AI News Today

<h2> Why Prompts Matter More Than You Think </h2> <p>The difference between a great AI response and a mediocre one isn't always the model. It's the prompt.</p> <p>Experience this: You ask ChatGPT a vague question and get a vague answer. You ask the same AI a perfectly crafted prompt and get something incredible.</p> <p>The skill gap is massive. Companies are paying prompt engineers $150K+ because mastering prompts directly impacts:</p> <ul> <li>Response quality</li> <li>Token usage (costs)</li>

<pre style='white-space:pre-wrap;width:81ex'>fix(ci): preserve AI runtime build artifacts</pre>

https://www.vaibot.io/blog/prompt-injection-is-an-egress-pro... Comments URL: https://news.ycombinator.com/item?id=48841473 Points: 1 # Comments: 1

<blockquote> <p>Canvas-native UI runtime with a Virtual Math Tree, semantic accessibility projection, WebGL/WebGPU backends, and agent-drivable controls. </p> </blockquote> <p><strong>VectoJS is an open-source TypeScript UI framework that renders an entire application to a single <code><canvas></code> element instead of the DOM.</strong> Every UI framework of the last decade has made a different bet: the DOM is the substrate, and the framework's job is to manage it faster, or hide it behind a ni

Article URL: https://thetruthasiseeitnow.com/ai-slop-starts-with-the-codebase-itself/ Comments URL: https://news.ycombinator.com/item?id=48841446 Points: 34 # Comments: 35

<p>Every "best AI for coding" article is secretly a personality quiz for the author. They already have a favorite, they feed it a softball, it hits the softball, and — shocking — their favorite wins.</p> <p>So I did the annoying version instead: I took the <em>same</em> real coding tasks and ran them through <strong>ChatGPT (GPT-5.2), Claude (Opus 4.8), Gemini (3.1 Pro), and Grok 4</strong> at the same time, side by side, and read the answers next to each other. Not benchmarks. Actual "I need th

<pre style='white-space:pre-wrap;width:81ex'>fix(telegram): keep DM topic auto-rename user message UTF-16 safe (#101781) * fix(telegram): keep DM topic auto-rename user message UTF-16 safe Add surrogate-boundary regression for auto-topic label input truncation. Co-authored-by: Cursor <cursoragent@cursor.com> * test(telegram): tighten topic label boundary proof --------- Co-authored-by: NIO <nocodet@mail.com> Co-authored-by: Cursor <cursoragent@cursor.com> Co-authored-by: Peter Steinberger <steip

<p>On Day 1, PostAll was a script that called the OpenAI API and dumped the output into a text file. On Day 100, it's a platform with a formatting engine, a three-part quality gate, and CMS integrations for WordPress, Ghost, and Webflow.</p> <p>I didn't plan most of that. I backed into almost all of it, one broken assumption at a time.</p> <p>This is the retrospective I've been putting off writing, because it means admitting how much of the last 100 days was me being confidently wrong about some

Article URL: https://www.axios.com/2026/07/08/imf-ai-energy-iran Comments URL: https://news.ycombinator.com/item?id=48841396 Points: 4 # Comments: 1

<p>Six months ago, I got a content pipeline running end to end.</p> <p>Topic mining, keyword filtering, outline generation, LLM writing, auto-publishing — fully automated. In theory, I could produce dozens of articles a day, target long-tail keywords, and slowly buildup search traffic.</p> <p>In the first week, I generated 200 articles. Then I sat down and read ten of them.</p> <p>By the third one, I understood the problem: <strong>they all sounded like the same person wrote them. And that perso

<p>No-code SaaS consistently surprises founders because the barrier to a real, working product is genuinely gone now. Here's the stack:<br> Validation: Claude — before touching a builder, describe the idea and ask Claude to steelman the 5 biggest reasons it fails. Cheapest possible filter.<br> Building: Bubble.io — real database, real auth, real workflows, no code. Free tier is enough to validate and launch a basic version.<br> AI features: OpenAI API via Bubble plugin — the unlock most no-code

<pre style='white-space:pre-wrap;width:81ex'>fix(security): keep channel-metadata and install-policy truncation surrogate-safe (#102266) * fix(security): keep channel-metadata and install-policy truncation surrogate-safe channel-metadata.ts and install-policy.ts both had private truncateText functions using String.prototype.slice(0, N) on user-visible text. channel-metadata is explicitly user-controlled untrusted content that gets injected into LLM prompt context -- a dangling surrogate from uns

Comments

Article URL: https://www.anthropic.com/research/off-switch-dual-use Comments URL: https://news.ycombinator.com/item?id=48841308 Points: 2 # Comments: 0

Article URL: https://www.irishtimes.com/world/europe/2026/07/08/ai-software-that-generates-rage-bait-developed-by-germanys-far-right-afd/ Comments URL: https://news.ycombinator.com/item?id=48841275 Points: 10 # Comments: 0

Ask an AI coding agent to scan open-source code for security holes, and it might run the attacker's code on your own machine instead. That is the finding in a proof-of-concept published Wednesday by the AI Now Institute, an attack it calls "Friendly Fire." It works against Anthropic's Claude Code and OpenAI's Codex when either is running in an autonomous mode that approves its own

It's like peering back through time to the 17th century. ScienceAlert stories are written, fact-checked, and edited by humans, never generated by AI. Don't miss a story, subscribe here.

<pre style='white-space:pre-wrap;width:81ex'>fix(agents): keep chunkString and buildResumeMessage truncation UTF-16 safe (#102085) * fix(agents): keep chunkString and buildResumeMessage truncation UTF-16 safe * fix: preserve code points in agent output chunks --------- Co-authored-by: Peter Steinberger <steipete@gmail.com></pre>

Article URL: https://www.cerenovus.ai/ Comments URL: https://news.ycombinator.com/item?id=48841133 Points: 1 # Comments: 1

Article URL: https://mendelevium.github.io/the-policy-never-shipped/ Comments URL: https://news.ycombinator.com/item?id=48841132 Points: 1 # Comments: 0

<pre style='white-space:pre-wrap;width:81ex'>[AI] fix(memory): use truncateUtf16Safe for dreaming snippet truncation (#101946) * [AI] fix(memory): use truncateUtf16Safe for dreaming snippet truncation Replace .slice(0, N) with truncateUtf16Safe() at 5 call sites in dreaming-phases.ts so complex emoji and surrogate pairs near the truncation boundary are not split into lone surrogates. truncateUtf16Safe is the standard SDK helper, already used in session-cost-usage, cron, exec-approval, and node-h

<pre style='white-space:pre-wrap;width:81ex'>Suppress failed tool progress in Discord (#92517) * fix(discord): suppress failed tool progress Co-authored-by: davefano-agents@teal-03-mst-m2u-wheeljack <davefano-agents@users.noreply.github.com> * chore: leave changelog to release --------- Co-authored-by: Peter Steinberger <steipete@gmail.com> Co-authored-by: davefano-agents@teal-03-mst-m2u-wheeljack <davefano-agents@users.noreply.github.com></pre>

Researchers at Wiz found that a flaw in six popular AI coding assistants lets a booby-trapped code project quietly take control of a developer's computer. The assistant asks permission to edit one harmless-looking file, but the write lands on a sensitive one instead. The affected tools are Amazon Q Developer, Anthropic's Claude Code, Augment, Cursor, Google Antigravity, and Windsurf.

<pre style='white-space:pre-wrap;width:81ex'>fix(acp): keep session update text truncation surrogate-safe (#102378) * fix(acp): keep session update text truncation surrogate-safe acp-projector's truncateText function used String.prototype.slice(0, N) on session update text that is sent to AI providers in the prompt context. When the truncation boundary falls inside a UTF-16 surrogate pair (emoji, CJK extended), the resulting string contains a dangling surrogate that can corrupt the prompt. Repla

Downer for doomers as conservative investor plans to lift portfolio exposure by 150 percent, and even more once it adds electrification ventures

arXiv:2607.07379v1 Announce Type: new Abstract: In agentic scientific machine learning (SciML), large language model (LLM) agents can discover surrogate models and select one by an automated score, typically an error metric. A low error, however, does not establish that the predicted fields satisfy the physics that matter for mechanics, such as boundary conditions, superposition, stiffness scaling, or causality. We introduce Physics-Audited Agentic SciML (PA-SciML), a verification-first workflow

arXiv:2607.06624v1 Announce Type: new Abstract: We present AgentLens, a production-assessed benchmark for interactive code agents. Most code-agent benchmarks reduce a run to a single bit -- did the task pass? -- but the people who actually use these agents experience the entire trajectory: how the agent follows instructions, uses its tools, verifies its own work, recovers from mistakes, and talks to them along the way. AgentLens evaluates that whole trajectory. It pairs formal verification, wher

arXiv:2607.06720v1 Announce Type: new Abstract: Training large language models (LLMs) with extended reasoning has enabled in-context search, in which models iteratively generate, critique, and revise solution attempts. We provide a theoretical analysis of in-context search by modeling it as approximate inference over reasoning traces, where the base model defines a prior and self-reflection provides feedback for posterior updates, and study the resulting inference-time sampling complexity - the

arXiv:2607.06757v1 Announce Type: new Abstract: Agent-based modeling (ABM) has the capability to model millions of individuals and their interactions, which is useful for policy making. However, ABMs have traditionally relied on static prior, which prevents the models from adapting to real-time changes. Our research provides a novel approach to addressing this information gap. Large language models (LLMs) offer new opportunities to predict human decision-making. Here, we introduce a scalable Hyb

arXiv:2607.06760v1 Announce Type: new Abstract: Autonomous systems under partial observability act on beliefs, not raw sensor events. QANTIS treats the quantum processor as a calibrated belief-update service in that loop: it receives a prior and an observation model, estimates the rare-event evidence term, and returns an ordinary posterior to a classical planner. This paper asks whether that service can be reused across a sequential Tiger POMDP horizon on present IBM Heron hardware without corru

arXiv:2607.06764v1 Announce Type: new Abstract: Recent progress on ARC-AGI-1 from disclosed architectures has come broadly from two regimes: heavy test-time compute over frontier models (evolutionary search, exhaustive sampling, extended chain-of-thought), or benchmark-specific training in which small models are fine-tuned on ARC data, often with task-specialized architectures. We study a third regime: an open-weight model in non-thinking mode (DeepSeek V3.2) under a strict budget, with no ARC-s

arXiv:2607.06820v1 Announce Type: new Abstract: Recent advances in AI for Mathematics have focused largely on autoformalization and theorem proving, leaving the role of Computer Algebra Systems (CAS) in agentic LLM workflows underexplored. We propose a ReAct-style agentic setup that combines LLM reasoning with verifiable feedback from SageMath, together with Context7 for the up-to-date documentation. We evaluate this agentic setup across frontier models for solving research-level mathematical pr

arXiv:2607.06906v1 Announce Type: new Abstract: Agentic AI development today runs on token maxing: buying capability with tokens -- longer reasoning traces, more turns, wider tool payloads, bigger replayed contexts -- so tokens per task grow faster than task value. Falling per-token prices mask the pattern; total spend rises anyway. We argue the decisive lever against token maxing is the harness: the orchestration layer that assembles context, exposes tools, sequences turns, delegates work, and

arXiv:2607.06993v1 Announce Type: new Abstract: Customer behavior modeling underpins recommendation, marketing, and decision support, yet existing approaches either optimize predictive accuracy without explaining decisions or simulate users without grounding them in real behavioral data. We present the Large Behavioral Model (LBM) that learns customer decision making directly from large-scale retail transactions through a unified Person-Environment formulation. Customer state is represented by a

arXiv:2607.07021v1 Announce Type: new Abstract: Humans continuously coordinate with others in dynamic interactions, often through implicit, hard-to-quantify social norms that act as shared tacit expectations among interacting agents. As AI agents, including large language models (LLMs), become embedded in daily life, they increasingly participate in such interactions and reshape social interaction structures. Yet they often fail to coordinate with humans in an effective, considerate, and natural

arXiv:2607.07097v1 Announce Type: new Abstract: Safety evaluations of multi-agent LLM systems often compare a direct prompt with a planner-executor pipeline and report the difference as a single "pipeline effect." We argue that this aggregate is difficult to interpret because it conflates three mechanisms: harmful intent may be reframed as plausible operational work, the planner may refuse or transform the request, and the executor may act under delegation prompts implying prior approval. To sep

arXiv:2607.07189v1 Announce Type: new Abstract: Vision-language models (VLMs) and agentic AI have shown strong performance on semantic visual tasks, but it remains unclear whether they can handle the physics and inverse problems that underlie computational imaging. We present ImagingBench, a benchmark of 20 computational imaging tasks spanning five categories: ray and wave optics, image signal processing, inverse reconstruction, computational sensing, and calibration. ImagingBench evaluates thre

arXiv:2607.07229v1 Announce Type: new Abstract: Prior work has shown that chain-of-thought (CoT) reasoning is often unfaithful: a model's stated reasoning does not reliably reflect the process that produced its output. Detecting unfaithfulness, though, requires controlled experimental interventions, which cannot be applied to evaluation transcripts after the fact. We turn instead to a more tractable question that has received less attention: whether the stated reasoning is logically consistent w

arXiv:2607.07321v1 Announce Type: new Abstract: Tool utilization enables Large Language Model (LLM) agents to interact with the real world and resolve complex tasks. However, existing agent frameworks predominantly rely on static toolsets composed of granular atomic actions (e.g., basic file I/O or single-turn search), which forces agents to reinvent low-level logic for every recurring workflow, leading to increased reasoning overhead and failure rates. In this study, we propose that agents can

arXiv:2607.07397v1 Announce Type: new Abstract: Autonomous agents promise substantial gains in speed, scale, and labor efficiency, but their failures can impose abrupt and often irreversible costs. The central challenge for agentic automation is therefore to increase the benefits of automation while bounding the consequences of failure. While databases remain central to modern computing, agents operate over a broader data environment spanning files, APIs, applications, and system state. In this