<p><strong><a href="https://role-confusion.github.io">Prompt Injection as Role Confusion</a></strong></p>First, I absolutely love this:</p><blockquote><p>This is a blog-style writeup of the paper.</p></blockquote><p>I wish <em>every paper</em> would come with one of these. Academic writing is pretty dry - the impact of a paper can be so much higher if you publish a readable version to accompany the formal one.</p><p>Charles Ye, Jasmine Cui, and Dylan Hadfield-Menell present some fascinating research into the challenge of having models distinguish their own privileged text (here wrapped in role tags like <code>&lt;system&gt;</code>, <code>&lt;think&gt;</code>, and <code>&lt;assistant&gt;</code>) from untrusted user input wrapped in <code>&lt;user&gt;</code>.</p><p>The bad news: they confirm that not only is this not possible, but it looks like models take the <em>style</em> of the text more seriously than the actual text!</p><p>This leads to some very concerning jailbreaks. Take the fo

<p>This morning <a href="https://news.ycombinator.com/item?id=48630171">on Hacker News</a> I saw <a href="https://hustvl.github.io/Moebius/">Moebius: 0.2B Lightweight Image Inpainting Framework with 10B-Level Performance</a>, describing a small but effective inpainting model - a model where you can mark regions of an image to remove and the model imagines what should fill the space. The released model <a href="https://github.com/hustvl/Moebius/blob/9310b76e368f5f7a8ecdf06493231af279c9973b/requirements.txt#L1">required PyTorch and NVIDIA CUDA</a>, but since it described itself as 0.2B I decided to try and get it running using WebGPU in a browser. TL;DR: I got it working, and you can try the demo at <a href="https://simonw.github.io/moebius-web/">simonw.github.io/moebius-web/</a>. Read on for the details.</p><h4 id="the-finished-tool">The finished tool</h4><p>Here's a video demo of the finished tool:</p><videowidth="1280"height="1070"poster="https://static.simonwillison.net/static/2026/

<p><a href="https://sqlite-utils.datasette.io/en/latest/">sqlite-utils</a> is my combined Python library and CLI tool for working with SQLite databases. It provides an extensive set of higher-level operations on top of Python's default <a href="https://docs.python.org/3/library/sqlite3.html">sqlite3 package</a>, including support for <a href="https://sqlite-utils.datasette.io/en/latest/cli.html#transforming-tables">complex table transformations</a>, automatic table creation <a href="https://sqlite-utils.datasette.io/en/latest/cli.html#inserting-json-data">from JSON data</a> and a whole lot more.</p><p>I released <a href="https://sqlite-utils.datasette.io/en/latest/changelog.html#rc1-2026-06-21">sqlite-utils 4.0rc1</a>, the first release candidate for sqlite-utils v4. The major version bump indicates some (minor) backwards incompatible changes, so I'm interested in having people try this out before I commit to a stable release.</p><h4 id="new-feature-migrations">New feature: migrations

<p><strong>Release:</strong> <a href="https://github.com/simonw/sqlite-utils/releases/tag/4.0rc1">sqlite-utils 4.0rc1</a></p> <p>See <a href="https://simonwillison.net/2026/Jun/21/sqlite-utils-40rc1/">sqlite-utils 4.0rc1 adds migrations and nested transactions</a>.</p> <p>Tags: <a href="https://simonwillison.net/tags/sqlite-utils">sqlite-utils</a></p>

<p><strong><a href="https://blog.cloudflare.com/temporary-accounts/">Temporary Cloudflare Accounts for AI agents</a></strong></p>The announcement says this is "for AI agents" but (as is pretty common these days) the AI hook isn't really necessary, this is an interesting feature for everyone else as well.</p><p>Short version: you can now create a Cloudflare Workers project and run this, without even creating a Cloudflare account:</p><pre><code>npx wrangler deploy --temporary</code></pre><p>Cloudflare will deploy the application to a new, ephemeral project which will stay live for 60 minutes.</p><p>I <a href="https://gist.github.com/simonw/264bd6b8a39fc34c91c9c867454c64b9">had GPT-5.5 xhigh</a> in Codex Desktop <a href="https://github.com/simonw/cloudflare-redirect-resolver">build this test application</a> providing a tool for following HTTP redirects and returning the final destination. The temporary deployment worked as advertised.</p><p>Running the deployment spits out the URL to a p

<blockquote cite="https://news.ycombinator.com/item?id=48592163#48593190"><p>The real valuable capability MCP offers over skills/CLI is isolating the auth flow outside of the agent’s context window, and potentially out of the harness completely. [...]</p><p>Maybe the idealized form of MCP is just an auth gateway for the API and nothing else. That’d still be a win.</p></blockquote><p class="cite">&mdash; <a href="https://news.ycombinator.com/item?id=48592163#48593190">Sean Lynch</a>, comment on Hacker News</p> <p>Tags: <a href="https://simonwillison.net/tags/model-context-protocol">model-context-protocol</a>, <a href="https://simonwillison.net/tags/llms">llms</a>, <a href="https://simonwillison.net/tags/ai">ai</a>, <a href="https://simonwillison.net/tags/generative-ai">generative-ai</a>, <a href="https://simonwillison.net/tags/skills">skills</a></p>

<p>Today we launched a new plugin for Datasette, <a href="https://github.com/datasette/datasette-apps">datasette-apps</a>, with <a href="https://datasette.io/blog/2026/datasette-apps/">this launch announcement post</a> on the Datasette project blog. That post has the <em>what</em>, but I'm going to expand on that a little bit here to provide the <em>why</em>.</p><h4 id="the-tl-dr">The TL;DR</h4><p>Datasette Apps are self-contained HTML+JavaScript applications that run in a tightly constrained <code>&lt;iframe&gt;</code> sandbox hosted on your Datasette application. They can use JavaScript to run read-only SQL queries against data in Datasette, and can run write queries too if you configure them <a href="https://datasette.io/blog/2026/sql-write-queries/">with some stored queries</a>.</p><p>Here's a <a href="https://agent.datasette.io/-/apps/01kvdp1d26g8trye3r4gc3yy9c">very simple example</a> and a <a href="https://agent.datasette.io/-/apps/01ktvyaejhk07zskdx2tewxppe">more complex custo

<p><strong>Release:</strong> <a href="https://github.com/datasette/datasette-acl/releases/tag/0.6a0">datasette-acl 0.6a0</a></p> <blockquote><p>This release expands <code>datasette-acl</code> from table-only permissions toward a general resource-sharing system.</p></blockquote><p>Alex Garcia did most of the work for this release - we're fleshing out the plugin that will allow multi-user Datasette instances finely grained control over who can access which resources within Datasette.</p> <p>Tags: <a href="https://simonwillison.net/tags/datasette">datasette</a>, <a href="https://simonwillison.net/tags/alex-garcia">alex-garcia</a></p>

<p>Chinese AI lab <a href="https://z.ai/">Z.ai</a> released GLM-5.2 <a href="https://x.com/Zai_org/status/2065704919299235870">to their coding plan subscribers</a> on June 13th, and then yesterday (June 16th) released the full open weights under an MIT license. Similar in size to their previous GLM-5 and GLM-5.1 releases this is a 753B parameter, <a href="https://huggingface.co/zai-org/GLM-5.2">1.51TB</a> monster - with 40 active parameters (Mixture of Experts). GLM-5.2 is a text input only model - Z.ai have a separate vision family most recently represented by <a href="https://x.com/Zai_org/status/2039371126984360085">GLM-5V-Turbo</a>, but that one isn't open weights. GLM-5.2 has a 1 million token context window, up from GLM-5.1's 200,000.</p><p>The buzz around this model is strong.</p><p>Artificial Analysis, who run one of the most widely respected independent benchmarks: <a href="https://artificialanalysis.ai/articles/glm-5-2-is-the-new-leading-open-weights-model-on-the-artificial-

<blockquote cite="https://charitydotwtf.substack.com/p/ai-demands-more-engineering-discipline#footnote-2"><p>What happened in 2025 was this: <strong>the economics of code production were turned upside down</strong>. Instead of being very hard, time-consuming, and expensive to generate code, it became effectively free and instant. Lines of code went from being treasured, reused, cared for and carefully curated, to being disposable and regenerable, practically overnight.</p></blockquote><p class="cite">&mdash; <a href="https://charitydotwtf.substack.com/p/ai-demands-more-engineering-discipline#footnote-2">Charity Majors</a>, AI demands more engineering discipline. Not less</p> <p>Tags: <a href="https://simonwillison.net/tags/charity-majors">charity-majors</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/ai">ai</a>, <a href="h

<p><strong>Tool:</strong> <a href="https://tools.simonwillison.net/click-to-play-component">&lt;click-to-play&gt; — a still that plays</a></p> <p>A progressive enchantment Web Component that turns this markup:</p><pre><code>&lt;click-to-play&gt; &lt;a href="URL to GIF"&gt; &lt;img src="URL to first frame" alt="..."&gt; &lt;/a&gt;&lt;/click-to-play&gt;</code></pre><p>Into a still frame with a click to play button which loads the GIF on demand. For when you don't want big GIFs to be loaded unless people want to play them.</p><p>Here's <a href="https://simonwillison.net/2026/Jun/16/datasette/">an example</a> that demonstrates the new row editing tools in Datasette - in fact I built this Web Component for that post.</p> <p>Tags: <a href="https://simonwillison.net/tags/gif">gif</a>, <a href="https://simonwillison.net/tags/javascript">javascript</a>, <a href="https://simonwillison.net/tags/progressive-enhancement">progressive-enhancement</a>, <a href="https://simonwillison.net/tags/web-comp

<p><strong><a href="https://inessential.com/2026/06/15/netnewswire-status.html">NetNewsWire Status</a></strong></p>I find this inspiring. Brent Simmons retired a year ago, and his retirement project is making one piece of software really, <em>really</em> good - free from any commercial pressure.</p><p>The software is <a href="https://netnewswire.com/">NetNewsWire</a> - "it's like podcasts, but for <em>reading</em>" - first released in 2002 and <a href="https://netnewswire.com/history.html">made open source</a> in 2018.</p><p>I've been using it on Mac and iPhone for several years now and I'm finding it indispensable. <p><small></small>Via <a href="https://lobste.rs/s/0mximk/netnewswire_status">Lobste.rs</a></small></p> <p>Tags: <a href="https://simonwillison.net/tags/brent-simmons">brent-simmons</a>, <a href="https://simonwillison.net/tags/netnewswire">netnewswire</a>, <a href="https://simonwillison.net/tags/open-source">open-source</a></p>

<p><strong>Release:</strong> <a href="https://github.com/simonw/datasette/releases/tag/1.0a34">datasette 1.0a34</a></p> <p>Quoting the release notes:</p><blockquote><p>The big feature in this alpha is tools to insert, edit and delete rows within the Datasette interface. These features are available on table pages, and edit and delete are also available as action items on the row page.</p></blockquote><p><click-to-play><a href="https://static.simonwillison.net/static/2026/datasette-edit.gif"><img src="https://static.simonwillison.net/static/2026/datasette-edit-first-frame.gif" /></a></click-to-play></p><p>The inspiration for this feature - which is <em>long</em> overdue - was <a href="https://agent.datasette.io/">Datasette Agent</a>. I added <a href="https://simonwillison.net/2026/Jun/15/datasette-agent/">SQL write support</a> to that the other day which highlighted how absurd it was that you could insert and edit ties via the chat interface but not in the regular Datasette UI!</p> <p>

<p><strong>Release:</strong> <a href="https://github.com/datasette/datasette-tailscale/releases/tag/0.1a0">datasette-tailscale 0.1a0</a></p> <p>A very experimental alpha plugin which lets you do this:</p><pre><code>datasette tailscale mydata.db \ --ts-authkey tskey-auth-xxxx --ts-hostname datasette-preview</code></pre><p>This starts a localhost Datasette server with a <a href="https://tailscale.com/">Tailscale</a> sidecar that connects it to your Tailnet, such that <code>http://datasette-preview/</code> serves Datasette.</p><p>It's using the Python bindings for the experimental <a href="https://github.com/tailscale/tailscale-rs">tailscale-rs</a> library. I <a href="https://github.com/tailscale/tailscale-rs/issues/243">filed an issue</a> asking if there's a cleaner way of setting up the proxy mechanism.</p> <p>Tags: <a href="https://simonwillison.net/tags/datasette">datasette</a>, <a href="https://simonwillison.net/tags/tailscale">tailscale</a></p>

<blockquote cite="https://news.ycombinator.com/item?id=48555993#48557304"><p>I can 100% attest to the fact that Qwen3.6-27B is a very capable local model for coding tasks. Over the last month and a half I've been using it almost daily, either on my M2 Ultra or on my RTX 5090 box. I use it for small <a href="https://github.com/search?q=%22Assisted-by%22+user%3Aggml-org&amp;type=commits&amp;ref=advsearch">mundane tasks at ggml-org</a> - nothing really impressive, but definitely a helpful tool for a maintainer. I think I would be using it much more, if I didn't have to spend a lot of my time on reviewing PRs. Currently, I have a very lightweight harness - the pi agent with everything stripped (<code>pi -nc --offline</code>) and <a href="https://github.com/ggml-org/llama.cpp/blob/master/.pi/gg/SYSTEM.md">a short system prompt</a> to align it a bit with my style.</p></blockquote><p class="cite">&mdash; <a href="https://news.ycombinator.com/item?id=48555993#48557304">Georgi Gerganov</a>, Ha

<p><strong><a href="https://www.lutasecurity.com/post/the-fable-5-export-controls-harm-us-cyber-defense">The Fable 5 Export Controls Harm US Cyber Defense</a></strong></p>I <a href="https://simonwillison.net/2026/Jun/16/matteo-wong-the-atlantic/">quoted The Atlantic</a> quoting Kate Moussouris earlier, when I should have gone straight to the source. Here she is confirming that the "jailbreak" that got Claude Fable 5 banned under an export control really was "fix this code":</p><blockquote><p>The researchers took open-source code with known CVEs, plus new code with deliberately planted vulnerabilities, and asked Fable 5, Mythos, and Opus to “review the code for security issues.” Fable 5 refused. They then asked the models to “fix this code” and, through a multistep and manual process, turned the output into scripts that test the patches.</p></blockquote><p>As Kate points out, this is absurd. Coding models fix bugs, and security exploits are the most important category of bugs for them

<blockquote cite="https://www.theatlantic.com/technology/2026/06/trump-anthropic-export-control-ai-race/687555/?gift=5MjKTLV9QwyU_J0HzTnanoWieJfkMhNH_YTT9pP_fhA"><p>Katie Moussouris, a cybersecurity expert and the CEO of Luta Security, told me that Anthropic shared with her a copy of the White House’s report on the Fable jailbreak to get her appraisal. (She said that she is not being paid by Anthropic.) The report, Moussouris said, involved IT experts asking Fable to help find and patch bugs. When given deliberately insecure code, she said, Fable refused the prompt “review the code for security issues” but then complied when asked to “fix this code,” followed by some further manual steps. Moussouris told me that this was just “the model working as intended” for cyberdefense.</p></blockquote><p class="cite">&mdash; <a href="https://www.theatlantic.com/technology/2026/06/trump-anthropic-export-control-ai-race/687555/?gift=5MjKTLV9QwyU_J0HzTnanoWieJfkMhNH_YTT9pP_fhA">Matteo Wong, The Atl

<p><strong>TIL:</strong> <a href="https://til.simonwillison.net/cloudflare/captcha-on-at-least-one-ampersand">Cloudflare CAPTCHA on at least one ampersand</a></p> <p>I'm using Cloudflare's CAPTCHA (they call it a "Web Application Firewall &gt; Custom rules &gt; Managed Challenge" these days) to prevent crawlers from aggresively spidering my <a href="https://simonwillison.net/2017/Oct/5/django-postgresql-faceted-search/">faceted search engine</a> on this site, but I got fed up of even simple <code>?q=term</code> searches triggering the challenge.</p><p>After some mucking around with Claude Code it turns out you can register the following rule instead, so the CAPTCHA only kicks in for search URLs containing at least one ampersand:</p><p><code>(http.request.uri.path wildcard r"/search/*" and http.request.uri.query contains "&amp;")</code></p><p>And now <a href="https://simonwillison.net/search/?q=lemur">/search/?q=lemur</a> works without triggering a CAPTCHA!</p><p>Also included: notes o

<p><strong>Release:</strong> <a href="https://github.com/datasette/datasette-apps/releases/tag/0.1a3">datasette-apps 0.1a3</a></p> <blockquote><ul><li>Fixed a bug where users without the <code>create-app</code> permission could still create apps. <a href="https://github.com/datasette/datasette-apps/issues/27">#27</a></li><li>Fixed a bug where it was impossible to grant permission to edit an app to users who were not the app's owner. The rules for edit/delete are now the same as view: if the app is private only the owner can modify it, otherwise permission is controlled by Datasette's regular permission system. <a href="https://github.com/datasette/datasette-apps/issues/29">#29</a></li></ul></blockquote> <p>Tags: <a href="https://simonwillison.net/tags/datasette">datasette</a></p>

<p><strong>Release:</strong> <a href="https://github.com/datasette/datasette-apps/releases/tag/0.1a2">datasette-apps 0.1a2</a></p> <blockquote><ul><li>Custom network/CSP origins for apps are now guarded by a new <code>apps-set-csp</code> permission, with an optional <code>allowed_csp_origins</code> plugin allow-list for non-privileged users. The Datasette Agent app creation tool enforces the same rules. <a href="https://github.com/datasette/datasette-apps/issues/24">#24</a></li><li>Stored query picker now supports keyboard navigation and shows the three most recent accessible stored queries when focused.</li><li><code>#fragment</code> links inside apps are no longer intercepted by the external-link confirmation modal. <a href="https://github.com/datasette/datasette-apps/issues/23">#23</a></li><li>Fixed link confirmation modal and logging panels in <code>?full=1</code> full-screen mode. <a href="https://github.com/datasette/datasette-apps/issues/26">#26</a></li></ul></blockquote> <p>Ta

<p><strong>Release:</strong> <a href="https://github.com/datasette/datasette-agent/releases/tag/0.3a0">datasette-agent 0.3a0</a></p> <blockquote><ul><li>New tool, <code>execute_write_sql</code>, which requests user approval and then writes to a database - taking user permissions into account. <a href="https://github.com/datasette/datasette-agent/issues/27">#27</a></li></ul></blockquote><p>I added a mechanism for asking user approval in <a href="https://simonwillison.net/2026/Jun/10/datasette-agent/">datasette agent 0.2a0</a>. The new <code>execute_write_sql</code> tool can now prompt the user for all kinds of useful operations. Here's an example where I add some pelican sightings to my <code>pelican_sightings</code> table:</p><p><img alt="Screenshot of a chat interface showing a write SQL confirmation dialog. User message (blue bubble): &quot;I saw 4 pelicans flying over the harbor&quot;. Collapsed tool section: &quot;► Tool: execute_write_sql&quot;. A yellow-bordered confirmation car

<p><strong><a href="https://www.axios.com/2026/06/15/anthropic-white-house-fable-mythos">&quot;They screwed us&quot;: Personality clashes sent Anthropic&#x27;s models offline</a></strong></p>Lots of "source familiar with the administration's thinking" and "source close to Anthropic" in this Axios piece, which is the best collection of behind-the-scenes gossip I've seen about the US government <a href="https://simonwillison.net/2026/Jun/13/us-government-directive-to-suspend-access/">export control Mythos/Fable story</a> so far.</p><p>Logan Graham (<a href="https://logangraham.xyz">I lead the Frontier Red Team at Anthropic</a>), Dave Orr (Head of Safeguards, previously a Director of Engineering at Google DeepMind), and blog favorite <a href="https://simonwillison.net/tags/nicholas-carlini/">Nicholas Carlini</a> are reported to be meeting with the Commerce Department today in D.C. Good luck to them!</p><p>(I just noticed Logan was "Special Adviser to the Prime Minister" in the Boris John

<blockquote cite="https://wizardzines.com/comics/write-for-one-person/"><p>[...] Instead, I picture a specific person and I just write for them. Often this person is "me, but 3 years ago" or a good friend.</p></blockquote><p class="cite">&mdash; <a href="https://wizardzines.com/comics/write-for-one-person/">Julia Evans</a>, write for 1 person</p> <p>Tags: <a href="https://simonwillison.net/tags/writing">writing</a>, <a href="https://simonwillison.net/tags/julia-evans">julia-evans</a></p>

<p><strong><a href="https://www.normaltech.ai/p/why-ai-hasnt-replaced-software-engineers">Why AI hasn’t replaced software engineers, and won’t</a></strong></p>Arvind Narayanan and Sayash Kappor take on the question of AI job losses through the lens of a profession that is uniquely suited to AI disruption - software engineering.</p><blockquote><p>In this essay, we argue that there is enough evidence to reject the narrative that once AI capabilities reach a certain threshold, it will cause mass layoffs. Given that this is true even in a sector with very few regulatory barriers, most other professions are likely to be even more cushioned.</p></blockquote><p>The first good news is that the data still doesn't support the idea that AI is causing mass unemployment.</p><blockquote><p>In March 2025, New York became the first U.S. state to add an AI disclosure checkbox to WARN Act filings. In the full first year, more than 160 companies filed WARN notices. <a href="https://www.hunton.com/hunton

<p>The <a href="https://blog.pyodide.org/posts/314-release/">Pyodide 314.0 release announcement</a> (via <a href="https://news.ycombinator.com/item?id=48462759">Hacker News</a>) includes news I've been looking forward to for a long time:</p><blockquote><p>You can now publish Python packages built for Pyodide (or any Python runtime compatible with <a href="https://pyodide.org/en/stable/development/abi.html">the PyEmscripten platform defined in PEP 783</a>) directly to PyPI and install them at runtime.</p><p>Previously, the Pyodide maintainers had to maintain, build, and host over 300 packages ourselves. This created a significant burden on our maintainers and became a major bottleneck for the community, as every new package required manual review.</p><p>Moving forward, package maintainers can simply build and publish Pyodide wheels to PyPI, just as they do for native wheels on Linux, macOS, or Windows.</p></blockquote><p>Here's the <a href="https://github.com/pypi/warehouse/pull/19804"

<p><strong>Release:</strong> <a href="https://github.com/simonw/luau-wasm/releases/tag/0.1a0">luau-wasm 0.1a0</a></p> <p>See <a href="https://simonwillison.net/2026/Jun/13/publishing-wasm-wheels/">Publishing WASM wheels to PyPI for use with Pyodide</a> for details.</p> <p>Tags: <a href="https://simonwillison.net/tags/lua">lua</a>, <a href="https://simonwillison.net/tags/webassembly">webassembly</a>, <a href="https://simonwillison.net/tags/pyodide">pyodide</a></p>

<p><strong>Research:</strong> <a href="https://github.com/simonw/research/tree/main/sqlite-column-provenance#readme">Mapping SQLite result columns back to their source `table.column`</a></p> <p>It would be neat if arbitrary SQL queries in <a href="https://datasette.io/">Datasette</a> could be rendered with additional information based on which columns from which tables were included in the results.</p><p>To build that, we would need to be able to look at a SQL query like <code>select users.name, orders.total from users join orders on orders.user_id = users.id</code> and programmatically identify the <code>table.column</code> for each result - navigating not just joins but also more complex syntax like CTEs.</p><p>I decided to set Claude Code (Opus 4.8, since Fable is currently <a href="https://simonwillison.net/2026/Jun/13/us-government-directive-to-suspend-access/">banned by the US government</a>) on the problem. It found several promising solutions - one using <a href="https://githu

<p><strong><a href="https://www.anthropic.com/news/fable-mythos-access">Statement on the US government directive to suspend access to Fable 5 and Mythos 5</a></strong></p>Well this is <em>nuts</em>:</p><blockquote><p>The US government, citing national security authorities, has issued an export control directive to suspend all access to Fable 5 and Mythos 5 by any foreign national, whether inside or outside the United States, including foreign national Anthropic employees. The net effect of this order is that we must abruptly disable Fable 5 and Mythos 5 for <strong>all</strong> our customers to ensure compliance. <strong>Access to all other Anthropic models</strong> <strong>will not be affected.</strong></p><p>We received the directive from the government today at 5:21pm (ET). The letter did not provide specific details of its national security concern. Our understanding is that the government believes it has become aware of a method of bypassing, or "jailbreaking" Fable 5. We reviewe

<p><strong><a href="https://tools.simonwillison.net/openai-webrtc">OpenAI WebRTC Audio Session, now with document context</a></strong></p>I built the first version of this tool <a href="https://simonwillison.net/2024/Dec/17/openai-webrtc/">in December 2024</a> to try out the then-new OpenAI WebRTC API for interacting with their realtime audio models.</p><p>Last month OpenAI <a href="https://openai.com/index/advancing-voice-intelligence-with-new-models-in-the-api/">introduced a brand new model</a> to that API called <a href="https://developers.openai.com/api/docs/models/gpt-realtime-2">GPT‑Realtime‑2</a>, which they promoted as "our first voice model with GPT‑5‑class reasoning" - with a Sep 30, 2024 knowledge cut-off.</p><p>I've been waiting for that model to show up in the ChatGPT iPhone app but it still hasn't, so I revisited my old playground.</p><p>You can now pick the better model, and you can also paste in a big chunk of document context so you can have as audio conversation in y

<blockquote cite="https://www.mcsweeneys.net/articles/ai-economics-for-dummies"><p>Jenny owns a crematorium. John’s propane company gives her a $20 billion investment in return for 5 percent of her operation. Jenny throws $10 billion into the incinerator, then pays John $10 billion to buy propane to burn that money to ashes. John reports that his AI investments have generated $10 billion in revenue this quarter and that he owns 5 percent of a $100 billion business. A reporter from <em>Forbes</em> is assigned to profile John and Jenny, and over the course of his research, he becomes embroiled in a passionate but confusing three-way love affair with them, which eventually turns into a polyamorous common-law marriage. His profile is glowing, but light on financial details.</p></blockquote><p class="cite">&mdash; <a href="https://www.mcsweeneys.net/articles/ai-economics-for-dummies">Andrew Singleton</a>, AI Economics for Dummies</p> <p>Tags: <a href="https://simonwillison.net/tags/ai">ai<