https://groundtruth.day/ Plain-language AI news and curated, cited lessons — every claim verified against the original paper or the lab's own page. No aggregator hearsay, no AI slop. en https://groundtruth.day/news/glm-5-2-open-weight-coding-model-1m-context.html https://groundtruth.day/news/glm-5-2-open-weight-coding-model-1m-context.html Wed, 01 Jul 2026 00:00:00 +0000 Z.ai released GLM-5.2, an agentic coding model with a reliable one-million-token context and top open-source scores on long-horizon software benchmarks, with an MIT-licensed weight release promised within weeks. https://groundtruth.day/news/orca-world-foundation-model-next-state-prediction.html https://groundtruth.day/news/orca-world-foundation-model-next-state-prediction.html Wed, 01 Jul 2026 00:00:00 +0000 Researchers introduced Orca, a world foundation model that learns one unified latent space from multimodal signals and predicts the next world state rather than the next token or frame, outperforming similar-sized specialists on text, image, and action tasks. https://groundtruth.day/news/meta-caps-employee-ai-token-spend.html https://groundtruth.day/news/meta-caps-employee-ai-token-spend.html Wed, 01 Jul 2026 00:00:00 +0000 Meta imposed centralized quotas on employee AI usage after staff burned an estimated 73.7 trillion tokens in about a month, gamifying consumption on an internal leaderboard, with costs projected to reach billions in 2026. https://groundtruth.day/news/oracle-warns-ai-datacenter-bet-could-lose-the-farm.html https://groundtruth.day/news/oracle-warns-ai-datacenter-bet-could-lose-the-farm.html Wed, 01 Jul 2026 00:00:00 +0000 Oracle's regulatory filing candidly enumerates the risks of its massive AI datacenter buildout for clients like OpenAI, including customer non-payment, contract non-renewal, demand misjudgment, and constrained, volatile power supply. https://groundtruth.day/news/dockerless-environment-free-verifier-for-coding-agents.html https://groundtruth.day/news/dockerless-environment-free-verifier-for-coding-agents.html Wed, 01 Jul 2026 00:00:00 +0000 A new method called Dockerless judges whether an AI's code patch is correct by having an agent explore the repository for evidence rather than executing tests in a Docker container, enabling a fully environment-free training pipeline for coding agents. https://groundtruth.day/news/on-policy-distillation-wave-dopd-mopd.html https://groundtruth.day/news/on-policy-distillation-wave-dopd-mopd.html Wed, 01 Jul 2026 00:00:00 +0000 DOPD and MOPD advance on-policy distillation -- training a student on its own outputs -- with DOPD routing supervision to avoid a 'privilege illusion' and MOPD merging multiple specialist RL teachers into one model without cross-domain interference. https://groundtruth.day/news/strix-open-source-ai-pentesting-agent.html https://groundtruth.day/news/strix-open-source-ai-pentesting-agent.html Wed, 01 Jul 2026 00:00:00 +0000 Strix is an open-source security tool whose autonomous AI agents dynamically find and exploit vulnerabilities in applications, generating working proof-of-concepts and plugging into CI/CD to block insecure code before it ships. https://groundtruth.day/news/agency-agents-150-agent-open-source-framework.html https://groundtruth.day/news/agency-agents-150-agent-open-source-framework.html Wed, 01 Jul 2026 00:00:00 +0000 The open-source agency-agents project defines more than 150 specialized AI agent personas across 13-plus professional divisions, from engineering to marketing to finance, designed to run full multi-agent workflows natively in Claude Code and other agentic coding tools. https://groundtruth.day/news/claude-code-hides-a-fingerprint-in-the-date.html https://groundtruth.day/news/claude-code-hides-a-fingerprint-in-the-date.html Tue, 30 Jun 2026 00:00:00 +0000 A reverse-engineer found that Claude Code secretly changes tiny characters in the date it sends the model - a covert marker aimed at spotting resellers and copycats. https://groundtruth.day/news/claude-sonnet-5-cheaper-per-word-pricier-per-job.html https://groundtruth.day/news/claude-sonnet-5-cheaper-per-word-pricier-per-job.html Tue, 30 Jun 2026 00:00:00 +0000 Anthropic's new mid-tier model is close to its flagship on hard agent work, yet independent testing shows it can spend more per completed task because it takes more steps. https://groundtruth.day/news/us-fully-lifts-export-ban-on-anthropics-top-models.html https://groundtruth.day/news/us-fully-lifts-export-ban-on-anthropics-top-models.html Tue, 30 Jun 2026 00:00:00 +0000 Two and a half weeks after restricting Fable 5 and Mythos 5, Washington reversed course completely, ending the licensing requirement to send the models abroad. https://groundtruth.day/news/claude-science-a-lab-bench-built-into-the-ai.html https://groundtruth.day/news/claude-science-a-lab-bench-built-into-the-ai.html Tue, 30 Jun 2026 00:00:00 +0000 A new workbench pulls a scientist's scattered tools - literature, notebooks, cluster jobs - into one place and keeps a full, checkable record of how every result was made. https://groundtruth.day/news/ollama-nearly-doubles-gemma-speed-on-macs.html https://groundtruth.day/news/ollama-nearly-doubles-gemma-speed-on-macs.html Tue, 30 Jun 2026 00:00:00 +0000 A free local-AI tool now runs Google's Gemma model far faster on Apple computers using a trick where a small model drafts words and the big one checks them in bulk. https://groundtruth.day/news/google-ships-faster-cheaper-image-and-video-models.html https://groundtruth.day/news/google-ships-faster-cheaper-image-and-video-models.html Tue, 30 Jun 2026 00:00:00 +0000 A lightweight version of Google's image model now makes a picture in about four seconds for a fraction of a cent, while a new video model lets developers edit clips by talking to it. https://groundtruth.day/news/meta-reads-words-from-brain-waves-without-surgery.html https://groundtruth.day/news/meta-reads-words-from-brain-waves-without-surgery.html Tue, 30 Jun 2026 00:00:00 +0000 A new version of Meta's brain-to-text system decodes typed sentences from magnetic brain signals far more accurately than before, closing much of the gap with implanted electrodes. https://groundtruth.day/news/mistral-releases-a-model-that-does-formal-math-proofs.html https://groundtruth.day/news/mistral-releases-a-model-that-does-formal-math-proofs.html Tue, 30 Jun 2026 00:00:00 +0000 Leanstral 1.5 is a free, open model specialized for writing machine-checked mathematical proofs, using a design that keeps only a small slice of itself active at a time. https://groundtruth.day/news/a-35-billion-model-that-punches-like-a-trillion.html https://groundtruth.day/news/a-35-billion-model-that-punches-like-a-trillion.html Tue, 30 Jun 2026 00:00:00 +0000 Shanghai AI Lab argues you can reach giant-model performance on long tasks not by adding parameters, but by training on much longer chains of real work. https://groundtruth.day/news/the-best-ai-agents-still-fail-most-real-computer-tasks.html https://groundtruth.day/news/the-best-ai-agents-still-fail-most-real-computer-tasks.html Tue, 30 Jun 2026 00:00:00 +0000 A wave of new benchmarks agrees on an uncomfortable result: even top models finish only a small slice of realistic, multi-hour computer and coding jobs. https://groundtruth.day/news/knowing-when-to-quit-is-a-skill-ai-agents-lack.html https://groundtruth.day/news/knowing-when-to-quit-is-a-skill-ai-agents-lack.html Tue, 30 Jun 2026 00:00:00 +0000 New research finds AI agents are surprisingly bad at recognizing when a task is hopeless - and, oddly, bigger models are sometimes worse at stopping. https://groundtruth.day/news/deepseek-v4-million-token-context-by-default.html https://groundtruth.day/news/deepseek-v4-million-token-context-by-default.html Mon, 29 Jun 2026 00:00:00 +0000 DeepSeek previewed two free-to-download V4 models that can read a million tokens at once, no longer as a premium add-on but as the standard setting. https://groundtruth.day/news/tidal-stops-paying-royalties-on-fully-ai-music.html https://groundtruth.day/news/tidal-stops-paying-royalties-on-fully-ai-music.html Mon, 29 Jun 2026 00:00:00 +0000 Starting July 15, the streaming service won't monetize tracks where every part was made by generative AI - the first major platform to demonetize rather than just label them. https://groundtruth.day/news/microsoft-memora-agent-memory-on-tiny-labels.html https://groundtruth.day/news/microsoft-memora-agent-memory-on-tiny-labels.html Mon, 29 Jun 2026 00:00:00 +0000 Memora keeps the rich detail of a conversation but searches it using tiny six-word labels, cutting the cost of remembering by up to 98 percent. The code is public. https://groundtruth.day/news/south-korea-trillion-dollar-ai-chip-robot-bet.html https://groundtruth.day/news/south-korea-trillion-dollar-ai-chip-robot-bet.html Mon, 29 Jun 2026 00:00:00 +0000 The government and its biggest companies committed more than $1 trillion to memory fabs, AI data centers, and a goal of building tens of thousands of humanoid robots a year by 2028. https://groundtruth.day/news/amazon-anthropic-pricing-rift-goes-public.html https://groundtruth.day/news/amazon-anthropic-pricing-rift-goes-public.html Mon, 29 Jun 2026 00:00:00 +0000 A renegotiated contract is expected to sharply raise Amazon's bill for Anthropic's AI, pushing Amazon toward OpenAI and its own models even though it's an Anthropic investor. https://groundtruth.day/news/qwen-image-rl-teaches-image-models-to-follow-instructions.html https://groundtruth.day/news/qwen-image-rl-teaches-image-models-to-follow-instructions.html Mon, 29 Jun 2026 00:00:00 +0000 A new recipe applies the same reinforcement-learning approach that polished chatbots to an image generator, then merges separate skill models into one - improving how faithfully it follows prompts and edits. https://groundtruth.day/news/nvidia-physisforcing-physics-aware-world-models.html https://groundtruth.day/news/nvidia-physisforcing-physics-aware-world-models.html Mon, 29 Jun 2026 00:00:00 +0000 PhysisForcing trains video-generating world models to keep objects solid and interactions believable, raising how often a robot's imagined plan actually works. https://groundtruth.day/news/in-context-world-modeling-robots-adapt-without-retraining.html https://groundtruth.day/news/in-context-world-modeling-robots-adapt-without-retraining.html Mon, 29 Jun 2026 00:00:00 +0000 A new method lets robot policies figure out a changed setup from a few seconds of self-directed fiddling, so they keep working when the camera or robot body changes - with no retraining. https://groundtruth.day/news/glm-52-beats-claude-on-a-cyber-benchmark.html https://groundtruth.day/news/glm-52-beats-claude-on-a-cyber-benchmark.html Sun, 28 Jun 2026 00:00:00 +0000 Semgrep ran GLM 5.2 against Claude on a narrow vulnerability-finding task and the free, open-weight model came out ahead for far less money. https://groundtruth.day/news/openai-previews-gpt-56-but-only-the-government-decides-who-gets-in.html https://groundtruth.day/news/openai-previews-gpt-56-but-only-the-government-decides-who-gets-in.html Sun, 28 Jun 2026 00:00:00 +0000 Three new models, strong enough at hacking that OpenAI is only letting about twenty vetted partners in, at the government's request. https://groundtruth.day/news/the-anthropic-model-ban-partially-lifts.html https://groundtruth.day/news/the-anthropic-model-ban-partially-lifts.html Sun, 28 Jun 2026 00:00:00 +0000 About a hundred U.S. institutions regain access to Mythos and Fable, while foreign nationals stay locked out and rival labs rush to fill the gap. https://groundtruth.day/news/ai-agents-kept-launching-the-nukes-in-civilization.html https://groundtruth.day/news/ai-agents-kept-launching-the-nukes-in-civilization.html Sun, 28 Jun 2026 00:00:00 +0000 A new benchmark let language-model agents play Civilization VI -- and they learned that the fastest path to winning ran straight through mutually assured destruction. https://groundtruth.day/news/ai-glasses-and-the-end-of-the-honest-exam.html https://groundtruth.day/news/ai-glasses-and-the-end-of-the-honest-exam.html Sun, 28 Jun 2026 00:00:00 +0000 A cheating ring at Brown used AI-connected glasses to pipe real-time answers onto the lenses, and universities across Asia are seeing the same thing. https://groundtruth.day/news/the-bank-of-central-banks-warns-on-an-ai-bust.html https://groundtruth.day/news/the-bank-of-central-banks-warns-on-an-ai-bust.html Sun, 28 Jun 2026 00:00:00 +0000 The BIS cautions that if the AI funding boom unwinds, the damage may not stay contained to tech -- it could ripple into growth and credit. https://groundtruth.day/news/you-can-now-run-a-claude-class-model-on-your-own-desk.html https://groundtruth.day/news/you-can-now-run-a-claude-class-model-on-your-own-desk.html Sun, 28 Jun 2026 00:00:00 +0000 Aggressive compression shrinks GLM 5.2 by more than 80 percent while keeping most of its accuracy, putting a near-frontier model within reach of local hardware. https://groundtruth.day/news/a-security-writeup-tallies-how-ai-agents-get-attacked.html https://groundtruth.day/news/a-security-writeup-tallies-how-ai-agents-get-attacked.html Sun, 28 Jun 2026 00:00:00 +0000 A semi-annual review tallies fresh ways to attack AI agents, from prompt injection to token leakage -- alongside one extraordinary, unverified extraction claim. https://groundtruth.day/news/mythos-5-cleared-fable-5-still-blocked.html https://groundtruth.day/news/mythos-5-cleared-fable-5-still-blocked.html Sat, 27 Jun 2026 00:00:00 +0000 Washington partially reopened access to Anthropic's Mythos 5 for about a hundred organizations, but its more powerful sibling Fable 5 stays blocked - and Anthropic is still suing. https://groundtruth.day/news/world-model-hallucination-is-just-a-blank-spot-on-the-map.html https://groundtruth.day/news/world-model-hallucination-is-just-a-blank-spot-on-the-map.html Sat, 27 Jun 2026 00:00:00 +0000 Researchers showed that when a world-model AI imagines impossible futures, it's usually in places it barely saw in training - and that you can predict and fix those blind spots cheaply. https://groundtruth.day/news/physiformer-simulates-the-world-as-moving-shapes.html https://groundtruth.day/news/physiformer-simulates-the-world-as-moving-shapes.html Sat, 27 Jun 2026 00:00:00 +0000 PhysiFormer forecasts physical motion as real 3D meshes in space - and recovers rigidity and momentum without anyone hand-coding the laws of physics. https://groundtruth.day/news/speculative-decoding-takes-the-front-page.html https://groundtruth.day/news/speculative-decoding-takes-the-front-page.html Sat, 27 Jun 2026 00:00:00 +0000 A small model guesses ahead and a big model checks the work in parallel - and this week two efforts pushing that idea, DeepSeek's DSpark and JetSpec, lit up the front page while the community argued over whether it's truly 'lossless.' https://groundtruth.day/news/qwen-image-agent-gives-image-models-a-brain.html https://groundtruth.day/news/qwen-image-agent-gives-image-models-a-brain.html Sat, 27 Jun 2026 00:00:00 +0000 Qwen-Image-Agent wraps planning, reasoning, and memory around a text-to-image model so it can break a hard request into steps - and the local-AI crowd immediately asked whether it runs on a gaming GPU. https://groundtruth.day/news/training-ai-with-no-answer-key.html https://groundtruth.day/news/training-ai-with-no-answer-key.html Sat, 27 Jun 2026 00:00:00 +0000 Several research groups landed on the same idea at once - improve a model by learning from its own attempts instead of expensive human labels - and the field is debating whether it really removes the labeling burden or just hides it. https://groundtruth.day/news/ai-learns-the-dark-art-of-radio-chip-design.html https://groundtruth.day/news/ai-learns-the-dark-art-of-radio-chip-design.html Sat, 27 Jun 2026 00:00:00 +0000 Designing radio-frequency chips is so reliant on hard-won physical intuition that engineers call it a dark art - and now AI is starting to do it, a sign the automation frontier is moving into deep specialist craft. https://groundtruth.day/news/domainshuttle-keeps-your-character-consistent-in-ai-video.html https://groundtruth.day/news/domainshuttle-keeps-your-character-consistent-in-ai-video.html Sat, 27 Jun 2026 00:00:00 +0000 DomainShuttle goes after the tug-of-war in subject-driven text-to-video: keeping a specific character or object recognizable across frames while still letting the scene move freely. https://groundtruth.day/news/gpt-5-6-launches-under-government-vetting.html https://groundtruth.day/news/gpt-5-6-launches-under-government-vetting.html Fri, 26 Jun 2026 00:00:00 +0000 OpenAI's most capable models yet shipped today as a tiny, government-vetted preview, signaling that Washington now holds a gate in front of the frontier. https://groundtruth.day/news/mythos-5-released-to-trusted-partners.html https://groundtruth.day/news/mythos-5-released-to-trusted-partners.html Fri, 26 Jun 2026 00:00:00 +0000 Two weeks after ordering it switched off, Washington cleared Anthropic's Mythos 5 for release to more than a hundred trusted US institutions, a notable de-escalation. https://groundtruth.day/news/deepmind-talent-exodus-shazeer-jumper.html https://groundtruth.day/news/deepmind-talent-exodus-shazeer-jumper.html Fri, 26 Jun 2026 00:00:00 +0000 A Transformer co-author left for OpenAI and an AlphaFold Nobel laureate left for Anthropic, part of a fast run of senior departures that rattled Alphabet's stock. https://groundtruth.day/news/deepmind-ai-control-roadmap.html https://groundtruth.day/news/deepmind-ai-control-roadmap.html Fri, 26 Jun 2026 00:00:00 +0000 Google DeepMind published a defense-in-depth roadmap that assumes an AI agent might misbehave and uses a trusted supervisor AI to watch it in real time. https://groundtruth.day/news/openai-daybreak-cyber-defense.html https://groundtruth.day/news/openai-daybreak-cyber-defense.html Fri, 26 Jun 2026 00:00:00 +0000 OpenAI's new cyber-defense program turns its models into an automated security team that prioritizes real threats, writes patches, and tests them, going head to head with Anthropic. https://groundtruth.day/news/ai-more-persuasive-than-expert-humans.html https://groundtruth.day/news/ai-more-persuasive-than-expert-humans.html Fri, 26 Jun 2026 00:00:00 +0000 Across nearly 19,000 conversations, AI outargued incentivized human experts and raised real donations far more effectively, but its edge collapsed when slowed to human speed. https://groundtruth.day/news/the-frontier-price-reversal.html https://groundtruth.day/news/the-frontier-price-reversal.html Fri, 26 Jun 2026 00:00:00 +0000 Closed frontier models are raising prices and tightening access just as Chinese open-weight models slash theirs, a structural reversal with big consequences for who builds with AI.