AI News Today

<h1> Ringkasan Tren GitHub Minggu Ini: Efisiensi, Pemrosesan Panjang, dan Evolusi Agentik </h1> <p>Minggu ini, lanskap open source menunjukkan pergeseran yang menarik. Fokus utama tidak lagi hanya pada kecerdasan buatan generatif yang "pintar", melainkan pada bagaimana alat-alat tersebut dapat bekerja lebih efisien, memproses data dalam skala makro, dan berkolaborasi secara otonom. Dari optimasi kode hingga pemrosesan optik karakter (OCR) jarak jauh, lima repository yang naik daun mewakili pilar

<!-- SC_OFF --><div class="md"><p><a href="https://www.gnosyslabs.com/case-studies/safety-classifier-sparse-labels">https://www.gnosyslabs.com/case-studies/safety-classifier-sparse-labels</a></p> <p><strong>Gnosys is an autonomous model engineer: it improves prompts and classifiers when ground truth is too sparse for conventional optimization. On ToxicChat, a public safety benchmark, under realistic label scarcity, it improved a classifier past both the team's starting point and GEPA (a standard

<p>They’re making an art of stealing intellectual property</p><ul><li><p><strong>See more of </strong><a href="https://www.theguardian.com/profile/fiona-katauskas"><strong>Fiona Katauskas’s cartoons here</strong></a></p></li></ul> <a href="https://www.theguardian.com/commentisfree/picture/2026/jul/02/are-ai-companies-getting-away-with-crime">Continue reading...</a>

<p>Every few months, a new AI model is released, and the same question comes up:</p> <p><strong>"Will AI replace software developers?"</strong></p> <p>My answer is simple: <strong>No, but it will change what it means to be a developer.</strong></p> <h2> AI Is Already Changing the Way We Work </h2> <p>Today's AI tools can write code, explain complex functions, generate tests, fix bugs, and even review pull requests. They're incredibly useful, and they've made developers more productive than ever.

<pre style='white-space:pre-wrap;width:81ex'>fix(discord): guard JSON.parse against malformed API response bodies (#97889) * fix(discord): guard JSON.parse against malformed API response bodies Wrap JSON.parse(text) in requestDiscord with try/catch to prevent a malformed Discord API response body from throwing an unhandled SyntaxError and crashing the process. On parse failure, throw a DiscordApiError with a descriptive message so the caller can handle it gracefully. Co-Authored-By: Claude <nore

Article URL: https://makerspet.com/blog/building-an-open-source-robot-vacuum-meet-oomwoo/ Comments URL: https://news.ycombinator.com/item?id=48755005 Points: 44 # Comments: 2

<pre style='white-space:pre-wrap;width:81ex'>fix(agents): don't inject A2A turns into isolated-cron sessions_send (#92257) (#92283) * fix(agents): don't inject A2A turns into isolated-cron sessions_send (#92257) Fire-and-forget sessions_send (timeoutSeconds === 0) with announce delivery runs the A2A ping-pong loop. For a cross-session send (requester != target) the loop's first iteration feeds the target agent's reply back into the requester session as a new turn. For a normal requester that rou

Article URL: https://werd.io/i-have-a-theory-about-ai-fake-news-site-the-editorial/ Comments URL: https://news.ycombinator.com/item?id=48754929 Points: 1 # Comments: 0

<p> Shared, searchable memory for every AI coding agent </p> <p> <a href="https://www.producthunt.com/products/scritty?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/1185930?app_id=339">Link</a> </p>

<pre style='white-space:pre-wrap;width:81ex'>docs(telegram): move maintainer decisions into scoped AGENTS.md</pre>

AI's externalities are growing faster than the industry can address them

Article URL: https://www.agentsessions.dev/ Comments URL: https://news.ycombinator.com/item?id=48754784 Points: 2 # Comments: 1

Sam Altman spent a year pitching the White House on taking a piece of OpenAI. The offer is now on the table, and it volunteers his rivals too. Add Fable 5's return, paid for in oversight concessions, and a pattern is hard to miss: this was the quarter the government stopped watching frontier AI from across the street and got a desk inside. Below: the security mess in agentic IDEs, courts and agencies improvising faster than legislators, and our new section, The Split — the week's biggest story a

Multi-agent LLM systems are increasingly deployed as autonomous collaborators, where agents interact freely rather than execute fixed, pre-specified workflows. In such settings, effective coordination cannot be fully designed in advance and must instead emerge through interaction. However, most prior work enforces coordination through fixed roles, workflows, or aggregation rules, leaving open the question of how well self-organizing teams perform when coordination is unconstrained. Drawing on or

Maximum inner product search (MIPS) is a crucial subroutine in machine learning, requiring the identification of a vector taken within a database (the keys) that best aligns with a given query. We propose amortized MIPS: a regression-based approach that trains neural networks to directly predict MIPS solutions, amortizing the cost of repeatedly solving MIPS for queries drawn from a known distribution over a fixed key database. Our key insight is that the MIPS value function is the support functi

Reinforcement learning (RL) finetuning has become a key technique for enhancing large language models (LLMs) on reasoning-intensive tasks, motivating its extension to vision language models (VLMs). While RL-tuned VLMs improve on visual reasoning benchmarks, they remain vulnerable to weak visual grounding, hallucinations, and over-reliance on textual cues. We show that simple, controlled textual perturbations—misleading captions or incorrect chain-of-thought (CoT) traces—cause substantial drops i

Understanding how transformer components operate in LLMs is important, as it is at the core of recent technological advances in artificial intelligence. In this work, we revisit the challenges associated with interpretability of feed-forward modules (FFNs) and propose MemoryLLM, which aims to decouple FFNs from self-attention and enables us to study the decoupled FFNs as context-free token-wise neural retrieval memory. In detail, we investigate how input tokens access memory locations within FFN

Large language models can exhibit emergent reasoning behaviors, often manifested as recurring lexical patterns (e.g., “wait,” indicating verification). However, complex reasoning trajectories remain sparse in unconstrained sampling, and standard RL often fails to guarantee the acquisition of diverse reasoning behaviors. We propose a systematic discovery and reinforcement of diverse reasoning patterns through structured reasoning, a paradigm that requires targeted exploration of specific reasonin

Diffusion (Large) Language Models (dLLMs) now match the downstream performance of their autoregressive counterparts on many tasks, while holding the promise of being more efficient during inference. One critical design aspect of dLLMs is the sampling procedure that selects which tokens to unmask at each diffusion step. Indeed, recent work has found that heuristic strategies such as confidence thresholding improve both sample quality and token throughput compared to random unmasking. However, suc

Diffusion Large Language Models (dLLMs) have emerged as a promising alternative to purely autoregressive language models because they can decode multiple tokens in parallel. However, state-of-the-art block-wise dLLMs rely on a “remasking” mechanism that decodes only the most confident tokens and discards the rest, effectively wasting computation. We demonstrate that recycling computation from the discarded tokens is beneficial, as these tokens retain contextual information useful for subsequent

Reasoning Large Language Models (LLMs) enable test-time scaling, with dataset-level accuracy improving as the token budget increases, motivating adaptive reasoning—spending tokens when they improve reliability and stopping early when additional computation is unlikely to help. However, setting the token budget, as well as the threshold for adaptive reasoning, is a practical challenge that entails a fundamental risk-accuracy trade-off. We re-frame the budget setting problem as risk control, limit

The problem of domain generalization concerns learning predictive models that are robust to distribution shifts when deployed in new, previously unseen environments. Existing methods typically require labeled data from multiple training environments, limiting their applicability when labeled data are scarce. In this work, we study domain generalization in an anti-causal setting, where the outcome causes the observed covariates. Under this structure, environment perturbations that affect the cova

Comments

Article URL: https://github.com/authsec-ai/authsec-ai Comments URL: https://news.ycombinator.com/item?id=48754628 Points: 1 # Comments: 0

Article URL: https://blog.akring.com/posts/fomo-is-the-cyberpsychosis-of-the-ai-era/ Comments URL: https://news.ycombinator.com/item?id=48754421 Points: 6 # Comments: 1

<p> Documentation that works for both humans and AI systems </p> <p> <a href="https://www.producthunt.com/products/docsalot-2?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/1185912?app_id=339">Link</a> </p>

I've been suckered into running claude code and codex independently and vaguely scanning the output before moving on. I don't love the code-base familiarity that produces. Anybody figure out a neat harness setup where one goes file by file, method by method with the agent? Looking for suggestions and what the SOTA is for this sort of thing. Comments URL: https://news.ycombinator.com/item?id=48754327 Points: 5 # Comments: 4

<p>AI-generated overview found to gloss over allegations of sexual harassment and describes hotel being sued over hygiene as ‘spotless’</p><p>A hotel being sued for mass food poisonings was described as “spotless” and a resort where guests complained of sexual harassment by staff was praised for “friendly” service by an AI intended to summarise millions of Tripadvisor reviews.</p><p>The overviews of customer feedback downplayed serious complaints, ranging from the stench of mould to a lack of ma

Article URL: https://pypi.org/project/toolnexus/ Comments URL: https://news.ycombinator.com/item?id=48754258 Points: 2 # Comments: 0

I've been a dev for close to 9 years now, always spending my time inside VSCode. When the AI extension came in, I started writing less code, reading more. When Claude Desktop app came in, I did not like it, stuck with the VSCode extension. Then the app got better, and the extension got worse. so I switched my focus to that instead of the code editor. I now view the diffs in the Claude app instead of VSCode, and only switch to VSCode for a more thorough review. Now VSCode ships with a new view, s

Hi HN! Over a year ago, we launched Banto as a party games website (with 4 game types). Since then, we’ve pivoted and made the idea a bit bigger. Now, anyone can generate and host their own live game room with just a prompt. From that, Banto will turn it into an interactive room people can join from their phone/laptop. This currently is our latest iteration but we’d love feedback on the product, positioning, and whether the “turn content into a live game room / experience” idea is clear at all t

Article URL: https://www.eventual.ai/blog/egodex-scenario-search Comments URL: https://news.ycombinator.com/item?id=48754121 Points: 2 # Comments: 0

Article URL: https://www.avclub.com/weird-al-yankovic-ai-commercial-exit Comments URL: https://news.ycombinator.com/item?id=48754099 Points: 19 # Comments: 0

Fable 5 is back today, baby! Premium subscribers have one week to use it within their subscriptions. First hit’s free. Then you pay by the token.

Article URL: https://blog.google/innovation-and-ai/technology/safety-security/opening-up-zero-knowledge-proof-technology-to-promote-privacy-in-age-assurance/ Comments URL: https://news.ycombinator.com/item?id=48753979 Points: 60 # Comments: 38

Risk factors galore

Fable 5 is back (for a week) but I just noticed that all the Claude models suddenly are failing with all the Tools ( search, fetch, read, write). Anyone else experienced the same? Comments URL: https://news.ycombinator.com/item?id=48753884 Points: 5 # Comments: 1

Article URL: https://www.orgonaut.co/tools/agentic-reorg-simulator/ Comments URL: https://news.ycombinator.com/item?id=48753823 Points: 2 # Comments: 0

I'm curious, has anyone encountered scenarios where Fable wasn't able to solve a problem, fix a bug, complete some task, etc. outside of the (questionable) restrictions in place? Every conversation I read about Fable devolves into either two conclusions: (1) Fable is insane and was able to do X in one or two prompts or fix issues that other models and humans had been grinding on for a while or (2) Fable sucks because my prompt triggered the guardrails. But I haven't seen much of "Fable wasn't ab

AI has partnered with the nation's first president, and the results aren't as bad as you might assume.