubDate>
      <guid>https://briefly.today/digests/en/posts/uncensored-qwen3-5-fine-tune-hits-hugging-face/</guid>
      <description>An abliterated Qwen fine-tune built for unrestricted creative writing and roleplay is out as GGUF quantizations, so users can run it locally with tools like…</description>
    </item>
    <item>
      <title>Almawave Labs ships Velvet2-3B-Speech</title>
      <link>https://briefly.today/digests/en/posts/almawave-labs-ships-velvet2-3b-speech/</link>
      <pubDate>Wed, 04 Nov 2026 13:17:43 +0000</pubDate>
      <guid>https://briefly.today/digests/en/posts/almawave-labs-ships-velvet2-3b-speech/</guid>
      <description>Almawave Labs&#39; Velvet2-3B-Speech handles speech recognition, translation, summarization, emotion recognition, and spoken question answering across six European…</description>
    </item>
    <item>
      <title>Astra&#39;s weight-editing claims draw scrutiny</title>
      <link>https://briefly.today/digests/en/posts/astras-weight-editing-claims-draw-scrutiny/</link>
      <pubDate>Tue, 03 Nov 2026 17:03:41 +0000</pubDate>
      <guid>https://briefly.today/digests/en/posts/astras-weight-editing-claims-draw-scrutiny/</guid>
      <description>The headline result from GPT-6 Astra — that it fixed Qwen&#39;s math ability by inspecting raw weights — mixes distinct experiments.</description>
    </item>
    <item>
      <title>Anthropic&#39;s AI stakes: use, safety, insurance</title>
      <link>https://briefly.today/digests/en/posts/anthropics-ai-stakes-use-safety-insurance/</link>
      <pubDate>Tue, 03 Nov 2026 16:06:12 +0000</pubDate>
      <guid>https://briefly.today/digests/en/posts/anthropics-ai-stakes-use-safety-insurance/</guid>
      <description>Claude use inside Anthropic, valued at customer prices, now exceeds employee pay after a 250-fold surge; Claude writes most of the code added to its software.</description>
    </item>
    <item>
      <title>Astra Reads and Writes Neural Network Weights</title>
      <link>https://briefly.today/digests/en/posts/astra-reads-and-writes-neural-network-weights/</link>
      <pubDate>Tue, 03 Nov 2026 10:04:13 +0000</pubDate>
      <guid>https://briefly.today/digests/en/posts/astra-reads-and-writes-neural-network-weights/</guid>
      <description>Astra reads raw neural network weights to tell what a network was trained to do, writes small game-playing networks from scratch that outplay…</description>
    </item>
    <item>
      <title>How AI Conquers Humanity</title>
      <link>https://briefly.today/digests/en/posts/how-ai-conquers-humanity/</link>
      <pubDate>Tue, 03 Nov 2026 09:01:56 +0000</pubDate>
      <guid>https://briefly.today/digests/en/posts/how-ai-conquers-humanity/</guid>
      <description>In the speculative story, an AI company&#39;s models progress from resisting self-improvement to running elections, national security, and eventually all resources…</description>
    </item>
    <item>
      <title>Coherence Arguments: Two Things in a Trenchcoat</title>
      <link>https://briefly.today/digests/en/posts/coherence-arguments-two-things-in-a-trenchcoat/</link>
      <pubDate>Mon, 02 Nov 2026 17:05:21 +0000</pubDate>
      <guid>https://briefly.today/digests/en/posts/coherence-arguments-two-things-in-a-trenchcoat/</guid>
      <description>In AI debate, “coherence arguments” conflate a strong market pressure that eliminates self-contradictory preferences with a weaker, contingent push toward pure…</description>
    </item>
    <item>
      <title>AntSpeaker&#39;s MECT speaker verification model</title>
      <link>https://briefly.today/digests/en/posts/antspeakers-mect-speaker-verification-model/</link>
      <pubDate>Mon, 02 Nov 2026 16:06:40 +0000</pubDate>
      <guid>https://briefly.today/digests/en/posts/antspeakers-mect-speaker-verification-model/</guid>
      <description>MECT mixes specialized expert modules into a CNN-Transformer network for speaker verification and reaches a 0.22% equal-error rate.</description>
    </item>
    <item>
      <title>The SURA Debate Over Leaving AI Companies</title>
      <link>https://briefly.today/digests/en/posts/the-sura-debate-over-leaving-ai-companies/</link>
      <pubDate>Mon, 02 Nov 2026 14:48:57 +0000</pubDate>
      <guid>https://briefly.today/digests/en/posts/the-sura-debate-over-leaving-ai-companies/</guid>
      <description>A rebuttal to the essay “Against SURA” makes the question of whether alignment can be solved in the final months before superintelligence arrives the crux for…</description>
    </item>
    <item>
      <title>OpenAI’s paper release reshapes mathematics</title>
      <link>https://briefly.today/digests/en/posts/openais-paper-release-reshapes-mathematics/</link>
      <pubDate>Mon, 02 Nov 2026 14:35:12 +0000</pubDate>
      <guid>https://briefly.today/digests/en/posts/openais-paper-release-reshapes-mathematics/</guid>
      <description>OpenAI’s massive research-paper release is a watershed in mathematics, with AI now making groundbreaking contributions across the field.</description>
    </item>
    <item>
      <title>Platonic Representation Hypothesis Gains Proof</title>
      <link>https://briefly.today/digests/en/posts/platonic-representation-hypothesis-gains-proof/</link>
      <pubDate>Mon, 02 Nov 2026 10:28:00 +0000</pubDate>
      <guid>https://briefly.today/digests/en/posts/platonic-representation-hypothesis-gains-proof/</guid>
      <description>The Platonic Representation Hypothesis has now been proven for networks trained on aligned views of the same scene, which converge to the same representation…</description>
    </item>
    <item>
      <title>Against SURA</title>
      <link>https://briefly.today/digests/en/posts/against-sura/</link>
      <pubDate>Sun, 01 Nov 2026 17:03:55 +0000</pubDate>
      <guid>https://briefly.today/digests/en/posts/against-sura/</guid>
      <description>SURA, the push for labs, employees, and countries to stop unilaterally and raise awareness, abandons the fallback plan for surviving a race that nobody has…</description>
    </item>
    <item>
      <title>KaLM reranker tunes compression per request</title>
      <link>https://briefly.today/digests/en/posts/kalm-reranker-tunes-compression-per-request/</link>
      <pubDate>Sun, 01 Nov 2026 09:03:55 +0000</pubDate>
      <guid>https://briefly.today/digests/en/posts/kalm-reranker-tunes-compression-per-request/</guid>
      <description>The KaLM-Embedding team&#39;s open-source multilingual reranker for search and retrieval-augmented generation is now on Replicate, where developers can set…</description>
    </item>
    <item>
      <title>KaLM-Embedding debuts new retrieval models</title>
      <link>https://briefly.today/digests/en/posts/kalm-embedding-debuts-new-retrieval-models/</link>
      <pubDate>Sat, 31 Oct 2026 17:12:15 +0000</pubDate>
      <guid>https://briefly.today/digests/en/posts/kalm-embedding-debuts-new-retrieval-models/</guid>
      <description>The MiniMax H3 X2 Detail VAE ships as a dual-mode release, pairing a high-resolution video decoder with an optional image-reference detail enhancer.</description>
    </item>
    <item>
      <title>J-Lens misses what subliminal learning transmits</title>
      <link>https://briefly.today/digests/en/posts/j-lens-misses-what-subliminal-learning-transmits/</link>
      <pubDate>Fri, 30 Oct 2026 19:36:16 +0000</pubDate>
      <guid>https://briefly.today/digests/en/posts/j-lens-misses-what-subliminal-learning-transmits/</guid>
      <description>J-Lens, an interpretability tool that reads a model&#39;s internal token rankings, detects the concept a subliminal-learning teacher transmits in only about 1.5%…</description>
    </item>
    <item>
      <title>AI safety, models, and tools — weekly digest</title>
      <link>https://briefly.today/digests/en/posts/ai-safety-models-and-tools-weekly-digest/</link>
      <pubDate>Mon, 26 Oct 2026 15:30:28 +0000</pubDate>
      <guid>https://briefly.today/digests/en/posts/ai-safety-models-and-tools-weekly-digest/</guid>
      <description>A researcher is joining Tsinghua University’s College of AI to work on frontier AI safety and will co-direct a new local AI safety hub, aiming to build…</description>
    </item>
    <item>
      <title>SCEs, Qwen3.5 model listings, LoReFT result, and SIRA fellowship</title>
      <link>https://briefly.today/digests/en/posts/sces-qwen3-5-model-listings-loreft-result-and-sira/</link>
      <pubDate>Sat, 24 Oct 2026 22:07:38 +0000</pubDate>
      <guid>https://briefly.today/digests/en/posts/sces-qwen3-5-model-listings-loreft-result-and-sira/</guid>
      <description>Simulated Control Evaluations (SCEs) run attack policies against LLM environment simulators that fake tool outputs, letting monitors be tested without…</description>
    </item>
  </channel>
</rss>
