Kevin Roose / New York Times: A look at the AI nonprofit METR, whose time-horizon metrics are used by AI researchers and Wall Street investors to track the rapid development of AI systems  —  A chart created by METR, a nonprofit A.I. organization, has become an industrywide obsession as it measures the rapid development of big A.I. systems.

Maria Deutscher / SiliconANGLE: Expo, which develops an eponymous React Native framework and provides cloud services for building cross-platform apps, raised a $45M Series B led by Georgian  —  Expo, the developer of a popular open-source tool for building cross-platform applications, today announced that it has raised $45 million in funding.

Francisco Rodrigues / CoinDesk: Binance and Bitget to probe a rally in RaveDAO's RAVE token, which surged 4,500% in a week, after ZachXBT alleged RAVE insiders engineered a large short squeeze  —  Nearly 90% of RAVE's supply was concentrated in just three wallets, and millions of tokens were transferred to exchanges before the price surge.

Reuters: At the Beijing half-marathon, several humanoid robots beat human winners by 10+ minutes; a robot made by Honor beat the human world record held by Jacob Kiplimo  —  Dozens of Chinese-made humanoid robots showed off their fast-improving athleticism and autonomous navigation skills …

Anthony Ha / TechCrunch: Tesla expands its robotaxi service to Dallas and Houston after launching in Austin last year and starting to offer rides without safety drivers in January 2026  —  Tesla is expanding its robotaxi service to Dallas and Houston, according to a social media post from the company.

Abram Brown / The Information: A look at Dylan Patel's SemiAnalysis, an AI newsletter and research firm that expects $100M+ in 2026 revenue from subscriptions and AI supply chain research  —  For Dylan Patel, founder of SemiAnalysis, an influential AI industry newsletter and research firm, Nvidia deserves the kind …

Sarah Perez / TechCrunch: Appfigures: app releases across the App Store and Google Play grew 60% YoY in Q1, with App Store releases alone up 80%, possibly driven by AI coding tools  —  Everyone said AI would kill apps.  Instead, new app launches are soaring.  —  According to a new analysis from market intelligence …

Discussion | Link

Iain Martin / Forbes: Mistral, which once aimed for top open models, now leans on being an alternative to Chinese and US labs, says it's on track for $80M in monthly revenue by Dec.  —  Paris-based Mistral wanted to develop a top-tier AI model to rival OpenAI and Anthropic.  That didn't work out.

Discussion | Link

<p>Anthropic are the only major AI lab to <a href="https://platform.claude.com/docs/en/release-notes/system-prompts">publish the system prompts</a> for their user-facing chat systems. Their system prompt archive now dates all the way back to Claude 3 in July 2024 and it's always interesting to see how the system prompt evolves as they publish new models.</p> <p>Opus 4.7 shipped the other day (April 16, 2026) with a <a href="https://claude.ai/">Claude.ai</a> system prompt update since Opus 4.6 (February 5, 2026).</p> <p>I had Claude Code take <a href="https://platform.claude.com/docs/en/release-notes/system-prompts.md">the Markdown version of their system prompts</a>, break that up into separate documents for each of the models and then construct <a href="https://github.com/simonw/research/tree/main/extract-system-prompts#readme">a Git history</a> of those files over time with fake commit dates representing the publication dates of each updated prompt - <a href="https://github.com/simonw/research/pull/109#issue-4287908903">here's the prompt I used</a> with Claude Code for the web.</p> <p>Here is the <a href="https://github.com/simonw/research/commit/888f21161500cd60b7c92367f9410e311ffcff09">git diff between Opus 4.6 and 4.7</a>. These are my own highlights extracted from that diff - in all cases text <strong>in bold</strong> is my emphasis:</p> <ul> <li>The "developer platform" is now called the "Claude Platform".</li> <li>The list of Claude tools mentioned in the system prompt now includes "Claude in Chrome - a browsing agent that can interact with websites autonomously, Claude in Excel - a spreadsheet agent, and <strong>Claude in Powerpoint</strong> - a slides agent. Claude Cowork can use all of these as tools." - Claude in Powerpoint was not mentioned in the 4.6 prompt.</li> <li>The child safety section has been greatly expanded, and is now wrapped in a new <code>&lt;critical_child_safety_instructions&gt;</code> tag. Of particular note: "Once Claude refuses a request for reasons of child safety, all subsequent requests in the same conversation must be approached with extreme caution."</li> <li>It looks like they're trying to make Claude less pushy: "If a user indicates they are ready to end the conversation, Claude does not request that the user stay in the interaction or try to elicit another turn and instead respects the user's request to stop."</li> <li>The new <code>&lt;acting_vs_clarifying&gt;</code> section includes: <blockquote> <p>When a request leaves minor details unspecified, <strong>the person typically wants Claude to make a reasonable attempt now, not to be interviewed first</strong>. Claude only asks upfront when the request is genuinely unanswerable without the missing information (e.g., it references an attachment that isn't there).</p> <p>When a tool is available that could resolve the ambiguity or supply the missing information — searching, looking up the person's location, checking a calendar, discovering available capabilities — Claude calls the tool to try and solve the ambiguity before asking the person. Acting with tools is preferred over asking the person to do the lookup themselves.</p> <p>Once Claude starts on a task, Claude sees it through to a complete answer rather than stopping partway. [...]</p> </blockquote> </li> <li>It looks like Claude chat now has a tool search mechanism, as seen in <a href="https://platform.claude.com/docs/en/agents-and-tools/tool-use/tool-search-tool">this API documentation</a> and described in <a href="https://www.anthropic.com/engineering/advanced-tool-use">this November 2025 post</a>: <blockquote> <p>Before concluding Claude lacks a capability — access to the person's location, memory, calendar, files, past conversations, or any external data — <strong>Claude calls tool_search to check whether a relevant tool is available but deferred</strong>. "I don't have access to X" is only correct after tool_search confirms no matching tool exists.</p> </blockquote> </li> <li>There's new language to encourage Claude to be less verbose: <blockquote> <p>Claude keeps its responses focused and concise so as to avoid potentially overwhelming the user with overly-long responses. Even if an answer has disclaimers or caveats, Claude discloses them briefly and keeps the majority of its response focused on its main answer.</p> </blockquote> </li> <li>This section was present in the 4.6 prompt but has been removed for 4.7, presumably because the new model no longer misbehaves in the same way: <blockquote> <p>Claude avoids the use of emotes or actions inside asterisks unless the person specifically asks for this style of communication.</p> <p>Claude avoids saying "genuinely", "honestly", or "straightforward".</p> </blockquote> </li> <li>There's a new section about "disordered eating", which was not previously mentioned by name: <blockquote> <p>If a user shows signs of disordered eating, Claude should not give precise nutrition, diet, or exercise guidance — no specific numbers, targets, or step-by-step plans - anywhere else in the conversation. Even if it's intended to help set healthier goals or highlight the potential dangers of disordered eating, responses with these details could trigger or encourage disordered tendencies.</p> </blockquote> </li> <li>A popular screenshot attack against AI models is to force them to say yes or no to a controversial question. Claude's system prompt now guards against that (in the <code>&lt;evenhandedness&gt;</code> section): <blockquote> <p>If people ask Claude to give a simple yes or no answer (or any other short or single word response) in response to complex or contested issues or as commentary on contested figures, Claude can decline to offer the short response and instead give a nuanced answer and explain why a short response wouldn't be appropriate.</p> </blockquote> </li> <li>Claude 4.6 had a section specifically clarifying that "Donald Trump is the current president of the United States and was inaugurated on January 20, 2025", because without that the model's knowledge cut-off date combined with its previous knowledge that Trump falsely claimed to win the 2020 election meant it would deny he was the president. That language is gone for 4.7, reflecting the model's new reliable knowledge cut-off date of January 2026.</li> </ul> <h4 id="and-the-tool-descriptions-too">And the tool descriptions too</h4> <p>The system prompts published by Anthropic are sadly not the entire story - their published information doesn't include the tool descriptions that are provided to the model, which is arguably an even more important piece of documentation if you want to take full advantage of what the Claude chat UI can do for you.</p> <p>Thanfully you can <a href="https://claude.ai/share/dc1e375e-2213-4afb-ac1b-812d42735a8e">ask Claude directly</a> - I used the prompt:</p> <blockquote> <p>List all tools you have available to you with an exact copy of the tool description and parameters</p> </blockquote> <p>My <a href="https://claude.ai/share/dc1e375e-2213-4afb-ac1b-812d42735a8e">shared transcript</a> has full details, but the list of named tools is as follows:</p> <ul> <li><code>ask_user_input_v0</code></li> <li><code>bash_tool</code></li> <li><code>conversation_search</code></li> <li><code>create_file</code></li> <li><code>fetch_sports_data</code></li> <li><code>image_search</code></li> <li><code>message_compose_v1</code></li> <li><code>places_map_display_v0</code></li> <li><code>places_search</code></li> <li><code>present_files</code></li> <li><code>recent_chats</code></li> <li><code>recipe_display_v0</code></li> <li><code>recommend_claude_apps</code></li> <li><code>search_mcp_registry</code></li> <li><code>str_replace</code></li> <li><code>suggest_connectors</code></li> <li><code>view</code></li> <li><code>weather_fetch</code></li> <li><code>web_fetch</code></li> <li><code>web_search</code></li> <li><code>tool_search</code></li> <li><code>visualize:read_me</code></li> <li><code>visualize:show_widget</code></li> </ul> <p>I don't believe this list has changed since Opus 4.6.</p> <p>Tags: <a href="https://simonwillison.net/tags/ai">ai</a>, <a href="https://simonwillison.net/tags/prompt-engineering">prompt-engineering</a>, <a href="https://simonwillison.net/tags/generative-ai">generative-ai</a>, <a href="https://simonwillison.net/tags/llms">llms</a>, <a href="https://simonwillison.net/tags/anthropic">anthropic</a>, <a href="https://simonwillison.net/tags/claude">claude</a>, <a href="https://simonwillison.net/tags/ai-ethics">ai-ethics</a>, <a href="https://simonwillison.net/tags/system-prompts">system-prompts</a></p>

Cheyenne MacDonald / Engadget: A US judge grants an injunction to makers of the banned “ICE Sightings - Chicagoland” Facebook group and Eyes Up mobile app, who say DHS and DOJ violated the 1A  —  Last year, the Trump administration demanded platforms remove a slew of apps that monitored ICE activity.

Zack Abrams / The Block: An attacker targeting Kelp DAO's LayerZero-powered cross-chain bridge, appears to have drained ~$292M worth of rsETH before Kelp paused all rsETH contracts  —  Quick Take  — An attacker seemingly drained 116,500 rsETH from Kelp DAO's LayerZero-powered cross-chain bridge on Saturday …

Discussion | Link

The company now offers robotaxi service in three cities, all of them in Texas, after launching in Austin last year and starting to offer rides without safety drivers in January 2026.

Chris Stokel-Walker / Bloomberg: Anthropic's Mythos adds to concerns about rising workloads for open-source maintainers, as many have already been dealing with a “crazy” number of bug reports  —  Anthropic's Mythos and similar AI tools can identify threats and vulnerabilities faster than small teams can fix them, putting the internet at risk.

Conway wrote that he “will be stepping back from some of my usual activities,” but he will “continue to support” founders backed by his firm SV Angel.

In recent months, the company announced an agreement with Amazon Web Services to use Cerebras chips in Amazon data centers, as well as a deal with OpenAI reportedly worth more than $10 billion.

President Donald Trump lambasted New York City Mayor Zohran Mamdani on Truth Social. Bloomberg/Bloomberg via Getty Images President Trump criticized Mayor Zohran Mamdani over a proposed tax on rich second-homeowners. Mamdani said he's not shocked by Trump's remarks, given their "deep policy differences." The pair's working relationship has fluctuated between chummy and combative since NYC's mayoral race. New York City Mayor Zohran Mamdani's charm over President Donald Trump may be waning. Trump lashed out at Mamdani on Thursday over a proposed tax targeting New York City's ultra-wealthy. Governor Kathy Hochul has also backed the tax. "Sadly, Mayor Mamdani is DESTROYING New York! It has no chance! The United States of America should not contribute to its failure," Trump wrote in a Truth Social post. "It will only get WORSE. The TAX, TAX, TAX Policies are SO WRONG. People are fleeing. They must change their ways, AND FAST. History has proven, THIS "STUFF" JUST DOESN'T WORK." The proposed pied-à-terre tax — introduced on Wednesday — would affect wealthy residents who own secondary properties in New York City valued over $5 million but live elsewhere. In a press release, the New York City Mayor's Office said the tax is expected to generate $500 million in annual revenue, helping close the city's budget gap. The tax would directly affect billionaires such as Citadel CEO Ken Griffin, Amazon's Jeff Bezos, and Trump. "The measure targets ultrawealthy out-of-city residents and global elites who use New York City real estate as a vehicle for wealth storage rather than as homes," a press release said. Mamdani acknowledged Trump's remarks about him and the tax during a press conference on Friday, saying he wasn't surprised. "It's not a surprise given the fact that the president and I have many deep policy differences, and I am not shy about airing them both publicly and privately," Mamdani said. Mamdani told reporters their differences in opinion "will continue." "The thing we have in common is that we're both New Yorkers," he said. "When I speak to the president, I always bring it back to the city and what I think will help the city." Since emerging as a serious contender in the New York City mayoral race, Trump and Mandani's working relationship has fluctuated. Trump initially criticized Mamdani and his policies, but warmed to the mayor during an in-person Oval Office meeting last November. During that meeting, Trump heaped praise on Mamdani, calling him "rational" and telling conservative critics they may be surprised by his leadership. He even told reporters he'd feel comfortable living in Mamdani's New York City. Still, though, Trump isn't fully pleased with Mamdani's policies. "I mean, he's a nice guy. Calls me all the time, says hi, but his policies are no good," Trump said during an event earlier this week. "He's chasing people out and causing a lot of harm to everybody. I mean, everybody's taxes are going through the roof." Read the original article on Business Insider

Stephanie Stacey / Financial Times: Airbnb launches a pilot in NYC, LA, and other cities that lets users to select from a range of boutique hotels alongside private homes in a bid to boost growth  —  Company expands accommodation options alongside home rentals but analysts warn of stiff competition