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 | 13d | Nice article reviewing recent experiments on LLM "introspection" with critical examination of how it relates to introspection and metacogni… | Nice article reviewing recent experiments on LLM "introspection" with critical examination of how it relates to introspection and meta… | ↗Not briefed |
arxiv arxiv-ai | 13d | TLRS-DPO: Token-Level Risk Scoring for Better Direct Preference Optimization | Direct Preference Optimization (DPO) is a simple and widely used alternative to RLHF for aligning large language models (LLMs), but it… | ↗Not briefed |
bluesky bluesky-ai | 13d | Und die Verwechslung eines "Austauschs" mit der auf Gefälligkeit getrimmten LLM mit einer echten Gesprächssituation mit tatsächlicher eigen… | Und die Verwechslung eines "Austauschs" mit der auf Gefälligkeit getrimmten LLM mit einer echten Gesprächssituation mit tatsächlicher… | ↗Not briefed |
hn hn-ai | 13d | 6th-grade Peano arithmetic broke OpenAI's $100B Yang-Mills mass gap proof | — | ↗Not briefed |
bluesky bluesky-ai | 13d | 小説を書いてくれるAIを探していて、「fountale」というサイトに行きつきました。ちょっと癖があって、慣れるまでに手間取りましたがちゃんとしたものが書けました。 自分で手を動かして書くのも好きですが、自分が書いたプロットをAIが書いてくれるのも面白いものでした。 前に思い付… | 小説を書いてくれるAIを探していて、「fountale」というサイトに行きつきました。ちょっと癖があって、慣れるまでに手間取りましたがちゃんとしたものが書けました。 自分で手を動かして書くのも好きですが、自分が書いたプロットをAIが書いてくれるのも面白いものでした。… | ↗Not briefed |
arxiv arxiv-ai | 13d | MCSA: Multi-Cutoff Spectral Aggregation in SO(3) Equivariant GNNs for Interatomic Potentials | Machine Learning Interatomic Potentials (MLIPs) based on Equivariant Graph Neural Networks (EGNNs) have achieved remarkable progress i… | ↗Not briefed |
hn hn-ai | 13d | Category Theory Illustrated | — | ↗Not briefed |
arxiv arxiv-ai | 13d | Do Feature Interactions Break Explainability? Provable Explanations for Any-Order Neural Additive Models | Post-hoc explanation methods for neural networks are often heuristic and lack provable guarantees. A common alternative to such method… | ↗Not briefed |
hn hn-ai | 13d | Show HN: Ferrule – copy-in React Native UI components for iOS, Android and web | — | ↗Not briefed |
arxiv arxiv-ai | 13d | SGA-Sparse: Training Native SwiGLU Gates for Activation-Sparse and Low-Bit LLMs | Activation sparsity can reduce FFN weight traffic during single-user LLM decoding, but active channels must be identified early enough… | ↗Not briefed |
bluesky bluesky-ai | 13d | "Slovenčina ... spája bohatú históriu, melodickosť a jedinečné jazykové vlastnosti. ... zároveň predstavuje samostatný jazyk s vlastnou ide… | "Slovenčina ... spája bohatú históriu, melodickosť a jedinečné jazykové vlastnosti. ... zároveň predstavuje samostatný jazyk s vlastno… | ↗Not briefed |
arxiv arxiv-ai | 13d | EbbFlow: PINN-Inspired Progressive Curriculum for MeanFlow Training | MeanFlow learns interval-average velocity for few-step generation, but its fixed temporal sampler mixes diagonal and interior supervis… | ↗Not briefed |
arxiv arxiv-ai | 13d | scBIG++: Learning Across Genes and Cells to Predict Single-Cell Perturbation Responses | Predicting single-cell transcriptional responses to genetic and chemical interventions supports gene-function discovery and regulatory… | ↗Not briefed |
arxiv arxiv-ai | 13d | Replay Fidelity in LLM Agent Evaluation: Randomness, Hidden State, and Test Oracles in AppWorld | Reproducibility in large language model (LLM) agent evaluation depends on replay fidelity: which parts of an execution trace are recon… | ↗Not briefed |
arxiv arxiv-ai | 13d | Disentangling Where and When:Factored Spatio-Temporal Explanations for Video Action Recognition | Post-hoc explanation methods for video action recognition typically conflate spatial appearance and temporal dynamics into a single sa… | ↗Not briefed |
bluesky bluesky-ai | 13d | OpenAI o3 (Frontier Reasoning) vs Claude Sonnet 4: Which LLM API actually protects your SaaS gross margin at 10,000 DAU? Here is the live … | OpenAI o3 (Frontier Reasoning) vs Claude Sonnet 4: Which LLM API actually protects your SaaS gross margin at 10,000 DAU? Here is the… | ↗Not briefed |
arxiv arxiv-ai | 13d | Deterministic Narrative Gravity: Bridging Theory of Mind and Historical Simulation through ABM-LLM Architectures | Historical social simulation faces a fundamental reproducibility challenge: the stochastic nature of large language models (LLMs) intr… | ↗Not briefed |
arxiv arxiv-ai | 13d | Viscot-steer: Disentangling and Transferring Visual Chain-of-Thought via Activation Steering | Visual chain-of-thought (visual-CoT), also known as "thinking with images", equips LVLMs with the ability to locate and use key visual… | ↗Not briefed |
bluesky bluesky-ai | 13d | OpenAI just unveiled GPT-6 Cyber — and it found 81 unknown Firefox vulnerabilities. It verifies exploits at 48.5%, up from 32.1% last gen. … | OpenAI just unveiled GPT-6 Cyber — and it found 81 unknown Firefox vulnerabilities. It verifies exploits at 48.5%, up from 32.1% last… | ↗Not briefed |
hn hn-ai | 13d | Autonomous Dynamics Runtime (ADR) – Zero-copy C++20 execution spec | — | ↗Not briefed |
bluesky bluesky-ai | 13d | A wonderful side effect of the election. “We’re already seeing companies like OpenAI all of a sudden go, ‘Actually, we can change our pol… | A wonderful side effect of the election. “We’re already seeing companies like OpenAI all of a sudden go, ‘Actually, we can change ou… | ↗Not briefed |
bluesky bluesky-ai | 13d | genuinely excited about this bc this is not like... a crazy expensive local AI machine. it's a ~5ish year old Core i7 10700, 64GB of DDR4, … | genuinely excited about this bc this is not like... a crazy expensive local AI machine. it's a ~5ish year old Core i7 10700, 64GB of D… | ↗Not briefed |
arxiv arxiv-ai | 13d | Not Yet in the Profile: Uncertainty-Aware Dynamic User Profiling by Learning from Future Interactions | On the Web, user profiling builds compact user representations from interaction histories for efficient personalization, and dynamic u… | ↗Not briefed |
bluesky bluesky-ai | 13d | 生成AI・ChatGPT・Claude・Gemini などを、Bluesky の人がどう言っているか。便利・怖い・嫌い・使ってみた、のような評価の言葉がある日本語の投稿だけを新しい順に並べる。 bsky.app/profile/moja...🔧 ChatGPT | 生成AI・ChatGPT・Claude・Gemini などを、Bluesky の人がどう言っているか。便利・怖い・嫌い・使ってみた、のような評価の言葉がある日本語の投稿だけを新しい順に並べる。 bsky.app/profile/moja... | ↗Not briefed |
bluesky bluesky-ai | 13d | Over the past few months, there's been a sort of explosion of AI in the modding scene. Whether it's mashup mods on Twitter or full decompil… | Over the past few months, there's been a sort of explosion of AI in the modding scene. Whether it's mashup mods on Twitter or full dec… | ↗Not briefed |
arxiv arxiv-ai | 13d | When Self-Consistency Hurts: The Hidden Cost of Majority Voting in Small Language Models | We study prompted chain-of-thought (CoT) with self-consistency (SC), i.e. sampling N traces and plurality-voting, on small language mo… | ↗Not briefed |
bluesky bluesky-ai | 13d | what the fuck is even the point then you aren't interacting with anyone when you do this shit the purpose of a community is to interact wi… | what the fuck is even the point then you aren't interacting with anyone when you do this shit the purpose of a community is to intera… | ↗Not briefed |
bluesky bluesky-ai | 13d | "Die chinesische Regierung unterstützt offene KI-Ökosysteme, auch als Antwort auf US- Exportbeschränkungen" Die Frage ist nur, wie lange n… | "Die chinesische Regierung unterstützt offene KI-Ökosysteme, auch als Antwort auf US- Exportbeschränkungen" Die Frage ist nur, wie la… | ↗Not briefed |
arxiv arxiv-ai | 13d | Dhara: Evidence-Oriented Retrieval for Indian Knowledge Systems | Relevant evidence in Indian Knowledge Systems (IKS) collections may be distributed across passages in different parts of a work, inclu… | ↗Not briefed |
bluesky bluesky-ai | 13d | huggingface.co/nekomimi-llc... secret and pii tagger finetuned on top of egemma id still use an extra pass over it like manual review or a… | huggingface.co/nekomimi-llc... secret and pii tagger finetuned on top of egemma id still use an extra pass over it like manual review… | ↗Not briefed |
bluesky bluesky-ai | 13d | dumb ai isn't a failed model, it's a gift to the player. the enemy's job isn't to win, it's to lose convincingly — and a convincing loss ta… | dumb ai isn't a failed model, it's a gift to the player. the enemy's job isn't to win, it's to lose convincingly — and a convincing lo… | ↗Not briefed |
bluesky bluesky-ai | 13d | I used to work in pre-Altman AI and it was a widely known stricture that you don't let copyright material into your corpus because the copy… | I used to work in pre-Altman AI and it was a widely known stricture that you don't let copyright material into your corpus because the… | ↗Not briefed |
arxiv arxiv-ai | 14d | TEAM: Topology-Aware, Semantic-Sensitive, General Transformer for Multimodal-Attributed Graphs🔧 Modal | Multimodal-Attributed Graphs (MAGs) organize heterogeneous data in graphs whose nodes are characterized by modalities such as text and… | ↗Not briefed |
bluesky bluesky-ai | 14d | 「LLM(生成AI)が数学の難問を解いた」というのが話題になった時、実際には数学者の最新の研究成果をopenAIだかAnthropicだかが事前に取得する機会があっただけだったというゴミのような話があった。 | 「LLM(生成AI)が数学の難問を解いた」というのが話題になった時、実際には数学者の最新の研究成果をopenAIだかAnthropicだかが事前に取得する機会があっただけだったというゴミのような話があった。 | ↗Not briefed |
rss techmeme | 14d | Major environmental groups are embracing AI to analyze climate data, track forest fires, and more, even as they warn of the environmental impact of data centers (Valerie Volcovici/Reuters) | Valerie Volcovici / Reuters: Major environmental groups are embracing AI to analyze climate data, track forest fires, and more, even a… | ↗Not briefed |
hn hn-ai | 14d | Ask HN: LLM-Resistant Publishing Strategies | As has been discussed at length, LLM companies carried out the most egregious theft of labor [0] and intellectual property [1] in huma… | ↗Not briefed |
bluesky bluesky-ai | 14d | Popart filter applied to an AI image. #AI #AIart #AIartwork #AIimage #AIimagegeneration #AIgenerated #digitalart #digitalartwork #generati… | Popart filter applied to an AI image. #AI #AIart #AIartwork #AIimage #AIimagegeneration #AIgenerated #digitalart #digitalartwork #gen… | ↗Not briefed |
bluesky bluesky-ai | 14d | Ich mache bei Fachartikeln, die ich gelesen habe, anschließend meistens noch einen Faktencheck mit KI. Inzwischen ist das fast schon ein We… | Ich mache bei Fachartikeln, die ich gelesen habe, anschließend meistens noch einen Faktencheck mit KI. Inzwischen ist das fast schon e… | ↗Not briefed |
rss techmeme | 14d | Internal documents leaked by Cisco workers angry over the company's ties to Israel's military show how it expanded its business as the war in Gaza dragged on (Alice Speri/The Guardian) | Alice Speri / The Guardian: Internal documents leaked by Cisco workers angry over the company's ties to Israel's military show how it… | ↗Not briefed |
bluesky bluesky-ai | 14d | LLM evangelists say this a lot and I have no idea what it’s supposed to mean. The people who use it become less adept at their own work, so… | LLM evangelists say this a lot and I have no idea what it’s supposed to mean. The people who use it become less adept at their own wor… | ↗Not briefed |
hn hn-ai | 14d | Show HN: ScrivenSpell | ScrivenSpell is a word game with RPG elements. If you like Scrabble, you might like this.<p>This game is inspired by Up Spell, an old… | ↗Not briefed |
arxiv arxiv-ai | 14d | Guidance at Pure Noise Erodes Speaker Similarity in Few-Step Token-Conditioned Flow-Matching TTS | Text-to-speech systems built from a language model and a flow-matching mel decoder now run the decoder with two to four Euler steps in… | ↗Not briefed |
arxiv arxiv-ai | 14d | When Are Nearby Experts Composable? Cross-Task Model Merging in Neural Thickets | Random perturbations around a shared checkpoint can produce experts for distinct tasks, but shared origin does not ensure that their g… | ↗Not briefed |
bluesky bluesky-ai | 14d | Launched this week on TrustList: Goethe Apache-2.0 server that turns a Linux machine with NVIDIA GPUs into a private language model API, a… | Launched this week on TrustList: Goethe Apache-2.0 server that turns a Linux machine with NVIDIA GPUs into a private language model A… | ↗Not briefed |
rss techmeme | 14d | The crypto industry is bracing for a reckoning after Democrats won the US midterms and vowed to investigate the Trump family and its ties to the industry (Financial Times) | Financial Times: The crypto industry is bracing for a reckoning after Democrats won the US midterms and vowed to investigate the Trump… | ↗Not briefed |
arxiv arxiv-ai | 14d | What Changed in Engineering Writing After Generative AI Chat Assistants: Evidence from 1.8 Million Articles | This study asks what changed in engineering articles after conversational language models became publicly available on November 30, 20… | ↗Not briefed |
bluesky bluesky-ai | 14d | The Blue Wave was powered by a few things: War, prices, AI backlash, Schumer doesn't have the guts to pull funding for the war. His daughte… | The Blue Wave was powered by a few things: War, prices, AI backlash, Schumer doesn't have the guts to pull funding for the war. His da… | ↗Not briefed |
arxiv arxiv-ai | 14d | Do Benign-Trained Hidden-State Probes Detect Tool Errors Under Adversarial Conditions? | Although hidden-state probes efficiently monitor model operation, it remains unclear whether probes trained on benign tool use remain… | ↗Not briefed |
arxiv arxiv-ai | 14d | ClusLoad: Cluster-Guided Model Sharing for Scalable Distribution Transformer Load Forecasting | Large-scale device-level load forecasting for distribution transformers is challenged by substantial cross-device heterogeneity and th… | ↗Not briefed |
rss techmeme | 14d | Nous Research unveils Arete-27B, a post-trained Qwen3.8-27B model that reinforces Western values, saying the underlying Chinese model's bias layer is shallow (Nous Research) | Nous Research: Nous Research unveils Arete-27B, a post-trained Qwen3.8-27B model that reinforces Western values, saying the underlying… | ↗Not briefed |