AINews Portal

News

50 items (filtered)

The raw AI news stream — everything we ingest across RSS, Hacker News, arXiv, GitHub, Bluesky, YouTube and more, newest first.

SourceAgeTitleSummaryStatus
arxiv
arxiv-ai
1moEvaluating Exact Output and Checkpoint-State Prediction in Real ProgramsWe present a benchmark for predicting final output and checkpoint state from source and input alone. It extends CRUXEval-style output…
↗Not briefed
hn
hn-ai
1moShow HN: I gave my Pi agent eyes on Xpi-x is a pi extension that lets Pi coding agent to interact with X feed to look up trends, news and many more stuff.
↗Not briefed
bluesky
bluesky-ai
1moever since the openai mathstra paper drop, this guy has had regular astra in a loop, tightening one of the proofs, dropping new improvement…ever since the openai mathstra paper drop, this guy has had regular astra in a loop, tightening one of the proofs, dropping new improv…
↗Not briefed
hn
hn-ai
1moShow HN: Phobos – a tiny kernel language that now runs LLMs on a 2019 GPU—
↗Not briefed
hn
hn-ai
1moShow HN: Wordminos – Crosswords Meet DominosI built a daily word game in 12 days.<p>The base mechanic only took an hour. Getting it to the finish line took the rest.<p>The work…
↗Not briefed
bluesky
bluesky-ai
1moVoice training an AI model to read to your kid is stupid because the words are only a portion of it. What does the child see in the pictu…Voice training an AI model to read to your kid is stupid because the words are only a portion of it. What does the child see in the…
↗Not briefed
bluesky
bluesky-ai
1moThis guy is gonna be so angry about his divorce and the way the judge barely gave him any time with the kid but makes him spend a ton on ch…This guy is gonna be so angry about his divorce and the way the judge barely gave him any time with the kid but makes him spend a ton…
↗Not briefed
hn
hn-ai
1moStages of migration to new Outlook for Windows—
↗Not briefed
hn
hn-ai
1moShow HN: Gauze fixes (some) open-weight LLM deficienciesIf you have worked with open-weight&#x2F;local LLMs, you&#x27;ve known both the highs and the lows. Lows there are many: malformed res…
↗Not briefed
arxiv
arxiv-ai
1moHow Is Automated Research Evaluated? A Survey of Benchmarks and Evaluation PracticesAutomated research systems support literature synthesis, ideation, experiments, writing, and peer review, but their evaluation is disp…
↗Not briefed
bluesky
bluesky-ai
1moIn eighteen years his kids will send an AI model to visit him for the holidays.In eighteen years his kids will send an AI model to visit him for the holidays.
↗Not briefed
bluesky
bluesky-ai
1moIts not. Human knowledge will *always* exceed that of an LLM because they can only be trained on things we've written down. One, we acquire…Its not. Human knowledge will *always* exceed that of an LLM because they can only be trained on things we've written down. One, we ac…
↗Not briefed
hn
hn-ai
1moSkipping Stones – Git history with the agent conversations behind each commit—
↗Not briefed
bluesky
bluesky-ai
1moand then they all get fed into the AI model with no discernment from the bad ones... 😬and then they all get fed into the AI model with no discernment from the bad ones... 😬
↗Not briefed
hn
hn-ai
1moShow HN: Momenial for Events — White-label ticketing appI built a white-label ticketing and registration app last year to run an event for a large state-owned corp, and I decided to turn it…
↗Not briefed
bluesky
bluesky-ai
1mo生成AI(LLM、VLM等)による資料画像の汚染、マジで迷惑すぎるよなあ 本当は写真を資料にしたいのに、出典がはっきりしてる漫画の方が資料として確実になってしまっているなんて生成AI(LLM、VLM等)による資料画像の汚染、マジで迷惑すぎるよなあ 本当は写真を資料にしたいのに、出典がはっきりしてる漫画の方が資料として確実になってしまっているなんて
↗Not briefed
bluesky
bluesky-ai
1moup next: how i replaced my wife with an AI modelup next: how i replaced my wife with an AI model
↗Not briefed
arxiv
arxiv-ai
1moLearning structured linear dynamical systems from missing observationsWe consider the problem of learning structured linear dynamical systems over convex sets $\mathcal{K}$, where only a small subset of t…
↗Not briefed
bluesky
bluesky-ai
1moOn dirait que l'annonce faite il y a deux jours par OpenAI à propos de la résolution d'un paquet de problèmes mathématiques ne passe pas co…On dirait que l'annonce faite il y a deux jours par OpenAI à propos de la résolution d'un paquet de problèmes mathématiques ne passe p…
↗Not briefed
bluesky
bluesky-ai
1mo韓国金融機関9社へのAI攻撃、Claude Code履歴が公開状態に。CrowdStrikeが攻撃者の身元手がかりを特定韓国金融機関9社へのAI攻撃、Claude Code履歴が公開状態に。CrowdStrikeが攻撃者の身元手がかりを特定
↗Not briefed
arxiv
arxiv-ai
1moUnderstanding Latent-Dimension Scaling in Dynamical-System Learning through Spectral ReliabilityIn deep learning, approximation theory motivates increasing representation size. We ask whether this benefit extends to dynamics learn…
↗Not briefed
arxiv
arxiv-ai
1moConditional Kernel Stein DiscrepancyKernel Stein discrepancies (KSDs) provide a versatile tool for comparing distributions. One of their main applications is in quantifyi…
↗Not briefed
arxiv
arxiv-ai
1moLEVER: Adaptive Cost-Aware Proof Search Over AND/OR GraphsMathematicians value proofs for more than correctness: among correct proofs, simplicity, purity and the computational cost of finding…
↗Not briefed
hn
hn-ai
1moSolo Developer Rebuilds Adobe Creative Suite in Rust Using Claude—
↗Not briefed
hn
hn-ai
1moShow HN: ScreenCI – Tutorial videos as E2E tests, written by your coding agentHi HN, I&#x27;m Olli. My friend Patrik and I are bootstrapping ScreenCI: tutorial videos of your web app that you can re-record in you…
↗Not briefed
arxiv
arxiv-ai
1moTrajectory-Guided Fault Localization for Agent Skill EvolutionAgent skills provide reusable guidance for code agents, but incomplete or unsuitable guidance can impair task execution. To reduce the…
↗Not briefed
arxiv
arxiv-ai
1moGRPODropout: Less is More for Online Reinforcement Learning RolloutsReinforcement learning (RL) methods such as GRPO substantially improve large language model reasoning but often suffer from policy ent…
↗Not briefed
arxiv
arxiv-ai
1moDADP: Dynamic Activity-Dependent Pruning, A Reverse Hebbian-Inspired Structural Pruning MethodModern neural networks are heavily over-parameterized. This redundancy incurs substantial compute and memory overhead during training…
↗Not briefed
bluesky
bluesky-ai
1moTrain a voice model? This guy is stuck in 2025. All the best parenting is done by agentic AI these days.Train a voice model? This guy is stuck in 2025. All the best parenting is done by agentic AI these days.
↗Not briefed
bluesky
bluesky-ai
1moI’m excited for OpenAI to be destroyed 3 out of almost 400 new math results retracted isn’t giving me anything tbhI’m excited for OpenAI to be destroyed 3 out of almost 400 new math results retracted isn’t giving me anything tbh
↗Not briefed
rss
techmeme
1moAI retail trading agent startup Catalyst raised a $30M seed led by Sequoia and says a pilot generated hundreds of millions in trading volume over two weeks (Allie Garfinkle/Fortune)Allie Garfinkle / Fortune: AI retail trading agent startup Catalyst raised a $30M seed led by Sequoia and says a pilot generated hundr…
↗Not briefed
hn
hn-ai
1moMellum2.1 Gets to Work: A Fast Open Model for Coding Agents—
↗Not briefed
arxiv
arxiv-ai
1moOpen-Vocabulary Audio-Visual Event Localization via Complex-Valued FusionOpen-Vocabulary Audio-Visual Event Localization (OV-AVEL) labels each video segment with an event class, including classes that were n…
↗Not briefed
arxiv
arxiv-ai
1moDetecting Spin in Clinical Trials with Large Language ModelsSpin in clinical trials includes reporting practices that distort the presentation of results. This is particularly critical in medici…
↗Not briefed
hn
hn-ai
1moShow HN: Bot Bowl Bench – Benchmarking LLMs with Blood BowlI wanted a meaningful benchmark for Mistral 4, and as a fan of Blood Bowl, I&#x27;m faintly aware that it&#x27;s a hugely challenging…
↗Not briefed
bluesky
bluesky-ai
1moSupplement hawkers, alien channelers, and New Age enema healers usually have a unique style, cadence, intensity of intrusive eye contact. A…Supplement hawkers, alien channelers, and New Age enema healers usually have a unique style, cadence, intensity of intrusive eye conta…
↗Not briefed
bluesky
bluesky-ai
1moThe kid absolutely shouldn't have used AI for his route. But Claude's UX design is deliberately seductive. AI products are defective produc…The kid absolutely shouldn't have used AI for his route. But Claude's UX design is deliberately seductive. AI products are defective p…
↗Not briefed
hn
hn-ai
1moShow HN: Etchv – API for invisible watermarks in images, PDFs and videosHello HN!<p>The inspiration for Etchv came directly from friends who raised concerns about the growing ease of unauthorized media copy…
↗Not briefed
bluesky
bluesky-ai
1moDTU-born PSV Tech closes Fund II, writing the next chapter in their 25 years historyof pre-seed investing techlensmedia.com/news/dtu-bor.…DTU-born PSV Tech closes Fund II, writing the next chapter in their 25 years historyof pre-seed investing techlensmedia.com/news/dtu…
↗Not briefed
bluesky
bluesky-ai
1moWhen you’re so caught up in AI that you forget what it means to be human…When you’re so caught up in AI that you forget what it means to be human…
↗Not briefed
arxiv
arxiv-ai
1moRecovery Guarantees for Posterior Sampling of One-Bit Compressed SensingWe study the sample complexity of noisy one-bit compressed sensing for signals drawn from a prior distribution. By characterizing the…
↗Not briefed
arxiv
arxiv-ai
1moProbability-Signature Dynamics: Unpacking Modular Addition Learning Within Two-Layer NetworksNeural networks trained on modular addition tasks often develop Fourier-structured representations that support exact generalization.…
↗Not briefed
hn
hn-ai
1moHow HN: Dialtrace – see where the time goes on a voice AI phone call—
↗Not briefed
bluesky
bluesky-ai
1moThere is no such thing as competent use of an ai, nor is there such a thing as a competent model. Competence necessitates consideration and…There is no such thing as competent use of an ai, nor is there such a thing as a competent model. Competence necessitates consideratio…
↗Not briefed
arxiv
arxiv-ai
1moFrom Suppression to Repair: Mitigating Object Hallucination in Large Vision-Language Models via Localized Distribution AlignmentObject hallucination remains a major obstacle for large vision-language models (LVLMs) to generate reliable content. An intuitive miti…
↗Not briefed
arxiv
arxiv-ai
1moSelf-Supervised Speech Representations for Cross-Speaker Dysarthria Detection During Awake CraniotomyDetecting intra-operative speech impairment during awake craniotomy is essential for preserving language function. However, automated…
↗Not briefed
rss
techmeme
1moDrones struck a data center owned by Russia's Yandex in the town of Sasovo, marking the first major attack on a Russian data hub since the Ukraine war began (Reuters)Reuters: Drones struck a data center owned by Russia's Yandex in the town of Sasovo, marking the first major attack on a Russian data…
↗Not briefed
hn
hn-ai
1moShow HN: Get YouTube Text – free YouTube transcript tool, no signup or ads—
↗Not briefed
bluesky
bluesky-ai
1moWhat that means is that poorer people will buy less tech and hang on to old stuff longer. Since manufacturers don’t make money on old tech,…What that means is that poorer people will buy less tech and hang on to old stuff longer. Since manufacturers don’t make money on old…
↗Not briefed
arxiv
arxiv-ai
1moFrom Pixels to Structure: Lightweight Vision-Language Models for Document OCR and Structured JSON ExtractionWhile massive, closed-source Vision-Language Models (VLMs) set strong benchmarks for document understanding, their dependence on comme…
↗Not briefed