arXiv:2604.14237v1 Announce Type: new Abstract: Transistor topology optimization is a critical step in standard cell design, directly dictating diffusion sharing efficiency and downstream routability. However, identifying optimal topologies remains a persistent bottleneck, as conventional exhaustive search methods become computationally intractable with increasing circuit complexity in advanced nodes. This paper introduces TOPCELL, a novel and scalable framework that reformulates high-dimensional topology exploration as a generative task using Large Language Models (LLMs). We employ Group Relative Policy Optimization (GRPO) to fine-tune the model, aligning its topology optimization strategy with logical (circuit) and spatial (layout) constraints. Experimental results within an industrial flow targeting an advanced 2nm technology node demonstrate that TOPCELL significantly outperforms foundation models in discovering routable, physically-aware topologies. When integrated into a state-of-the-art (SOTA) automation flow for a 7nm library generation task, TOPCELL exhibits robust zero-shot generalization and matches the layout quality of exhaustive solvers while achieving an 85.91x speedup.

arXiv:2604.14231v1 Announce Type: new Abstract: Financial crime costs U.S. institutions over $32 billion each year. Although AI tools for fraud detection have become more advanced, their use in real-world systems still faces a major obstacle: many of these models operate as black boxes that cannot provide the transparent, auditable explanations required by regulations such as OCC Bulletin 2011-12 and Federal Reserve SR 11-7. This study makes three main contributions. First, it offers a thorough evaluation of explanation quality across faithfulness (sufficiency and comprehensiveness at k=5, 10, and 15) and stability (Kendall's W across 30 bootstrap samples). XGBoost paired with TreeExplainer achieves near-perfect stability (W=0.9912), while LSTM with DeepExplainer shows weak results (W=0.4962). Second, the paper introduces the SHAP-Guided Adaptive Ensemble (SGAE), which dynamically adjusts per-transaction ensemble weights based on SHAP attribution agreement, achieving the highest AUC-ROC among all tested models (0.8837 held-out; 0.9245 cross-validation). Third, a complete three-architecture evaluation of LSTM, Transformer, and GNN-GraphSAGE on the full 590,540-transaction IEEE-CIS dataset is provided, with GNN-GraphSAGE achieving AUC-ROC 0.9248 and F1=0.6013. All results are mapped directly to OCC, SR 11-7, and BSA-AML regulatory compliance requirements.

arXiv:2604.14232v1 Announce Type: new Abstract: The Spatial-Temporal Graph Attention Network (ST-GAT) framework was created to serve as an explainable GNN-based solution for detecting bank distress early warning signs and for conducting macro-prudential surveillance of the interbank system in the United States. The ST-GAT framework models 8,103 FDIC insured institutions across 58 quarterly snapshots (2010Q1-2024Q2). Bilateral exposures were reconstructed from publicly available FDIC Call Reports using maximum entropy estimation to produce a dynamic directed weighted graph. The framework achieves the highest AUPRC among all GNN architectures (0.939 +/- 0.010), trailing only XGBoost (0.944). Ablation analysis confirms the BiLSTM temporal component contributes +0.020 AUPRC; temporal attention weights exhibit a monotonically decreasing pattern consistent with long-run structural vulnerability weighting. Permutation importance identifies ROA (0.309) and NPL Ratio (0.252) as dominant predictors, consistent with post-mortem analyses of the 2023 regional banking crisis. All data are publicly available FDIC Call Reports and FRED series; all code and results are released.

arXiv:2604.14206v1 Announce Type: new Abstract: This paper proposes a machine learning assisted portfolio optimization framework designed for low data environments and regime uncertainty. We construct a teacher student learning pipeline in which a Conditional Value at Risk (CVaR) optimizer generates supervisory labels, and neural models (Bayesian and deterministic) are trained using both real and synthetically augmented data. The synthetic data is generated using a factor based model with t copula residuals, enabling training beyond the limited real sample of 104 labeled observations. We evaluate four student models under a structured experimental framework comprising (i) controlled synthetic experiments (3 x 5 seed grid), (ii) in-distribution real market evaluation (C2A) and (iii) cross-universe generalization (D2A). In real-market settings, models are deployed using a rolling evaluation protocol where a frozen pretrained model is periodically fine tuned on recent observations and reset to its base state, ensuring stability while allowing limited adaptation. Results show that student models can match or outperform the CVaR teacher in several settings, while achieving improved robustness under regime shifts and reduced turnover. These findings suggest that hybrid optimization learning approaches can enhance portfolio construction in data constrained environments

arXiv:2604.14198v1 Announce Type: new Abstract: Domain reweighting can improve sample efficiency and downstream generalization, but data-mixture optimization for multimodal midtraining remains largely unexplored. Current multimodal training recipes tune mixtures along a single dimension, typically data format or task type. We introduce MixAtlas, a method that produces benchmark-targeted data recipes that can be inspected, adapted, and transferred to new corpora. MixAtlas decomposes the training corpus along two axes: image concepts (10 visual-domain clusters discovered via CLIP embeddings) and task supervision (5 objective types including captioning, OCR, grounding, detection, and VQA). Using small proxy models (Qwen2-0.5B) paired with a Gaussian-process surrogate and GP-UCB acquisition, MixAtlas searches the resulting mixture space with the same proxy budget as regression-based baselines but finds better-performing mixtures. We evaluate on 10 benchmarks spanning visual understanding, document reasoning, and multimodal reasoning. On Qwen2-7B, optimized mixtures improve average performance by 8.5%-17.6% over the strongest baseline; on Qwen2.5-7B, gains are 1.0%-3.3%. Both settings reach baseline-equivalent training loss in up to 2 times fewer steps. Recipes discovered on 0.5B proxies transfer to 7B-scale training across Qwen model families.

arXiv:2604.14176v1 Announce Type: new Abstract: Generalized Category Discovery (GCD) leverages labeled data to categorize unlabeled samples from known or unknown classes. Most previous methods jointly optimize supervised and unsupervised objectives and achieve promising results. However, inherent optimization interference still limits their ability to improve further. Through quantitative analysis, we identify a key issue, i.e., gradient entanglement, which 1) distorts supervised gradients and weakens discrimination among known classes, and 2) induces representation-subspace overlap between known and novel classes, reducing the separability of novel categories. To address this issue, we propose the Energy-Aware Gradient Coordinator (EAGC), a plug-and-play gradient-level module that explicitly regulates the optimization process. EAGC comprises two components: Anchor-based Gradient Alignment (AGA) and Energy-aware Elastic Projection (EEP). AGA introduces a reference model to anchor the gradient directions of labeled samples, preserving the discriminative structure of known classes against the interference of unlabeled gradients. EEP softly projects unlabeled gradients onto the complement of the known-class subspace and derives an energy-based coefficient to adaptively scale the projection for each unlabeled sample according to its degree of alignment with the known subspace, thereby reducing subspace overlap without suppressing unlabeled samples that likely belong to known classes. Experiments show that EAGC consistently boosts existing methods and establishes new state-of-the-art results. Code is available at https://haiyangzheng.github.io/EAGC.

arXiv:2604.14235v1 Announce Type: new Abstract: Fraud detection on graph data can be viewed as a demanding task that requires distinguishing between different types of nodes. Because graph neural networks (GNNs) are naturally suited for processing information encoded in graph form through their message-passing operations, methods based on GNN models have increasingly attracted attention in the fraud detection domain. However, fraud graphs inherently exhibit relation camouflage, high heterophily, and class imbalance, causing most GNNs to underperform in fraud detection tasks. To address these challenges, this paper proposes a Graph-Based Fraud Detection Model with Dual-Path Graph Filtering (DPF-GFD). DPF-GFD first applies a beta wavelet-based operator to the original graph to capture key structural patterns. It then constructs a similarity graph from distance-based node representations and applies an improved low-pass filter. The embeddings from the original and similarity graphs are fused through supervised representation learning to obtain node features, which are finally used by an ensemble tree model to assess the fraud risk of unlabeled nodes. Unlike existing single-graph smoothing approaches, DPF-GFD introduces a frequency-complementary dual-path filtering paradigm tailored for fraud detection, explicitly decoupling structural anomaly modeling and feature similarity modeling. This design enables more discriminative and stable node representations in highly heterophilous and imbalanced fraud graphs. Comprehensive experiments on four real-world financial fraud detection datasets demonstrate the effectiveness of our proposed method.

arXiv:2604.14209v1 Announce Type: new Abstract: As deep neural networks are deployed in safety-critical domains such as autonomous driving and medical diagnosis, stakeholders need explanations that are interpretable but also trustworthy with formal guarantees. Existing XAI methods fall short: heuristic attribution techniques (e.g., LIME, Integrated Gradients) highlight influential features but offer no mathematical guarantees about decision boundaries, while formal methods verify robustness yet remain untargeted, analyzing the nearest boundary regardless of whether it represents a critical risk. In safety-critical systems, not all misclassifications carry equal consequences; confusing a "Stop" sign for a "60 kph" sign is far more dangerous than confusing it with a "No Passing" sign. We introduce ViTaX (Verified and Targeted Explanations), a formal XAI framework that generates targeted semifactual explanations with mathematical guarantees. For a given input (class y) and a user-specified critical alternative (class t), ViTaX: (1) identifies the minimal feature subset most sensitive to the y->t transition, and (2) applies formal reachability analysis to guarantee that perturbing these features by epsilon cannot flip the classification to t. We formalize this through Targeted epsilon-Robustness, certifying whether a feature subset remains robust under perturbation toward a specific target class. ViTaX is the first method to provide formally guaranteed explanations of a model's resilience against user-identified alternatives. Evaluations on MNIST, GTSRB, EMNIST, and TaxiNet demonstrate over 30% fidelity improvement with minimal explanation cardinality.

Researchers have shown that blending quantum computing with AI can dramatically improve predictions of complex, chaotic systems. By letting a quantum computer identify hidden patterns in data, the AI becomes more accurate and stable over time. The method outperformed standard models while using far less memory. This could have big implications for fields like climate science, energy, and medicine.

Financial Times: Sources suggest Anthropic is holding off from a wider Mythos release until it can reliably serve it to customers; Anthropic has suffered outages in recent weeks  —  Dario Amodei met with Susie Wiles despite lawsuits over whether AI lab is a national security threat

<p><em><a href="https://simonwillison.net/guides/agentic-engineering-patterns/">Agentic Engineering Patterns</a> &gt;</em></p> <p>Here's an example of a deceptively short prompt that got a quite a lot of work done in a single shot.</p> <p>First, some background. I send out a <a href="https://simonw.substack.com/">free Substack newsletter</a> around once a week containing content copied-and-pasted from my blog. I'm effectively using Substack as a lightweight way to allow people to subscribe to my blog via email.</p> <p>I generate the newsletter with my <a href="https://tools.simonwillison.net/blog-to-newsletter">blog-to-newsletter</a> tool - an HTML and JavaScript app that fetches my latest content from <a href="https://datasette.simonwillison.net/">this Datasette instance</a> and formats it as rich text HTML, which I can then copy to my clipboard and paste into the Substack editor. Here's a <a href="https://simonwillison.net/2023/Apr/4/substack-observable/">detailed explanation of how that works</a>.</p> <p>I recently <a href="https://simonwillison.net/2026/Feb/20/beats/">added a new type of content</a> to my blog to capture content that I post elsewhere, which I called "beats". These include things like releases of my open source projects, new tools that I've built, museums that I've visited (from <a href="https://www.niche-museums.com/">niche-museums.com</a>) and other external content.</p> <p>I wanted to include these in the generated newsletter. Here's the prompt I ran against the <a href="https://github.com/simonw/tools">simonw/tools</a> repository that hosts my <code>blog-to-newsletter</code> tool, using <a href="https://code.claude.com/docs/en/claude-code-on-the-web">Claude Code on the web</a>.</p> <p><pre>Clone simonw/simonwillisonblog from github to /tmp for reference Update blog-to-newsletter.html to include beats that have descriptions - similar to how the Atom everything feed on the blog works Run it with python -m http.server and use `uvx rodney --help` to test it - compare what shows up in the newsletter with what&#x27;s on the homepage of https://simonwillison.net</pre> This got me the <a href="https://github.com/simonw/tools/pull/268">exact solution</a> I needed. Let's break down the prompt.</p> <blockquote> <p><code>Clone simonw/simonwillisonblog from github to /tmp for reference</code></p> </blockquote> <p>I use this pattern a lot. Coding agents can clone code from GitHub, and the best way to explain a problem is often to have them look at relevant code. By telling them to clone to <code>/tmp</code> I ensure they don't accidentally end up including that reference code in their own commit later on.</p> <p>The <a href="https://github.com/simonw/simonwillisonblog">simonw/simonwillisonblog</a> repository contains the source code for my Django-powered <a href="https://simonwillison.net/">simonwillison.net</a> blog. This includes the logic and database schema for my new "beats" feature.</p> <blockquote> <p><code>Update blog-to-newsletter.html to include beats that have descriptions - similar to how the Atom everything feed on the blog works</code></p> </blockquote> <p>Referencing <code>blog-to-newsletter.html</code> is all I need here to tell Claude which of the 200+ HTML apps in that <code>simonw/tools</code> repo it should be modifying.</p> <p>Beats are automatically imported from multiple sources. Often they aren't very interesting - a dot-release bug fix for one of my smaller open source projects, for example.</p> <p>My blog includes a way for me to add additional descriptions to any beat, which provides extra commentary but also marks that beat as being more interesting than those that I haven't annotated in some way.</p> <p>I already use this as a distinction to decide which beats end up in my site's <a href="https://simonwillison.net/about/#atom">Atom feed</a>. Telling Claude to imitate that saves me from having to describe the logic in any extra detail.</p> <blockquote> <p><code>Run it with python -m http.server and use `uvx rodney --help` to test it - compare what shows up in the newsletter with what's on the homepage of https://simonwillison.net</code></p> </blockquote> <p>Coding agents always work best if they have some kind of validation mechanism they can use to test their own work.</p> <p>In this case I wanted Claude Code to actively check that the changes it made to my tool would correctly fetch and display the latest data.</p> <p>I reminded it to use <code>python -m http.server</code> as a static server because I've had issues in the past with applications that fetch data and break when served as a file from disk instead of a localhost server. In this particular case that may not have been necessary, but my prompting muscle memory has <code>python -m http.server</code> baked in at this point!</p> <p>I described the <code>uvx rodney --help</code> trick in <a href="https://simonwillison.net/guides/agentic-engineering-patterns/agentic-manual-testing/#using-browser-automation-for-web-uis">the agentic manual testing chapter</a>. Rodney is browser automation software that can be installed using <code>uvx</code>, and that has <code>--help</code> output designed to teach an agent everything it needs to know in order to use the tool.</p> <p>I figured that telling Claude to compare the results in the newsletter to the content of my blog's homepage would be enough for it to confidently verify that the new changes were working correctly, since I had recently posted content that matched the new requirements.</p> <p>You can see <a href="https://claude.ai/code/session_01BibYBuvJi2qNUyCYGaY3Ss">the full session here</a>, or if that doesn't work I have an <a href="https://gisthost.github.io/?e906e938100ab42f4d6a932505219324/page-001.html#msg-2026-04-18T00-13-57-081Z">alternative transcript</a> showing all of the individual tool calls.</p> <p>The <a href="https://github.com/simonw/tools/pull/268">resulting PR</a> made exactly the right change. It added an additional UNION clause to the SQL query that fetched the blog's content, filtering out draft beats and beats that have nothing in their <code>note</code> column:</p> <p><div class="codehilite"><pre><span></span><code><span class="p">...</span> <span class="k">union</span><span class="w"> </span><span class="k">all</span> <span class="k">select</span> <span class="w"> </span><span class="n">id</span><span class="p">,</span> <span class="w"> </span><span class="s1">&#39;beat&#39;</span><span class="w"> </span><span class="k">as</span><span class="w"> </span><span class="k">type</span><span class="p">,</span> <span class="w"> </span><span class="n">title</span><span class="p">,</span> <span class="w"> </span><span class="n">created</span><span class="p">,</span> <span class="w"> </span><span class="n">slug</span><span class="p">,</span> <span class="w"> </span><span class="s1">&#39;No HTML&#39;</span><span class="w"> </span><span class="k">as</span><span class="w"> </span><span class="n">html</span><span class="p">,</span> <span class="w"> </span><span class="n">json_object</span><span class="p">(</span> <span class="w"> </span><span class="s1">&#39;created&#39;</span><span class="p">,</span><span class="w"> </span><span class="nb">date</span><span class="p">(</span><span class="n">created</span><span class="p">),</span> <span class="w"> </span><span class="s1">&#39;beat_type&#39;</span><span class="p">,</span><span class="w"> </span><span class="n">beat_type</span><span class="p">,</span> <span class="w"> </span><span class="s1">&#39;title&#39;</span><span class="p">,</span><span class="w"> </span><span class="n">title</span><span class="p">,</span> <span class="w"> </span><span class="s1">&#39;url&#39;</span><span class="p">,</span><span class="w"> </span><span class="n">url</span><span class="p">,</span> <span class="w"> </span><span class="s1">&#39;commentary&#39;</span><span class="p">,</span><span class="w"> </span><span class="n">commentary</span><span class="p">,</span> <span class="w"> </span><span class="s1">&#39;note&#39;</span><span class="p">,</span><span class="w"> </span><span class="n">note</span> <span class="w"> </span><span class="p">)</span><span class="w"> </span><span class="k">as</span><span class="w"> </span><span class="n">json</span><span class="p">,</span> <span class="w"> </span><span class="n">url</span><span class="w"> </span><span class="k">as</span><span class="w"> </span><span class="n">external_url</span> <span class="k">from</span><span class="w"> </span><span class="n">blog_beat</span> <span class="k">where</span><span class="w"> </span><span class="k">coalesce</span><span class="p">(</span><span class="n">note</span><span class="p">,</span><span class="w"> </span><span class="s1">&#39;&#39;</span><span class="p">)</span><span class="w"> </span><span class="o">!=</span><span class="w"> </span><span class="s1">&#39;&#39;</span><span class="w"> </span><span class="k">and</span><span class="w"> </span><span class="n">is_draft</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="mi">0</span> <span class="k">union</span><span class="w"> </span><span class="k">all</span> <span class="p">...</span> </code></pre></div> And it figured out a mapping of beat types to their formal names, presumably derived from the <a href="https://github.com/simonw/simonwillisonblog/blob/2e9d7ebe64da799b3927e61b4f85d98f7e9bc9aa/blog/models.py#L545-L551">Django ORM definition</a> that it read while it was exploring the reference codebase: <div class="codehilite"><pre><span></span><code>const beatTypeDisplay = { release: &#39;Release&#39;, til: &#39;TIL&#39;, til_update: &#39;TIL updated&#39;, research: &#39;Research&#39;, tool: &#39;Tool&#39;, museum: &#39;Museum&#39; }; </code></pre></div> Telling agents to use another codebase as reference is a powerful shortcut for communicating complex concepts with minimal additional information needed in the prompt.</p> <p>Tags: <a href="https://simonwillison.net/tags/ai">ai</a>, <a href="https://simonwillison.net/tags/llms">llms</a>, <a href="https://simonwillison.net/tags/prompt-engineering">prompt-engineering</a>, <a href="https://simonwillison.net/tags/coding-agents">coding-agents</a>, <a href="https://simonwillison.net/tags/ai-assisted-programming">ai-assisted-programming</a>, <a href="https://simonwillison.net/tags/generative-ai">generative-ai</a>, <a href="https://simonwillison.net/tags/agentic-engineering">agentic-engineering</a>, <a href="https://simonwillison.net/tags/github">github</a></p>

Iran attacked the Erbil region in northern Iraq over 400 times during its war with the US and Israel. WELT's Chief Correspondent Ibrahim Naber got exclusive access to the wreckage of Iranian weapons used to strike the Kurdistan region. The cache of damaged weapons includes the Zolfaghar ballistic missile and the Shahed drone, which has become Iran's mainstay weapon during the war. Read the original article on Business Insider

Siladitya Ray / Forbes: What some leaders think of using universal income to mitigate AI-fueled layoffs: Musk calls it the “best way”, OpenAI's policy doc mentions a Public Wealth Fund  —  Topline  —  Elon Musk on Friday touted what he described as “Universal HIGH INCOME” as a solution to deal …

OpenAI CEO Sam Altman. Bloomberg/Getty Images OpenAI lost three executives on Friday. Under applications CEO Fidji Simo, OpenAI has shifted focus by cutting side projects. Anthropic has been gaining traction and challenging OpenAI's market position. OpenAI lost three top executives on Friday. Kevin Weil, who headed OpenAI's scientific research efforts after serving as chief product officer, posted on LinkedIn that his team, OpenAI for Science, is being decentralized into other research teams and that he is leaving the company. Bill Peebles, who headed OpenAI's AI video app Sora, also announced his departure. Although Peebles didn't explain why in his post, OpenAI shut down Sora last month due to cost and compute constraints. "I'm proud of all the sleepless nights before and after the launch this team endured in order to deploy the technology in a responsible way and help steer societal norms," Peebles wrote. An OpenAI spokesperson said the company is unifying its business and product strategy. Prism, an AI workplace for scientists that Weil oversaw, is moving to Codex, OpenAI's AI developer assistant, which is expanding beyond coding. Srinivas Narayanan, OpenAI's chief technology officer for its B2B applications, is also leaving to spend more time with his family, he wrote on LinkedIn. Narayanan's departure is unrelated to the other two, according to a person familiar with the matter. The shakeup comes as OpenAI narrows its focus by cutting back on "side quests" and doubling down on selling to businesses. It's a move spearheaded by its CEO of applications, Fidji Simo, to make the company profitable as it moves towards an IPO. Simo is on medical leave for several weeks. OpenAI has been losing some of its thunder to Anthropic, as its latest releases like Claude Code have been gaining traction with businesses and sparking fears of a 'SaaS-pocalypse.' Anthropic has seen funding offers valuing it at up to $800 billion, Business Insider reported, more than twice its most recent valuation in February. OpenAI was valued at $852 billion in a funding round it announced last month. Have a tip? Contact Charles Rollet via email at crollet@businessinsider.com or on Signal and WhatsApp at 628-282-2811. Use a personal email address, a nonwork WiFi network, and a nonwork device; here's our guide to sharing information securely. Read the original article on Business Insider

<p>This year's <a href="https://us.pycon.org/2026/">PyCon US</a> is coming up next month from May 13th to May 19th, with the core conference talks from Friday 15th to Sunday 17th and tutorial and sprint days either side. It's in Long Beach, California this year, the first time PyCon US has come to the West Coast since Portland, Oregon in 2017 and the first time in California since Santa Clara in 2013.</p> <p>If you're based in California this is a great opportunity to catch up with the Python community, meet a whole lot of interesting people and learn a ton of interesting things.</p> <p>In addition to regular PyCon programming we have two new dedicated tracks at the conference this year: an <a href="https://us.pycon.org/2026/tracks/ai/">AI track</a> on Friday and a <a href="https://us.pycon.org/2026/tracks/security/">Security track</a> on Saturday.</p> <p>The AI program was put together by track chairs Silona Bonewald (CitableAI) and Zac Hatfield-Dodds (Anthropic). I'll be an in-the-room chair this year, introducing speakers and helping everything run as smoothly as possible.</p> <p>Here's <a href="https://us.pycon.org/2026/schedule/talks/#May15">the AI track schedule</a> in full:</p> <ul> <li>11:00: <a href="https://us.pycon.org/2026/schedule/presentation/105/">AI-Assisted Contributions and Maintainer Load</a> - Paolo Melchiorre</li> <li>11:45: <a href="https://us.pycon.org/2026/schedule/presentation/66/">AI-Powered Python Education : Towards Adaptive and Inclusive Learning</a> - Sonny Mupfuni</li> <li>12:30: <a href="https://us.pycon.org/2026/schedule/presentation/23/">Making African Languages Visible: A Python-Based Guide to Low-Resource Language ID</a> - Gift Ojeabulu</li> <li>2:00: <a href="https://us.pycon.org/2026/schedule/presentation/138/">Running Large Language Models on Laptops: Practical Quantization Techniques in Python</a> - Aayush Kumar JVS</li> <li>2:45: <a href="https://us.pycon.org/2026/schedule/presentation/126/">Distributing AI with Python in the Browser: Edge Inference and Flexibility Without Infrastructure</a> - Fabio Pliger</li> <li>3:30: <a href="https://us.pycon.org/2026/schedule/presentation/110/">Don't Block the Loop: Python Async Patterns for AI Agents</a> - Aditya Mehra</li> <li>4:30: <a href="https://us.pycon.org/2026/schedule/presentation/81/">What Python Developers Need to Know About Hardware: A Practical Guide to GPU Memory, Kernel Scheduling, and Execution Models</a> - Santosh Appachu Devanira Poovaiah</li> <li>5:15: <a href="https://us.pycon.org/2026/schedule/presentation/101/">How to Build Your First Real-Time Voice Agent in Python (Without Losing Your Mind)</a> - Camila Hinojosa Añez, Elizabeth Fuentes</li> </ul> <p>(And here's <a href="https://gisthost.github.io/?dab27f61d85eb98f60db5991aa21ec89">how I scraped that as a Markdown list</a> from the schedule page using Claude Code and <a href="https://github.com/simonw/rodney">Rodney</a>.)</p> <h4 id="you-should-come-to-pycon-">You should come to PyCon US!</h4> <p>I've been going to PyCon for over twenty years now - I first went <a href="https://simonwillison.net/2005/Mar/28/pycon/">back in 2005</a>. It's one of my all-time favourite conference series. Even as it's grown to more than 2,000 attendees PyCon US has remained a heavily community-focused conference - it's the least <em>corporate</em> feeling large event I've ever attended.</p> <p>The talks are always great, but it's the add-ons around the talks that really make it work for me. The <a href="https://us.pycon.org/2026/events/lightning-talks/">lightning talks</a> slots are some of the most heavily attended sessions. The PyLadies auction is always deeply entertaining. The sprints are an incredible opportunity to contribute directly to projects that you use, coached by their maintainers.</p> <p>In addition to scheduled talks, the event has <strong>open spaces</strong>, where anyone can reserve space for a conversation about a topic - effectively PyCon's version of an <a href="https://en.wikipedia.org/wiki/Unconference">unconference</a>. I plan to spend a lot of my time in the open spaces this year - I'm hoping to join or instigate sessions about both <a href="https://datasette.io/">Datasette</a> and <a href="https://simonwillison.net/guides/agentic-engineering-patterns/">agentic engineering</a>.</p> <p>I'm on the board of the Python Software Foundation, and PyCon US remains one of our most important responsibilities - in the past it's been a key source of funding for the organization, but it's also core to our mission to "promote, protect, and advance the Python programming language, and to support and facilitate the growth of a diverse and international community of Python programmers".</p> <p><small>If you do come to Long Beach, we'd really appreciate it if you could book accommodation in the official hotel block, for reasons <a href="https://pyfound.blogspot.com/2026/04/pycon-us-2026-hotels.html">outlined in this post on the PSF blog</a>.</small></p> <p>Tags: <a href="https://simonwillison.net/tags/conferences">conferences</a>, <a href="https://simonwillison.net/tags/open-source">open-source</a>, <a href="https://simonwillison.net/tags/pycon">pycon</a>, <a href="https://simonwillison.net/tags/python">python</a>, <a href="https://simonwillison.net/tags/ai">ai</a>, <a href="https://simonwillison.net/tags/psf">psf</a></p>

A commercial vessel seen sailing toward the Strait of Hormuz amid the ongoing ceasefire. Shadi J. H. Alassar/Anadolu/Getty Images The Strait of Hormuz is "completely open" to commercial traffic, Iran announced Friday. Real-time data showed a small number of vessels sailing through the waterway, though others turned around. Tehran appears to be steering ships through its waters, and a US Navy blockade remains in place. Iran announced Friday morning that the Strait of Hormuz is open for maritime traffic, and mid-afternoon data showed that commercial vessels were passing through the strategic waterway, though only a small number. Friday evening, a large group of vessels sailed toward the strait but inexplicably turned away, ship-tracking data shows, raising questions about the status of this critical chokepoint. Opening the strait has been a central demand since the start of a fragile ceasefire earlier this month and amid the ongoing negotiations between Washington and Tehran. Seyed Abbas Araghchi, Iran's foreign minister, said in a statement on social media Friday that "the passage for all commercial vessels through Strait of Hormuz is declared completely open for the remaining period of ceasefire, on the coordinated route as already announced by Ports and Maritime Organization of the Islamic Rep. of Iran." The announcement follows a ceasefire between Israel and Lebanon. Iran's coordinated route appears to refer to previously released guidance that takes ships through the northern part of the waterway closest to Iran, which has said that the route was determined by security needs. Separate from Iranian considerations, US forces have been working to establish safe passage in the strait, where mines continue to represent a potential threat. Shipping organizations have had mixed reactions. Some welcomed the news as a positive signal, while others urged caution. It is unclear how much has changed with Friday's announcement from Iran. President Donald Trump acknowledged Iran's Friday statement in a Truth Social post, writing that "Iran has just announced that the Strait of Iran is fully open and ready for passage." The market reaction to the announcement saw stocks climb higher and oil fall. The strategic Strait of Hormuz is a narrow chokepoint linking the Persian Gulf to the Gulf of Oman and a major artery for global shipments of oil and natural gas. MarineTraffic data from Friday afternoon showed that just over half a dozen commercial vessels had transited the strait since the announcement. Data from KPler, a real-time trade intelligence platform, showed that eight commercial vessels had crossed as of 2 pm EST Friday compared to five on Thursday. The data refers to commercial fleets of crude tankers, LPG, LNG, and dry bulk vessels. Containerships are not included in the count. But by the evening, new data showed ships were turning back. The limited number of ships that have gone through is a fraction of the pre-war normal. Before the war, the strait saw an average of over 120 transits daily. Earlier in the week, days before the announcement on the strait, US Navy warships set up a blockade of Iranian ships and ports. Trump said the blockade will remain in effect until Washington and Tehran reach a deal to end the war. US Air Force Gen. Dan Caine, chairman of the Joint Chiefs of Staff, has highlighted the role of the Navy's Arleigh Burke-class destroyers in the blockade during a briefing Thursday but didn't specify numbers involved. Dozens of surveillance aircraft, refueling planes, and drones are also involved in the blockade. US Central Command, which oversees operations in the Middle East, said Friday that 19 vessels have turned around to comply with blockade restrictions since it began on Monday. The US set up the blockade amid a tenuous ceasefire between the US and Iran, which was reached after more than a month of war. US military leaders have asserted that American forces remain poised to resume combat operations should the negotiations fail. Read the original article on Business Insider

Coding agents are starting to write production code at scale. Stripe’s agents generate 1,300+ PRs per week. Ramp attributes 30% of merged PRs to agents.... Coding agents are starting to write production code at scale. Stripe’s agents generate 1,300+ PRs per week. Ramp attributes 30% of merged PRs to agents. Spotify reports 650+ agent-generated PRs per month. Tools like Claude Code and Codex make hundreds of API calls per coding session, each carrying the full conversation history. Behind every one of these workflows is an inference stack under… Source

World, which has raised eyebrows (but also a lot of interest) with its Orb-centered anonymous verification project, is looking to expand its influence via a bevy of new partnerships.

Financial Times: Sources: Recursive Superintelligence, a four-month-old start-up developing self-teaching AI and founded by ex-DeepMind and OpenAI engineers, has raised $500M+  —  Group founded by former engineers at DeepMind and OpenAI secures $4bn valuation in deal with Google's venture arm and Nvidia

Axios: White House says a meeting between Chief of Staff Susie Wiles and Dario Amodei had been “productive and constructive”; source: Scott Bessent joined the meeting  —  Treasury Secretary Scott Bessent joined a meeting on Friday between White House Chief of Staff Susie Wiles …