AI News Today

arXiv:2409.13007v3 Announce Type: replace Abstract: Class imbalance poses a significant challenge in classification tasks, often causing standard learning algorithms to become biased toward the majority class. Cost-sensitive learning (CSL) addresses this issue by assigning higher penalties to minority-class misclassifications. However, conventional CSL typically applies a uniform penalty to all minority-class instances, ignoring the fact that minority samples may differ substantially in terms of

arXiv:2501.07400v2 Announce Type: replace Abstract: We derive explicit equations governing the cumulative biases and weights in Deep Learning with ReLU activation function, based on gradient descent for the Euclidean loss in the input layer, and under the assumption that the weights are, in a precise sense, adapted to the coordinate system distinguished by the activations. We show that gradient descent corresponds to a dynamical process in the input layer, whereby clusters of data are progressiv

arXiv:2508.03898v2 Announce Type: replace Abstract: Accurate prediction of grape phenology is essential for timely vineyard management decisions, such as scheduling irrigation and fertilization, to maximize crop yield and quality. While traditional biophysical models calibrated on historical field data can be used for season-long predictions, they lack the precision required for fine-grained vineyard management. Deep learning methods are a compelling alternative but their performance is hindered

arXiv:2510.03243v3 Announce Type: replace Abstract: Efficient scheduling of large language model (LLM) inference tasks is critical for achieving low latency and high throughput, a challenge that is becoming increasingly acute with the rise of reasoning-capable LLMs whose generation lengths are highly variable. Traditional strategies like First Come, First-Serve (FCFS) often suffer from Head-of-Line (HOL) blocking, where long-running tasks delay shorter ones queued behind them. In this paper, we

arXiv:2510.09685v2 Announce Type: replace Abstract: Deep learning has become a pivotal technology in fields such as computer vision, scientific computing, and dynamical systems, significantly advancing these disciplines. However, neural Networks persistently face challenges related to theoretical understanding, interpretability, and generalization. To address these issues, researchers are increasingly adopting a differential equations perspective to propose a unified theoretical framework and sy

arXiv:2512.23075v5 Announce Type: replace Abstract: Policy gradient methods for Large Language Models optimize a policy $\pi_\theta$ via a surrogate objective computed from samples of a rollout policy $\pi_{\text{roll}}$. However, modern LLM-RL pipelines suffer from unavoidable implementation divergences -- backend discrepancies, Mixture-of-Experts routing discontinuities, and distributed training staleness -- causing off-policy mismatch ($\pi_{\text{roll}} \neq \pi_\theta$) and approximation er

arXiv:2601.16632v5 Announce Type: replace Abstract: Time series forecasting has witnessed significant progress with deep learning. While prevailing approaches enhance forecasting performance by modifying architectures or introducing novel enhancement strategies, they often fail to dynamically disentangle and leverage the complex, intertwined temporal patterns inherent in time series, thus resulting in the learning of static, averaged representations that lack context-aware capabilities. To addre

arXiv:2602.16065v2 Announce Type: replace Abstract: As artificial intelligence (AI)-generated content proliferates, models are increasingly trained on their own outputs, risking progressive degradation or collapse. In this article, we provide the first positive, rigorous theoretical results, to the best of our knowledge, showing that under model-agnostic mild conditions, the model converges to the true data-generating distribution. The convergence rate is the minimum of the model's intrinsic rat

arXiv:2605.27482v2 Announce Type: replace Abstract: While orthogonal subspace methods try to mitigate task interference in Continual Learning (CL), they often suffer from energy diffusion across the basis, hindering knowledge compaction and exhausting capacity for future tasks. We observe that output feature drift induced by parameter updates is inherently low-rank, and theoretically prove that preserving parameters along the principal directions of this drift minimizes the output reconstruction

arXiv:2605.30719v2 Announce Type: replace Abstract: We study when large language models (LLMs) can serve as effective black-box policy optimizers for reinforcement learning (RL) tasks, i.e., when can we replace classical RL algorithms with an LLM? We explore this question by introducing Prompted Policy Optimization (PromptPO), an iterative method that prompts an LLM with Python descriptions of the state space, action space, and reward function, then has it generate and refine executable policies

arXiv:2606.21641v2 Announce Type: replace Abstract: Large language models (LLMs) have been proposed as hyperparameter-optimization (HPO) advisors that "warm-start" search from prior knowledge, proposing strong configurations in very few evaluations. We test that claim under a budget-matched, multi-seed protocol on eight PMLB tabular benchmarks, comparing an LLM advisor (LLM-OptFlow) against four classical baselines (random search, Optuna-TPE, Gaussian-process Bayesian optimization, and successiv

<p>Last month, I had a problem with a popular mobile banking app in Southeast Asia. Nothing exotic. A transaction didn't go through, and my support ticket had been sitting untouched for two weeks.</p> <p>So I opened the app's chatbot. It greeted me warmly, asked how it could help, and then couldn't do a single useful thing. It couldn't look up my transaction. It couldn't check the status of my ticket. It couldn't tell me why my issue was unresolved. It could answer FAQ questions, and that was it

<pre style='white-space:pre-wrap;width:81ex'>fix(mistral): bound streaming response bodies Bounds Mistral SDK response streams at 16 MiB using the existing streaming byte guard.</pre>

<p>Over the past year, one concept has fundamentally changed how I think about AI applications.</p> <p>Not larger language models.</p> <p>Not better prompts.</p> <p>Not even AI agents.</p> <p>It's <strong>Model Context Protocol (MCP)</strong>.</p> <p>For a long time, most AI applications lived inside a closed environment. They could generate text, answer questions, or write code, but they couldn't easily interact with external systems.</p> <p>MCP changes that.</p> <p>It provides a standardized w

<p>Real numbers: XEdge has 700+ users, $0 revenue, and a $29 playbook that hasn't sold a single copy in a month, even with a free coupon code live.<br> Here's my honest read on why.<br> I assumed "real, lived experience" would sell itself. It doesn't. I wrote a genuinely specific document — actual TAM/SAM/SOM math from XEdge, the exact validation prompts I used, the pricing mistake I almost made — but my posts about it stayed vague. "Check out my playbook" doesn't tell anyone why it's different

<pre style='white-space:pre-wrap;width:81ex'>fix(provider-usage): bound Anthropic usage error response reads to prevent OOM (#97614) Replace unbounded res.json() with readProviderJsonResponse in the fetchClaudeUsage error path to cap error body reads at 16 MiB.</pre>

Article URL: https://github.com/raiyanyahya/llmaker Comments URL: https://news.ycombinator.com/item?id=48714397 Points: 3 # Comments: 1

<p>Value of some chip manufacturers have tripled, or more, driving Asia Pacific stock markets sharply higher</p><p>Shares in chipmakers have surged in the first half of this year as investors piled into companies that make the hardware underpinning the AI boom, according to analysis.</p><p>Investors have driven up the value of semiconductor and memory chip manufacturers, whose profits have soared during 2026, at the expense of some large software companies, which have fallen out of favour this y

<pre style='white-space:pre-wrap;width:81ex'>fix(secrets): skip PLAINTEXT_FOUND for known non-secret apiKey markers (#97622) Exclude models.providers.*.apiKey values that match isNonSecretApiKeyMarker (e.g., lmstudio-local, ollama-local) from secrets audit plaintext warnings. Regression test covers marker bypass and real-key flagging. Closes #89233</pre>

Recently studying agent and workflow. Decided to spawn a /goal command on Claude code and made this. Comments URL: https://news.ycombinator.com/item?id=48714073 Points: 1 # Comments: 0

Article URL: https://fuckui.com/fuckui Comments URL: https://news.ycombinator.com/item?id=48714065 Points: 2 # Comments: 1

Hi, we have open-sourced a signed https://emem.dev [ https://github.com/Vortx-AI/emem ] to enable ai agents leverage the physical world in a repetitive, cite-able manner. Think of us as https for the real world intelligence. Do give it a try by adding https://emem.dev/mcp in your agentic workforce. Comments URL: https://news.ycombinator.com/item?id=48713981 Points: 2 # Comments: 0

The people said 360 One is expected to invest US$20 million to US$25 million, and other investors could join the round.

Article URL: https://gurulabs.com/blogs/edgeos-dhcrelay-binary-patch/ Comments URL: https://news.ycombinator.com/item?id=48713705 Points: 3 # Comments: 0

Comments

Article URL: https://www.phantafield.com/whitepaper Comments URL: https://news.ycombinator.com/item?id=48713686 Points: 26 # Comments: 29

Article URL: https://twitter.com/bbuddha_xyz/status/2061606383435620527 Comments URL: https://news.ycombinator.com/item?id=48713682 Points: 2 # Comments: 1

Article URL: https://www.youtube.com/watch?v=PNU3YILDyYc Comments URL: https://news.ycombinator.com/item?id=48713671 Points: 3 # Comments: 0

XSDR is a unified pipeline for monitoring activity on X and the web. I built it because I wanted my agent to do things based on real-time events that were taking place instead of polling the web or scheduling cron jobs. With XSDR, you can trigger agentic loops based on real-time events. All you need is an API key, a webhook that can receive POST requests, and an idea of what you’re looking for. It currently supports X and Firehose (aka a persistent web crawler). Comments URL: https://news.ycombi

A wild tale of rediscovery. ScienceAlert stories are written, fact-checked, and edited by humans, never generated by AI. Don't miss a story, subscribe here.

Article URL: https://www.youtube.com/watch?v=9tvJ_GYJA-o Comments URL: https://news.ycombinator.com/item?id=48713470 Points: 1 # Comments: 0

Article URL: https://www.ft.com/content/e81ce414-e4bd-4e8c-bac7-94f7bf17def4 Comments URL: https://news.ycombinator.com/item?id=48713221 Points: 1 # Comments: 1

wavecat is a fully local personal agent that watches your screen. It develops a rich understanding of your needs and goals by constantly viewing your activity. Don't worry, none of your personal data ever leaves your computer since all the models run locally. I think it's something cool to show a feature that local agent systems can do that cloud systems can't. If you have a beefy Apple Silicon Mac or a nice GPU, you should be able to run it! Comments URL: https://news.ycombinator.com/item?id=48

xAI's audio models are now live on AI Gateway. Realtime voice, text to speech, and speech to text are all available through the AI SDK with the same routing, observability, and spend controls as your other models. These capabilities are available on the AI SDK 7 release. Available models Capability Models Realtime voice xai/grok-voice-think-fast-1.0 Text to speech xai/grok-tts Speech to text xai/grok-stt Realtime A voice agent has two pieces: a server route that mints a short-lived token, so you

AI Gateway now supports voice and audio models. You can build realtime voice agents, generate speech from text, and transcribe audio to text. This provides the same observability, spend controls, and bring-your-own-key support as text, image, and video models in AI Gateway, with no markup or platform fees. These capabilities are in beta and available via AI SDK 7. With realtime support, a single model takes audio in and audio out, so a user can talk and hear a reply back in near real time instea

Three years into the productivity promise, there's finally enough hard evidence to answer the question plainly: does working with AI actually make you more productive? Yes — spectacularly, for some people on some tasks. And no, or worse, for others. The gains are real. They're just not flowing where the marketing said they would. This issue maps who wins, who pays, and why the line between them isn't where you think.

Vercel AI Gateway now supports routing rules . Routing rules are firewall-style rules that control which models your team can use, applied at the gateway level instead of in your application code. When a model goes down or gets retired, you usually have to ship a code change to move off it. With routing rules, you push one rule and every request reroutes instantly. There are two types: Type What it does Use it to Rewrite Serves a request for one model using another Keep traffic flowing when a mo

Article URL: https://betterimagesofai.org/ Comments URL: https://news.ycombinator.com/item?id=48713051 Points: 2 # Comments: 0

Its now clear that we need to preserve tech press for non AI related things. Techmeme for example is now completely overrun with AI stories. HN is getting closer to that every day. If AI kickback deals, phony new model ratings, high RAM prices and your surprise at how you think you coded something with AI and it was AMAZING! even though it doesnt work is all there is count me out. We need a filter on existing tech news sites or an alternative press. Comments URL: https://news.ycombinator.com/ite

Article URL: https://devfortress.net/blog/semi-annual-2026 Comments URL: https://news.ycombinator.com/item?id=48713032 Points: 1 # Comments: 0