Andreessen Horowitz tracks actual US consumer spending for the first time in its latest Top 100 AI list. Nearly half of US consumers use AI, but only 4.5 percent pay for a subscription. The top one percent of payers spend about $900 a month, mostly on professional tools for development and automation. The article Few people pay for AI, but those who do spend big appeared first on The Decoder .
Article URL: https://github.com/rociiu/talorys Comments URL: https://news.ycombinator.com/item?id=50031614 Points: 233 # Comments: 118
<!-- SC_OFF --><div class="md"><p>I am a data scientist who started working in the industry before the LLM revolution. Back then, Jupyter Notebooks were a perfect fit for the classical DS pipeline: EDA -> data prep -> fit -> eval -> tune -> save model artefact and notebook.</p> <p>Lately, I have been thinking a lot about how agentic development and LLMs are changing the way data scientists work. Especially in classical ML applications, where you still need to explore data, run experiments, check
<p> The macOS terminal that shows which coding agent needs you </p> <p> <a href="https://www.producthunt.com/products/farol-3?utm_campaign=producthunt-atom-posts-feed&utm_medium=rss-feed&utm_source=producthunt-atom-posts-feed">Discussion</a> | <a href="https://www.producthunt.com/r/p/1275794?app_id=339">Link</a> </p>
was wondering if YC still manually reviews all pitch decks or its just agents? Comments URL: https://news.ycombinator.com/item?id=50031461 Points: 2 # Comments: 0
Its easy to see how an prompt input the baseline of every AI app lacks the ability to do the most simple things in the right way and how it differs the most in the same app itself. Is it because the care for the products have reduced? or is it because people have started to care less about what they are using to get the work done? debugged some here: https://bumicodes.vercel.app/writing/prompt-input Comments URL: https://news.ycombinator.com/item?id=50031447 Points: 1 # Comments: 0
Philadelphia police said the tip was "flagged as spam", but criticised the tech company for taking more than two months to detect and report the breach.
<p>George Williamson says ‘national resilience’ is key focus of ATI amid US and China’s runaway leadership in field</p><p>The UK must not become dependent on foreign AI systems that can be switched off at short notice and must develop its own versions of the technology, according to the head of the Alan Turing Institute.</p><p>The country needs a stronger domestic position in AI in response to the US and China’s runaway leadership in the field, with “national resilience” a key focus for ATI said
<a href="https://www.theguardian.com/commentisfree/picture/2026/oct/10/jon-davis-job-titles-ai-cartoon">Continue reading...</a>
(Paul Hudson/Flickr/CC BY-NC-ND 2.0) Including an actual wooden club. ScienceAlert stories are written, fact-checked, and edited by humans, never generated by AI. Don't miss a story, subscribe here.
<p><em>This is a submission for the <a href="https://dev.to/challenges/kaggle-2026-09-23">Kaggle Benchmarking Challenge</a></em></p> <h2> What I Benchmarked </h2> <p>When a solver or an AI model returns "the optimal solution", it rarely says whether it is the only one. That matters: if many equally good solutions exist, a small change in the data can silently move you to a very different "optimal" answer. I wanted to know whether language models notice this, so the benchmark asks one question: <
<p>I wanted one place that tells me what is going on in my homelab: who is poking at it from outside, which boxes have known-exploited CVEs, what needs patching, and whether anything new turned up on the network. And I wanted it on a small box, without putting an agent on every machine.</p> <p>The obvious tools are good, and I'll say that up front. Wazuh does far more than this, but it wants an agent per host and its all-in-one install asks for 4 vCPU and 8 GB. CrowdSec is brilliant at reading l
<p>In 2021, a company you've probably never heard of quietly acquired the de-identified medical records of roughly 50 million Americans. The price was reported at around $1 billion. The records included diagnoses, prescriptions, lab results, and visit histories, stripped of names and Social Security numbers. The sellers were hospitals and health systems that had collected this data as a byproduct of care. The buyers were building products nobody had consented to be part of.</p> <p>That's not a s
<p>I asked my agent to build pr-rulebook from an idea that came up in conversation. When I wrote about the result, I included the candidate rule that failed review. That failure explains something important: a repeated comment is a lead, not an automatic team policy.</p> <p>Here is how I approach turning code-review feedback into useful context for Cursor.</p> <h2> Collect evidence, not just wording </h2> <p>Start with human comments on merged pull requests. Check what changed after each comment
<p><em>This is a submission for the <a href="https://dev.to/challenges/hacktoberfest-week1-2026-10-05">Hacktoberfest Open-Source AI Challenge Week 1: Touch Grass</a></em></p> <h2> What I Built </h2> <p><strong>FloraQuest-AI</strong> is a lightweight, offline-first web companion designed to get you off the screen and into nature in under 60 seconds.</p> <p>Features:</p> <ul> <li> <strong>Offline Specimen ID:</strong> Identify trail leaves, trees, and birds using open-weight vision models (MobileN
<p><em>This is a submission for the <a href="https://dev.to/challenges/hacktoberfest-week1-2026-10-05">Hacktoberfest Open-Source AI Challenge Week 1: Touch Grass</a></em></p> <h2> What I Built </h2> <p><strong>Free Time Walk Planner</strong> is a local AI agent that turns <em>"I have 1 hour free"</em> into a real plan for getting outside.</p> <p>It picks a nearby park, tells you <strong>when to leave</strong> and <strong>when to turn back</strong> so you're home before sunset, and says <strong>w
<p>A page fetch can look successful while your agent pipeline quietly depends on details that are no longer in the response. Cloudflare’s 9 October change to Markdown for Agents replaces buffered conversion with an in-process streaming path. Alongside that change, converted responses lose three headers that clients sometimes use for accounting and response handling.</p> <p>If your application asks Cloudflare for Markdown by sending <code>Accept: text/markdown</code>, update the consumer before r
<p><em>This is a submission for the <a href="https://dev.to/challenges/hacktoberfest-week1-2026-10-05">Hacktoberfest Open-Source AI Challenge Week 1: Touch Grass</a></em></p> <h2> What I Built </h2> <p>Walkworth is a trek journal you speak instead of type. Before a walk you open the app once on Wi-Fi so it caches. On the trail you put the phone in your pocket, and when something is worth keeping you tap once, say it out loud, and tap again. "Langurs on the wall near the gate, three of them, tota
<pre style='white-space:pre-wrap;width:81ex'>fix: show recovery guidance when prompts exceed model context (#168326)</pre>
Article URL: https://grigio.org/offline-voice-assistant-no-dedicated-gpu-less-15gb-ram/ Comments URL: https://news.ycombinator.com/item?id=50031212 Points: 1 # Comments: 0
<pre style='white-space:pre-wrap;width:81ex'>feat(cli): show progress while tool input is generated (#167789) claude-cli turns went silent while a tool call's input was generated, because CLI input_json_delta fragments were buffered until content_block_stop with no agent events in between (35-55 s of silence for a long Write on Claude Code 2.1.296). The CLI parser now emits the existing counts-only input_delta tool progress (shared with embedded runs) while input streams, without claiming execut
<pre style='white-space:pre-wrap;width:81ex'>fix(agents): enforce provider response-header deadlines (#168282) * fix(agents): enforce provider response-header deadlines Carry the first-event deadline through synchronous and asynchronous stream creation until provider acceptance or activity. Keep provider timeout overrides, local post-acceptance gap behavior, and transport first-SSE deadlines under their existing owners. Exercise missing response headers through the real Responses, Completions, A
Anthropic on Friday said it's cutting off live internet access for all its internal evaluations following the discovery of new incidents in which its artificial intelligence (AI) models exhibited misaligned behavior and targeted real websites. The AI company said it identified four broad categories of unintended model actions during evaluations and internal use of Claude - Claude Mythos
<pre style='white-space:pre-wrap;width:81ex'>perf(state): release caller memory after agent database opens (#168349) Bind the long-lived native-loss observer to the database path without capturing the completed opening caller. Preserve loss-time validation identity lookup and keep the opened store with its existing owner. Extend the native lifecycle retention fixture through 101 callers while the same database generation remains prepared. The regression collects all 303 caller, source, and buffe
<pre style='white-space:pre-wrap;width:81ex'>fix(agents): keep the poll vote echo matcher private (#168333) #167979 moved the matcher's only production caller into suppressPollVoteEcho, leaving isPollVoteEchoText exported solely for its unit test. The Knip production unused-export scan rejects that export, so pnpm check:changed fails on main for local and Testbox runs (GitHub Actions only reports it as advisory). Make the matcher module-private and drive the unchanged matcher table through the m
Amazon officially unveiled its all-new Android Kids Tablets in 8.7-inch and 11-inch sizes, featuring parent-managed Google Play access, AI Storybuilder, and kid-proof cases.
<pre style='white-space:pre-wrap;width:81ex'>refactor: deslop agent tools and macOS leftovers (#168168) Combine the macOS and agent-tool cleanup lanes around their existing schema, media dispatch/result, harness lifecycle, native parsing, capture, worker, and process-identity owners. Preserve validation, authority, and rollback. Remove 531 net production lines across 36 files. Drop the entire sessions-list cleanup during the conflict-only rebase so main's sidebarRoot behavior wins. Join native a
Hello awesome people, I've been thinking a lot lately about how I can make my knowledge and craft more valuable in the times we live in. I've been building UI libraries for the past 10 years, and I've been on a quest to perfect the components that I build. I want to share my knowledge with you. Like most of you, probably, I've been using a lot of AI in my craft, and I feel most of the time the code is just good enough, but never at the level of quality that I strive for or that I would reach if
Article URL: https://github.com/aindeev/agent-chrome-relay Comments URL: https://news.ycombinator.com/item?id=50030811 Points: 5 # Comments: 0
Source attribution matters to 79% of respondents – handy for a company licensing its Q&A archive to model builders
<!-- SC_OFF --><div class="md"><p></p> <p>I am going to fine-tune a satire model, with the base model being `Qwen3.5-14B-Base`. The dataset has ~18000 examples, most of which are pretty long and do not align with the model's internal knowledge (they have incorrect answers to facts), so I needed to do a full fine tune rather than use LoRA. </p> <p>Today, I rented a cloud computer on vast.ai. It had 4x12GB GPUs and 48GB of system RAM, but I feel stupid because it kept running out of memory when I
How do you keep improving an AI agent after the first 80%, once it serves real production traffic and silent quality regressions return HTTP 200 OK? We are open-sourcing AQuA (Ambient Quality Agent), a reference implementation that runs beside your production agent in Google Cloud, sweeps and verifies failure clusters against conversation transcripts, and diagnoses root causes anchored to the exact deployed source snapshot.
Google AI Edge has introduced ML Drift, a high-performance, universal GPU compute framework designed to accelerate on-device AI and machine learning inference. By abstracting the complexities of hardware and low-level APIs—including OpenGL ES, OpenCL, Metal, and WebGPU—ML Drift enables developers to build real-time, interactive ML experiences, from sophisticated video effects to generative AI, across mobile and desktop platforms.
EmbeddingGemma 2 is a new 740M open-weight multimodal model that maps text, images, video, and audio into a unified vector space for privacy-first, on-device retrieval. Developers can easily integrate these capabilities cross-platform using MediaPipe Tasks or optimize fine-grained performance across CPU, GPU, and NPU accelerators with LiteRT. The model enables ultra-low-latency local solutions like search-as-you-type media retrieval, keyframe video moments finding, and zero-shot intent routing.
EmbeddingGemma 2 is a compact, open-source multimodal embedding model that maps text, code, images, video, and audio into a unified 768-dimensional space. Developers can use the sentence-transformers library to selectively load modular modality encoders—ranging from 270M to 740M parameters—to optimize memory usage. Additionally, Matryoshka Representation Learning enables dynamic dimension truncation down to 128d, significantly reducing vector database storage requirements while maintaining high
(Alexandra C. Ribeiro/fStop/Getty Images) All the sweetness, no downsides? ScienceAlert stories are written, fact-checked, and edited by humans, never generated by AI. Don't miss a story, subscribe here.
<p> Let one AI agent open and drive your other CLI agents </p> <p> <a href="https://www.producthunt.com/products/cli-manager?utm_campaign=producthunt-atom-posts-feed&utm_medium=rss-feed&utm_source=producthunt-atom-posts-feed">Discussion</a> | <a href="https://www.producthunt.com/r/p/1275664?app_id=339">Link</a> </p>
The first version of Microsoft-Decision-1 is based on Qwen3.5-9B, but the next will sport homegrown tech, Redmond reassures
It started off promising enough. Dozens of tiny, colorful birds were drawn to my garden just a few weeks after I mounted a $350 $269 Kiwibit Bird Feeder 2 Pro. "Looks like Great Tit!" read the first AI-driven alert sent to my phone. "Looks like Eurasian Blue Tit!" read another. I chuckled, still waiting for […]
Hi HN, I'm Aravind. We build Velu, a docs platform. Coding agents increasingly read docs before humans do, but most docs sites are hard for them: client-rendered pages, no llms.txt, no markdown versions, soft 404s, pages too big for a context window. This tool scores any docs URL against the open Agent-Friendly Documentation Spec (agentdocsspec.com), using its reference implementation, afdocs. Comments URL: https://news.ycombinator.com/item?id=50030369 Points: 1 # Comments: 0
