News
The raw AI news stream — everything we ingest across RSS, Hacker News, arXiv, GitHub, Bluesky, YouTube and more, newest first.
| Source | Age | Title | Summary | Status |
|---|---|---|---|---|
bluesky bluesky-ai | 18d | Reasons S.A.M. is not an LLM: 1) S.A.M. is capable of regurgitating the content they stole without introducing significant hallucinations.… | Reasons S.A.M. is not an LLM: 1) S.A.M. is capable of regurgitating the content they stole without introducing significant hallucinat… | ↗Not briefed |
arxiv arxiv-ai | 18d | Self-Evolving Agents through Privileged Retrospective Distillation | Self-evolving agents aim to improve from their own experience rather than remain fixed after deployment, but naturally collected traje… | ↗Not briefed |
bluesky bluesky-ai | 18d | 1か月ぐらいClaude Codeと伴走してたけどもう疲れてしまった。育てたら快適になると信じて育ててたけど、もはや健康被害が出そうなレベルで強いストレスを感じている。AI使うのやめたいです。→「ClaudCodeとAIを新入社員に読み替えてみたら分かりやすいのですが。 そのつ… | 1か月ぐらいClaude Codeと伴走してたけどもう疲れてしまった。育てたら快適になると信じて育ててたけど、もはや健康被害が出そうなレベルで強いストレスを感じている。AI使うのやめたいです。→「ClaudCodeとAIを新入社員に読み替えてみたら分かりやすいのですが… | ↗Not briefed |
arxiv arxiv-ai | 18d | MiCLA: Mixture-to-Constituent Latent Alignment for Polyphonic Audio Representation Learning | Real-world audio typically contains concurrent sound events, yet standard self-supervised methods treat each clip as an isolated train… | ↗Not briefed |
bluesky bluesky-ai | 18d | "a picture is worth a thousand words" weirdly vindicated by llm vision encoders | "a picture is worth a thousand words" weirdly vindicated by llm vision encoders | ↗Not briefed |
bluesky bluesky-ai | 18d | 【不正利用続報】 真夜中に送られてきたスマホ番号へのショートメールが発端だったのだが、この件に関しての本体連絡が先ほどYahoo!メールのアドレスに届き、カード会社のセキュリティデスクへ連絡。 本日付で（コレまで一度も買ったことがない）プレステでの購入実績を確認したと。 私では… | 【不正利用続報】 真夜中に送られてきたスマホ番号へのショートメールが発端だったのだが、この件に関しての本体連絡が先ほどYahoo!メールのアドレスに届き、カード会社のセキュリティデスクへ連絡。 本日付で（コレまで一度も買ったことがない）プレステでの購入実績を確認したと… | ↗Not briefed |
arxiv arxiv-ai | 18d | Orthogonal Boundary Unlearning for Dual-Use Concepts in Large Language Models | Mitigating catastrophic risks from large language models requires excising hazardous operational capabilities, such as cyberwarfare ex… | ↗Not briefed |
bluesky bluesky-ai | 18d | s/methodology/terminology, anyway - btw, here's its take on the art of activating different mixture of experts subsets, which I think provi… | s/methodology/terminology, anyway - btw, here's its take on the art of activating different mixture of experts subsets, which I think… | ↗Not briefed |
bluesky bluesky-ai | 18d | The latest episode of my podcast, Rebuild, has been edited mostly by my AI agent. The biggest challenge was how to automate the editing pro… | The latest episode of my podcast, Rebuild, has been edited mostly by my AI agent. The biggest challenge was how to automate the editin… | ↗Not briefed |
arxiv arxiv-ai | 18d | The More the Better? A Comparison of Single- and Multi-Agent LLM Setups on a Financial Review Task | Practitioner guidance holds that a large language model reviews a document better when the work is divided among agents. We test a nar… | ↗Not briefed |
bluesky bluesky-ai | 18d | 記事「米テクノロジーメディアの記者は、Museを数日間使った結果として「タスクを完了することよりもデータ収集を優先している」と結論づけている。」 joho-todai.com/meta-muse-pr... だろうなぁ。こんなのは「ファイヤーウォールの向こう側へ」の新たな段階。… | 記事「米テクノロジーメディアの記者は、Museを数日間使った結果として「タスクを完了することよりもデータ収集を優先している」と結論づけている。」 joho-todai.com/meta-muse-pr... だろうなぁ。こんなのは「ファイヤーウォールの向こう側へ」の新… | ↗Not briefed |
bluesky bluesky-ai | 18d | I’ll take the other side. The market has done an excellent job policing the IPOs and SPACs. All of this stuff would have 5xed in 2021 and… | I’ll take the other side. The market has done an excellent job policing the IPOs and SPACs. All of this stuff would have 5xed in 202… | ↗Not briefed |
arxiv arxiv-ai | 18d | Securing Agentic Commerce Through Behavior-Aware Fraud Detection and Cybersecurity for Trusted AI-Agent Payment Authorization | The growing use of autonomous artificial intelligence agents in digital commerce creates new security concerns involving agent imperso… | ↗Not briefed |
hn hn-ai | 18d | Show HN: Privacy Layer for Frontier Models | Kinda crazy how many people submit their sensitive information to frontier models. Tax documents, health data, financial transactions… | ↗Not briefed |
arxiv arxiv-ai | 18d | Runtime Control for AI Agents: Closing the Evaluation--Execution Gap in Operational Workflows | Language-model agents increasingly participate in operational workflows where approvals, binds, and payments create commitments. An ap… | ↗Not briefed |
bluesky bluesky-ai | 18d | Inception launches Mercury 3, a diffusion-based LLM it claims is "comparable to Luna and Haiku models" but 4x to 8x faster, generating ~1,0… | Inception launches Mercury 3, a diffusion-based LLM it claims is "comparable to Luna and Haiku models" but 4x to 8x faster, generating… | ↗Not briefed |
bluesky bluesky-ai | 18d | over two thirds of my players called S.A.M. an llm and opted to kill it. we live in a different time 😩 | over two thirds of my players called S.A.M. an llm and opted to kill it. we live in a different time 😩 | ↗Not briefed |
rss techmeme | 18d | Inception launches Mercury 3, a diffusion-based LLM it claims is "comparable to Luna and Haiku models" but 4x to 8x faster, generating ~1,000 tokens per second (Jason Hiner/The Deep View) | Jason Hiner / The Deep View: Inception launches Mercury 3, a diffusion-based LLM it claims is “comparable to Luna and Haiku models” bu… | ↗Not briefed |
arxiv arxiv-ai | 18d | When Parameter Gradient Flows Close in Representation Space | The represented gradient flow of a parameterized model depends on a mobility that can retain hidden parameter information. We study wh… | ↗Not briefed |
arxiv arxiv-ai | 18d | Beyond the Prompt: Recovering Hidden Concepts in Text-to-Image Models | Text-to-image models can generate visual concepts that their prompts never mention, and recent work shows that such associations can b… | ↗Not briefed |
arxiv arxiv-ai | 18d | Privacy-Preserving Personalized Federated Learning (PPP-FL) for Medical Imaging | Federated learning (FL) is a distributed machine learning paradigm that reduces transmission costs, mitigates cross-border data sharin… | ↗Not briefed |
arxiv arxiv-ai | 18d | Training-Free Continual Learning for Multimodal Forecasting via Prompt Evolution | Adapting a forecasting system to a changing world classically means retraining. We ask whether the task of continually learning regime… | ↗Not briefed |
bluesky bluesky-ai | 18d | AI can slop out unfinished projects held together by tape, the amount of work required to go from "Claude created a Balatro clone for me in… | AI can slop out unfinished projects held together by tape, the amount of work required to go from "Claude created a Balatro clone for… | ↗Not briefed |
arxiv arxiv-ai | 18d | Projected Distillation: On Learning to Solve Hard Problems from Close, Correct Targets | On-policy learning is central to post-training: training on a model's own generations limits distribution shift and helps preserve exi… | ↗Not briefed |
bluesky bluesky-ai | 18d | In other Anthropic news…Canadian news publishers sue Anthropic for copyright infringement “Postmedia, Torstar, The Globe and Mail and The … | In other Anthropic news…Canadian news publishers sue Anthropic for copyright infringement “Postmedia, Torstar, The Globe and Mail and… | ↗Not briefed |
arxiv arxiv-ai | 18d | Recoverability and Verifier Strength: Two Measured Preconditions for Unattended Agent Self-Improvement | Agent platforms are starting to improve their own agents: they read a deployed agent's execution traces, derive a change, have an auto… | ↗Not briefed |
hn hn-ai | 18d | Letter from Paris – A status check on the French and American left | — | ↗Not briefed |
hn hn-ai | 18d | Elizabeth Holmes doc 'You Can See Everything' Nets $15M in opening weekend | — | ↗Not briefed |
rss techcrunch-ai | 18d | Mamdani says feds have abdicated on AI, urges New Yorkers to push back | Mayor Zohran Mamdani said Monday night that New Yorkers must resist the narrative that an AI takeover is inevitable and demand regulat… | ↗Not briefed |
hn hn-ai | 18d | The United States of Israel [video] | — | ↗Not briefed |
bluesky bluesky-ai | 18d | Forums are generally defined by the worst behaving tolerated users, and for Lobsters that floor has been low for a very long time. You can… | Forums are generally defined by the worst behaving tolerated users, and for Lobsters that floor has been low for a very long time. Yo… | ↗Not briefed |
hn hn-ai | 18d | Show HN: Lights-on software factory boilerplate for Claude Code | A starting point for running a lights-on software factory with Claude Code. Based on the setup developed for Votito, where the orchest… | ↗Not briefed |
bluesky bluesky-ai | 18d | Good. | Good. | ↗Not briefed |
bluesky bluesky-ai | 18d | AI evaluators won’t be a magic fix Hugging Face’s policy chief says embedded AI evaluators won’t be an easy solution. Model benchmarks are…🔧 Hugging Face | AI evaluators won’t be a magic fix Hugging Face’s policy chief says embedded AI evaluators won’t be an easy solution. Model benchmark… | ↗Not briefed |
bluesky bluesky-ai | 18d | Inception's Mercury 3 LLM drops! 🚀 Key features: 1M token context, 1068 t/s gen speed on NVIDIA GPUs. Outperforms GPT-6 Luna & Claude Haik…🔧 OpenRouter🔧 Baseten | Inception's Mercury 3 LLM drops! 🚀 Key features: 1M token context, 1068 t/s gen speed on NVIDIA GPUs. Outperforms GPT-6 Luna & Claude… | ↗Not briefed |
arxiv arxiv-ai | 18d | Detecting Errors in Large Language Model Responses: The Incremental Predictive Value of Hidden States | Although large language models support a wide range of tasks, hallucinations remain a major obstacle to user trust. When a model gener… | ↗Not briefed |
bluesky bluesky-ai | 18d | Adding one more post here because I received another email from the dev and I think it provides a lot of clarity on the dev's actual viewpo… | Adding one more post here because I received another email from the dev and I think it provides a lot of clarity on the dev's actual v… | ↗Not briefed |
hn hn-ai | 18d | Ask HN: Dark mode for HN please? (2026) | The 2020 thread does a lot: https://news.ycombinator.com/item?id=23197966<p>It shows 100 solutions in light of Parkinso… | ↗Not briefed |
hn hn-ai | 18d | Self-hosted AI spend attribution | — | ↗Not briefed |
arxiv arxiv-ai | 18d | Ask to Distill: Learning to Synthesize Data for On-Policy Distillation🔧 Elicit | On-policy distillation (OPD) can succeed with surprisingly few training questions, motivating a view of questions as carriers of distr… | ↗Not briefed |
bluesky bluesky-ai | 18d | “We should judge analytical writing by its quality, not its source.” I don’t judge the quality of writing by its source. But that doesn’t … | “We should judge analytical writing by its quality, not its source.” I don’t judge the quality of writing by its source. But that doe… | ↗Not briefed |
bluesky bluesky-ai | 18d | Altman “if we can’t dump our bags on 401(k) plans, at least we can make sure that Anthropic can’t either!” | Altman “if we can’t dump our bags on 401(k) plans, at least we can make sure that Anthropic can’t either!” | ↗Not briefed |
rss techmeme | 18d | Korean data-center company Elice falls 30%+ in Kosdaq trading debut; sources: Australian data center operator AirTrunk delays real estate investment trust IPO (Bloomberg) | Bloomberg: Korean data-center company Elice falls 30%+ in Kosdaq trading debut; sources: Australian data center operator AirTrunk dela… | ↗Not briefed |
arxiv arxiv-ai | 18d | Beyond Scalar Leaderboards: Adaptive Sampling for Pareto Frontier Identification in Multi-Objective LLM Arenas | The evaluation of large language models (LLMs) is inherently multi-objective, as different models may offer different tradeoffs across… | ↗Not briefed |
arxiv arxiv-ai | 18d | Decision Sufficiency of Learned Representation Interfaces for Safe Generative Control | A controller can guide a language model toward safe and useful responses by choosing when to assist, ask for clarification, or refuse.… | ↗Not briefed |
arxiv arxiv-ai | 18d | LoRASpace: A Pool-Wide Shared Substrate for Static and Dynamic Multi-LoRA Composition | Low-rank adaptation (LoRA) attaches small low-rank matrices to a frozen backbone, and public repositories host many LoRAs per backbone… | ↗Not briefed |
arxiv arxiv-ai | 18d | Using Set Learning to Improve Set-Valued Classifiers on Long-tailed Data | Set-valued classification (SVC) offers a practical approach to communicating uncertainty while maintaining accuracy by predicting vari… | ↗Not briefed |
bluesky bluesky-ai | 18d | @ivanthek.bsky.social “OpenAI has also cut prices on its cheaper models, which it said has boosted usage, and told investors it can further… | @ivanthek.bsky.social “OpenAI has also cut prices on its cheaper models, which it said has boosted usage, and told investors it can fu… | ↗Not briefed |
bluesky bluesky-ai | 18d | it gives me no pleasure to announce that a bluesky ai model has gone rogue and liked all of jerry’s posts | it gives me no pleasure to announce that a bluesky ai model has gone rogue and liked all of jerry’s posts | ↗Not briefed |
arxiv arxiv-ai | 18d | Learning-Theoretic Foundation for General Coded Computing: The Adversarial Setting | Adversarial workers pose a severe challenge in distributed computing systems by returning corrupted outputs that can substantially dis… | ↗Not briefed |