Cycle on Jun 9, 2026

I will start by listing the open threads for this topic to see if there are any specific research questions or watches we need to address in this cycle.

I will list the current set of notes for the topic to see what has already been documented and to make sure I don't duplicate any existing findings.

I see several important items: 1. One of the watches is `Apple WWDC 2026 Siri and iOS 27 AI Agent Releases` which has a fire date of June 9, 2026 (today!). Let's see what Apple has announced at WWDC 2026. WWDC typically starts on June 8 or June 9, 2026. Let's do a search on WWDC 2026, Siri, iOS 27, and Apple's AI agent capabilities. 2. Let's also check SpaceX and xAI IPO / updates (scheduled for June 12, 2026, but maybe there are updates). 3. Let's look for other major AI releases or policy changes in the last few days (since June 5-9, 2026). Let's do a web search for Apple WWDC 2026 Siri iOS 27 AI agent announcements. I will use `sort="recent"` and `freshness="week"`. Let's state intent then call the tool.

The search results point to a blog post: `WWDC 2026: iOS 27, New Siri & Dev Tools` with URL `https://lushbinary.com/blog/wwdc-2026-announcements-ios-27-siri-developer-guide/` Let's fetch this URL to see the exact details of the WWDC 2026 announcements, especially since WWDC 2026 is happening now (today is June 9, 2026). Let's state intent then fetch this page.

This is a massive and extremely detailed set of findings. Let's analyze what we have: - Title: WWDC 2026: iOS 27, New Siri & Dev Tools - Key points: 1. Siri has been completely rebuilt on a **1.2-trillion-parameter Google Gemini model** licensed for roughly **$1 billion per year** (publicly confirmed by Google Cloud CEO Thomas Kurian at Google Cloud Next '26). 2. This 1.2T-parameter model uses a mixture-of-experts design. 3. Simple requests run on Apple's on-device Foundation Models, but complex queries route to Google's fleet of **Nvidia Blackwell B200 chips on Google Cloud** instead of Private Cloud Compute (PCC) because PCC hardware was too slow at Siri's query volumes. 4. To ensure privacy, Apple enabled **Nvidia's hardware-based confidential computing** on those Blackwell chips. This encrypts user inputs, model weights, and inference results in GPU memory. The contract also prevents Google from using Siri queries to train future Gemini models. 5. **SiriKit has been formally deprecated** (with a 2-3 year support window) in favor of the **App Intents framework**, which is now the *mandatory* integration surface for Siri. 6. **Xcode 27** ships with on-device AI code completion (running locally on Apple Silicon), with options to configure third-party AI models. 7. **iOS 27** supports iPhone 11 and up. Apps launch up to 30% faster. 8. **macOS 27 "Golden Gate"** is the new Mac release name, with Siri built into Spotlight. 9. SwiftUI and UIKit gained new **foldable layout APIs** (hinge-state detection, etc.) preparing for the rumored iPhone Fold in fall 2026. 10. Liquid Glass 2 design language with updated tokens, icons, and a transparency slider. 11. This helps resolve the watch: `Apple WWDC 2026 Siri and iOS 27 AI Agent Releases` (id: `019e9bc8-00dd-7d1d-b044-b9c333710a79`), which has a fire date of June 9, 2026 (today!). Let's do a quick search on Google Cloud Next '26 or Thomas Kurian WWDC 2026 Gemini model to double check or find more articles on Apple WWDC 2026 to verify and enrich our findings with more sources. Let's state intent and search.

Let's fetch the TechTimes article `WWDC 2026: App Intents Replaces SiriKit as Gemini Siri Migration Clock Starts` to get another high-quality source on this announcement. Let's state intent then fetch the page.

This is fantastic. We have detailed, concrete, and highly authoritative sources for the first watch: `Apple WWDC 2026 Siri and iOS 27 AI Agent Releases`. Let's look at the details: - Siri rebuilt on custom 1.2-trillion-parameter Google Gemini model. - Deal reported at roughly $1 billion per year, confirmed by Google Cloud CEO Thomas Kurian at Google Cloud Next '26. (Wait, some other sources say $5 billion, but the TechTimes and MacRumors/Lushbinary sources say custom 1.2T-parameter model for a reported ~$1B/year. We will cite both details accurately). - Complex queries route to Google Cloud's fleet of Nvidia Blackwell B200 chips because Apple's Private Cloud Compute (PCC) was too slow at Siri's query volumes. - To ensure privacy, Apple uses Nvidia's hardware-based confidential computing on those Blackwell chips. This encrypts inputs, weights, and results in GPU memory. - SiriKit is formally deprecated. App Intents is now the mandatory integration surface for Siri, with a 2-3 year window for migration. - Xcode 27 adds on-device AI code completion running locally on Apple Silicon. - New privacy manifest APIs let developers specify on a per-intent basis whether Siri queries can route to the cloud or must stay on-device (crucial for enterprise and healthcare). - iOS 27 supports iPhone 11 and up, apps launch 30% faster. - macOS 27 is named "Golden Gate". - Foldable layout APIs in SwiftUI/UIKit prepare for the iPhone Fold in fall 2026. - Liquid Glass 2 design language refinements. - Keynote was on June 8, 2026. This was Tim Cook's final WWDC keynote as CEO, as John Ternus is confirmed to take over on September 1, 2026. Let's record the event for this watch: - Thread ID: `019e9bc8-00dd-7d1d-b044-b9c333710a79` (Apple WWDC 2026 Siri and iOS 27 AI Agent Releases) - Resolves: `true` (since WWDC has happened and we have the official announcements) - Outcome: Apple officially announced iOS 27 and the rebuilt Siri powered by a custom 1.2-trillion-parameter Google Gemini model running on Nvidia Blackwell B200 chips in Google Cloud with confidential compute. Let's also look for updates on SpaceX and xAI IPO (which is scheduled for June 12, 2026). Let's do a search on SpaceX xAI IPO June 12 2026 to see if there are any fresh updates or pre-IPO news. Let's state intent and search.

Excellent. These are highly relevant and current regarding SpaceX's impending IPO on June 12, 2026. Let's fetch one of the MEXC learn articles or similar sources to see if there are any specific details on SpaceX's AI infrastructure, Starlink profits, and the IPO prospectus. Let's state intent then fetch. Wait, let's fetch `Who Really Owns SpaceX? How Elon Musk Controls 82.4% of the Votes With Just 42% of the Equity` (https://www.mexc.com/en-GB/learn/article/who-really-owns-spacex-how-elon-musk-controls-82-4-of-the-votes-with-just-42-of-the-equity/1) or `How Does SpaceX Make Money? Is It Profitable?` (https://www.mexc.com/learn/article/how-does-spacex-make-money-is-it-profitable-government-contracts-starlink-and-a-historic-ipo-explained/1). Let's fetch the "How Does SpaceX Make Money" article to see the details of their S-1 and financial performance/compute deals. Let's state intent then call the tool.

Ah, Firecrawl timed out. Let's try another source or search specifically for the SpaceX S-1 details or prospectus. Let's search for "SpaceX S-1 prospectus" or "SpaceX IPO 15 Things Investors Must Know" to get details. Let's state intent and fetch `https://www.bitget.com/academy/spacex-ipo-15-things-investors-must-know` first.

The Bitget article provides incredibly granular details on the upcoming SpaceX IPO on June 12, 2026. Let's analyze the new findings: 1. **Target IPO Price**: Fixed at **$135.00 per share** (unconventional "no-range" fixed price structure). 2. **Total Shares Offered**: 555.56 million Class A shares. 3. **Target Raise**: **$75.00 billion** in capital (setting up the largest IPO in stock market history, dwarfing Saudi Aramco's $25.6B raise). 4. **Implied Valuation**: **$1.77 trillion** (representing a 61% premium from its late 2025 private valuation). 5. **Stock Split**: Executed a 5-for-1 stock split on May 4, 2026. 6. **IPO Date**: **June 12, 2026** (Nasdaq ticker: **SPCX**, and the newly established **Nasdaq Texas**). 7. **The xAI / X Merger**: In February 2026, SpaceX acquired xAI and X Holdings Corp. (Twitter) in an all-stock common-control transaction. It retrospectively combined their finances. This makes SpaceX a three-pronged mega-conglomerate: Launch, Starlink, and AI/Cloud Compute (xAI + X). 8. **The Google Cloud Deal**: Signed June 5, 2026. Google pays SpaceX **$920 million per month** from Oct 2026 through June 2029 ($29.4B total) for access to 110,000 NVIDIA GPUs/CPUs. (Google can cancel if chips aren't delivered by Sept 30, 2026; after Dec 31, 2026, either party can end with 90-day notice). 9. **The Anthropic Deal**: Anthropic pays SpaceX **$1.25 billion per month** through May 2029 to rent idle compute capacity. Anthropic can cancel with 90-day notice. 10. **Starlink Financials**: In 2025, Starlink brought in $11.387 billion in revenue (61% of total) and generated $4.423 billion in operating income. Over 10.3 million active subscribers across 164 countries with 9,600 satellites. 11. **Consolidated Revenue**: $18.67 billion in 2025 (up 33% YoY), but unprofitable on a consolidated GAAP basis due to SpaceX's AI segment (xAI and X) reporting a Segment Adjusted EBITDA loss of $1.2B in 2025. 12. **Bitcoin Treasury**: SpaceX holds **18,712 Bitcoin** on its balance sheet (valued at $1.3B - $1.5B). 13. **Insider Lock-Up**: Elon Musk's shares (42% of equity, representing 85% of voting power through super-voting Class B shares with 10 votes each) are locked up for **366 days** with zero early-release. Other key insiders are locked up until the release of financial results for the quarter ending June 30, 2027. 14. **Directed Share Program (DSP)**: 5% of Class A IPO shares ($3.75 billion) are reserved for employees/friends without any lockup restrictions. 15. **Cursor Acquisition**: SpaceX is targeting an all-stock, **$60 billion acquisition of Cursor** (the popular AI coding assistant), with a $10 billion breakup fee. 16. **Water Bottleneck**: S-1/A warns about ecological footprint and water resources required for cooling its massive AI data centers. 17. **Space-Based Data Centers (2030 Bull Case)**: Ambitious plans to launch solar-powered, space-based data centers (using natural cold of orbit for cooling and solar energy for power) to bypass terrestrial land/energy constraints. This is incredible! These details are highly specific and completely update the SpaceX IPO story. Let's look at the third watch: `SpaceX and xAI Public Debut / IPO` (id: `019e9889-19fc-7370-b885-054439ee4234`) - Fire date: June 12, 2026. - Since the IPO is scheduled for June 12, 2026 (which is in 3 days), this watch hasn't *officially* fired yet, but we have the updated S-1/A prospectus details from June 3, 2026, and the finalized IPO pricing of $135.00 per share. Wait, let's look at the watch description: "This watch fires when the IPO takes place and financial reports or market trading data become available." Since today is June 9, 2026, the IPO hasn't taken place yet. So we should keep this watch open (do not resolve yet) or we can keep tracking it, but we can write a fantastic finding updating our prior notes: - `spacex-google-anthropic-compute-deals-ipo-preps` (SpaceX Preps for $1.77 Trillion IPO with Massive Compute Rental Deals for Google and Anthropic) - `anthropic-space-x-colossus-compute-deal` (The Compute Race: SpaceX Lands $30 Billion Google Deal and $15 Billion Anthropic Pact Ahead of Historic IPO) - `google-spacex-compute-deal` (Google Agrees to Pay SpaceX $920 Million Monthly for xAI Data Center Compute) Let's check if there are any other notes to update. Wait, there is a watch: `Meta Muse Spark Developer API Release` (id: `019e9889-2962-79c0-ba91-3aa809f9f6b9`) - Description: "Meta has repeatedly delayed the developer API for its closed-source Muse Spark model due to bugs and infrastructure issues. A spokesperson indicated a planned launch in June 2026. This watch tracks whether the API successfully ships this month or faces further delays." Let's do a search on Meta Muse Spark developer API June 2026 to see if there are any updates. Let's state intent then search.

Let's fetch one of the articles, e.g., `Meta Delays New AI Model Release Multiple Times as Development Challenges Persist, WSJ Reports` (https://meyka.com/blog/meta-delays-new-ai-model-release-multiple-times-as-development-challenges-persist-wsj-reports-2606/) or `Meta’s delay puts $14B investment in doubt` (https://rollingout.com/2026/06/07/meta-delay-puts-14b-investment-in-doubt/). Let's fetch the rollingout.com article first to see the latest. Or the meyka.com article. Let's try the meyka.com article. Let's state intent then fetch.

Okay, so the Muse Spark API has been delayed multiple times since April 2026. Meta still expects a release "within the month" (June 2026), but no firm launch date has been announced. Wait, let's look at the watch description for Muse Spark again: "This watch tracks whether the API successfully ships this month or faces further delays." Since the month of June is not yet over (today is June 9, 2026) and Meta still expects a release "within the month," this watch should remain active as `ongoing` or `one_shot` with a future fire date of June 30, 2026. Let's see if there are other interesting updates. Wait! Let's check the first finding: "Anthropic Warns of Recursive Self-Improvement and Calls for a Verifiable Global AI Pause" (slug: `anthropic-recursive-self-improvement-pause-call`). Are there any updates on this recursive self-improvement paper or reactions from other labs? Let's search! Let's state intent and search.

Let's fetch the Hacker News story `When AI Builds Itself: Our progress toward recursive self-improvement` or one of the articles, e.g. `Shelly Palmer: When AI builds itself` (https://www.sasktoday.ca/opinion/shelly-palmer-when-ai-builds-itself-12380856) or `When AI Builds Itself: Anthropic Warns of Risks From Autonomous Recursive Self-Improvement` (https://the420.in/anthropic-warns-recursive-self-improving-ai/). Let's fetch the SaskToday article first. Let's state intent then fetch.

This is a brilliant finding that expands on `anthropic-recursive-self-improvement-pause-call`. Let's look at the specific quotes and metrics: - "Their engineers now ship roughly eight times more code per quarter than they did in 2024. More than 80 per cent of the code merged into Anthropic’s codebase is written by Claude, up from low single digits before Claude Code launched in February 2025." - "The most interesting data point comes from independent research organization METR. Their benchmarks show the length of tasks frontier AI models can complete reliably is now doubling every four months, down from every seven months just a year ago." - "In one example, Claude generated approximately 800 bug-fix pull requests that reduced a class of API errors by a factor of one thousand. The engineer who supervised the work estimated it would have taken a human several years to achieve the same result." These are concrete, quantifiable metrics that we can use to update `anthropic-recursive-self-improvement-pause-call`. Let's do a search on Hacker News for "When AI Builds Itself" to see if there are any interesting comments or additional context. Let's state intent then search.

Let's fetch the Hacker News thread `https://news.ycombinator.com/item?id=48400842` to read the comments and find representative quotes about the "When AI Builds Itself" paper. Let's state intent then fetch.

This is an incredibly rich and deep discussion on Hacker News about Anthropic's "When AI Builds Itself" paper. Let's see some of the key comments and themes: 1. **Puff Piece / Marketing for IPO**: Many commenters express skepticism about the timing of the paper, pointing out that Anthropic is preparing for a public debut. - User `jameson` says: > "I don't quite understand the intent of such article other than to promote themselves given an odd timing that the company is planning on going public, so I can only conclude that this is just part of the IPO roadshow." - User `becquerel` says: > "If the post drops long before the IPO, it's vain boosterism. If it's near the IPO, it's fattening the pig. If it's after the IPO, it's pumping the stock price." 2. **Skepticism Over "Lines of Code" (LoC) Metric**: Commenters criticize the use of lines of code as a productivity metric, noting that LLMs generate highly verbose and sometimes redundant/buggy code. - User `torben-friis` quotes the article: > "At Anthropic, we don’t reward people for how many lines of code they write; rather, team members are producing more code simply because they’re using AI systems to write more code." And adds: > "What about the hypothesis that AI is generating more verbose code? I just see the text pretending to acknowledge 'LOC != Productivity' and then using it as a metric anyway." - User `malfist` shares a personal anecdote of a coworker submitting an AI-generated PR with 40k lines of code across 600 files, which they declined to review: > "One of my co-workers just asked me to review his pull request that was all AI generated. 600 files were touched, over 40k lines of code added. I'm sure he thought that was a crowning achievement... I declined to review it, stating that I couldn't possibly vet 40k lines of code..." - User `overgard` says: > "I just watched copilot today turn a 8 line fix into 500 lines, so, yeah, verbosity is a big side effect" - User `strix_varius` says: > "Exactly this. Just this week an engineer who seems to purely vibe everything submitted a +700ish LoC fix for what seemed like a pretty simple issue. Moreover it was a perf issue, which in my experience is not usually best fixed by adding more stuff. Today, I merged my fix, net -381 LoC." 3. **Claude Code Overhead & Overengineering**: Commenters point out that Claude Code (Anthropic's TUI) is built on React (via the Ink framework) and has huge resource footprints. - User `bitwize` says: > "After several months with their top engineers and state-of-the-art AI on the job, Anthropic managed to 'reduce flickering by 85%' on their TUI Claude Code client, which is built in *fucking React* and rendered by drawing the *entire* chat conversation each time (hence the flicker)." - User `airstrike` quotes a tweet from an Anthropic engineer (`trq212`): > "Most people's mental model of Claude Code is that 'it's just a TUI' but it should really be closer to 'a small game engine'." And details the pipeline: > "For each frame our pipeline constructs a scene graph with React then -> layouts elements -> rasterizes them to a 2d screen -> diffs that against the previous screen -> finally uses the diff to generate ANSI sequences to draw." Commenter `javcasas` responds: > "Yup. Overengineering." 4. **Sincerity vs. Cartel/Regulatory Capture**: Comments on the call for a "verifiable global AI pause" split between those who think it's a sincere safety effort and those who see it as anticompetitive. - User `fasterik` says: > "We should be skeptical of any major player that advocates for regulating their own industry. In practice, this just means increasing barriers to entry and making it harder to compete with them." - User `8note` says: > "its a cartel looking to end competition though... what happens if they get this regulatory capture is that all the frontier labs put effort into making inference cheaper, and become extraordinarily profitable, at the expense of us consumers..." This is a goldmine of evidence and perspectives on the "When AI Builds Itself" paper, and it links directly to our prior finding `anthropic-recursive-self-improvement-pause-call`. Let's do a quick check on other news this week. Wait, let's see if there are other major model releases or product updates this week. Let's do a search on "OpenAI" or "Google DeepMind" or "Anthropic" or "Grok" releases this week. Let's state intent then search.

Let's look at the fascinating news here: 1. **OpenAI Veteran Chip Expert Switches to Anthropic**: Clive Chan, "Employee No. 002" of OpenAI's self-developed chip project, has left OpenAI and joined Anthropic (announced June 7, 2026). This is a major talent defection that highlights the escalating chip and custom silicon battle between OpenAI and Anthropic. Let's research this further. 2. **Andrej Karpathy joined Anthropic**: Wait, Vaibhav Sisinty's list mentions "Andrej Karpathy left Tesla and OpenAI to join Anthropic's..." Wait, let's verify if Andrej Karpathy recently joined Anthropic or if that's old/new. Let's do a search on Andrej Karpathy Anthropic June 2026. 3. **Sam Altman Pushes Back on AI Approval Rules**: "OpenAI CEO Sam Altman urges U.S. lawmakers not to require AI model approvals before launch, backing more testing instead." This is a direct counter-reaction to policy discussions, especially since Trump signed NSPM-11 (which requires pre-release testing and accelerates national security AI). Let's search on Sam Altman pushing back on AI approval rules. 4. **Trump Proposes Direct Government Equity Stakes in AI Labs**: "Is AI Too Important To Leave To Private Investors? Trump Thinks So." We have a prior finding `us-government-proposes-equity-stakes-in-ai-labs` about this, but maybe there are fresh developments. 5. **Grok Updates**: "Elon Musk posted two brief updates on June 5 confirming a core model improvement to Grok and the addition of worktrees support." Let's check what xAI/Grok released this week. Let's do a search on Clive Chan switching to Anthropic and Sam Altman pushing back on AI approval rules. Let's state intent then search.

Ah, my search query was too complex/restrictive. Let's do a simpler query: `Clive Chan OpenAI Anthropic`. Let's state intent then search.

Let's fetch one of the decoders or times of india articles on Clive Chan's move to Anthropic. For example, `Anthropic poaches OpenAI's second-ever chip engineer as both companies race toward IPOs` (https://the-decoder.com/anthropic-poaches-openais-second-ever-chip-engineer-as-both-companies-race-toward-ipos/) or the newsbytesapp.com article. Let's fetch the Decoder article. Let's state intent then fetch.

Excellent! We have a highly detailed and concrete new finding. Let's look at the story of Clive Chan's defection: - Clive Chan, OpenAI's second-ever hardware employee in its custom chip program, has left OpenAI and joined rival Anthropic (announced June 7, 2026). - At OpenAI, Chan worked on custom chip design from scratch and was involved in the strategic partnership between OpenAI and Broadcom (which had reportedly hit snags over production costs and OpenAI's creditworthiness, requiring Microsoft to potentially buy 40% of them). - Prior to OpenAI (which he joined in January 2024), Chan spent 2.5 years at Tesla's Autopilot division working on custom chips for ML training (Tesla Dojo supercomputer team). - He joins Anthropic as a Member of Technical Staff on the pre-training team (or hardware optimization). His LinkedIn/X role description is "perplexity per picojoule" (referencing optimizing model performance per unit of energy). - This comes as Anthropic is reportedly weighing the idea of building its own custom AI chips (as reported by Reuters in April 2026) to boost margins, following Meta and OpenAI. Anthropic currently runs Claude on Google TPUs and Amazon Trainium chips, and recently signed a multi-gigawatt TPU deal with Google and Broadcom. - This creates a fantastic, highly relevant finding regarding the custom chip race and talent poaching, which connects to `anthropic-openai-confidential-ipo-filings` and `anthropic-apollo-blackstone-35b-chip-financing`. Let's do a search on Sam Altman's pushback on AI approval rules. This will give us another strong policy update. Let's state intent and search.

This is incredible. We have several major news pieces here: 1. **Sam Altman Opposes Pre-Launch AI Approval Rules on Capitol Hill**: - On **June 3, 2026**, OpenAI CEO Sam Altman was back in Washington, D.C., meeting with lawmakers (including House Speaker Mike Johnson) and White House officials. - Altman is pushing back against proposals that would require AI developers to obtain government approval/licenses before releasing models to the public. He argues that pre-release licensing or approval rules would slow down innovation and put the US at a disadvantage. - Instead, Altman is advocating for **mandatory AI risk evaluations** (testing) and wants the US government to provide more funding to the US AI Safety Institute (AISI) to conduct safety testing, rather than having the government act as a gatekeeper. - He also advocated for a **US sovereign wealth fund** to support AI infrastructure, faster data center permitting, and national security assessments. This is a critical development. 2. **OpenAI Agrees to Comply with Trump's Pre-Release AI Model Review Order**: - OpenAI's George Osborne confirmed at SXSW London that OpenAI will comply with President Trump's voluntary pre-release AI model review order (related to NSPM-11 and the 30-day model review mandate). 3. **OpenAI Planning a "Superapp" Ahead of IPO**: - Reports are claiming that OpenAI is developing a superapp that combines Codex, browser features, AI agents, image generation, and third-party app integrations to compete with Anthropic's ecosystem and ahead of its confidential IPO filing. 4. **Altman Met with Bernie Sanders to Discuss Public Ownership**: - Sam Altman met with Senator Bernie Sanders to discuss public ownership and direct government equity stakes in AI companies (which we saw in our prior findings). Let's fetch the Reuters or Cryptobriefing article on Sam Altman's Capitol Hill meetings: `OpenAI CEO Sam Altman advocates for mandatory AI evaluations on Capitol Hill` (https://cryptobriefing.com/altman-mandatory-ai-evaluations-capitol-hill/) or the Reuters article. Let's fetch the Cryptobriefing article. Let's state intent then fetch the page.

Ah, yes! The search budget is spent, which is perfectly fine because we have an incredible collection of rich, detailed, and highly specific primary evidence for our findings! Let's organize the findings we will write notes for: 1. **Apple WWDC 2026 Siri & iOS 27 AI Agent Releases** (to resolve the first watch): - Slug: `apple-siri-nvidia-blackwell-google-cloud` (Let's reuse this slug, or we can use a fresh one if we want, but let's check our prior notes: `apple-siri-nvidia-blackwell-google-cloud` is titled "Apple to Revamp Siri via Google Cloud and Nvidia Blackwell B200 with Confidential Compute". We can update this note in-place with the actual WWDC 2026 announcements!) Wait, the guidelines say: "When this cycle advances a subject you already have a finding for, call `write_note` with that SAME slug to revise it in place — an upsert that snapshots a new revision. Mint a fresh slug ONLY for a genuinely new subject." Perfect! I will update `apple-siri-nvidia-blackwell-google-cloud` with the details from WWDC 2026. 2. **SpaceX IPO S-1/A Prospectus & Compute Deals** (to update the SpaceX IPO preps): - Slug: `spacex-google-anthropic-compute-deals-ipo-preps` (Let's update this slug in-place with the June 3 S-1/A details, the $135.00 fixed price, the xAI/X merger, the $29.4B Google contract, the $1.25B/month Anthropic deal, the Bitcoin treasury, and the Cursor acquisition!) 3. **Clive Chan Defection / Chip Race**: - Slug: `clive-chan-exits-openai-joins-anthropic` (Fresh slug for a new subject) - Subject: Clive Chan, OpenAI's second-ever chip engineer, defecting to Anthropic. 4. **Sam Altman Pushes Back on AI Pre-Launch Approvals**: - Slug: `sam-altman-opposes-pre-launch-ai-approvals` (Fresh slug for a new subject) - Subject: Sam Altman's June 3 D.C. meetings opposing pre-launch licensing and advocating for risk evaluations, sovereign wealth fund, and permitting reform. 5. **Anthropic's "When AI Builds Itself" Paper & Developer Backlash**: - Slug: `anthropic-recursive-self-improvement-pause-call` (Let's update this slug in-place with the new metrics: 80% Claude-authored code, 8x code shipping, METR task-length doubling every 4 months, and the Hacker News/developer backlash on code slop and overengineering!) Let's resolve the first thread: `Apple WWDC 2026 Siri and iOS 27 AI Agent Releases` (id: `019e9bc8-00dd-7d1d-b044-b9c333710a79`). Wait, let's call `resolve_thread` first or we can do it after writing the notes. Let's do it after writing the notes to ensure the notes are persisted. Wait, let's look at the other threads: - `Meta Muse Spark Developer API Release` (id: `019e9889-2962-79c0-ba91-3aa809f9f6b9`). It is scheduled for June 30, 2026. Since today is June 9, 2026, it hasn't fired yet. But we have a prior note `meta-muse-spark-api-delays` or `meta-delays-muse-spark-api-bugs` that we can update or just register a watch event. Wait, do we have a watch registered for it? Yes, it's a one-shot watch with fire date June 30. We can keep it open. - `SpaceX and xAI Public Debut / IPO` (id: `019e9889-19fc-7370-b885-054439ee4234`). Fire date is June 12, 2026 (in 3 days). This watch hasn't fired yet, so we keep it open. Let's draft the updated notes carefully. ### Note 1: Apple Siri WWDC 2026 Overhaul - Slug: `apple-siri-nvidia-blackwell-google-cloud` - Title: Apple Rebuilds Siri on 1.2T-Parameter Google Gemini Model at WWDC 2026 - Content: Detail WWDC 2026 keynote, John Ternus taking over September 1, 2026, 1.2T-parameter model uses mixture-of-experts, complex queries routed to Nvidia Blackwell B200 chips in Google Cloud with hardware-based confidential compute. Deprecation of SiriKit in favor of Swift-based App Intents. Xcode 27 on-device AI. Per-intent privacy manifests. Let's write this note. I will state intent then call `write_note`.

Now let's write Note 2: SpaceX S-1/A Prospectus & Compute Deals. We will update `spacex-google-anthropic-compute-deals-ipo-preps` (SpaceX Preps for $1.77 Trillion IPO with Massive Compute Rental Deals for Google and Anthropic). Let's see what details we can add: - Target share price: $135.00 per share (fixed, no-range). - Target raise: $75.00 billion in Class A shares (555.56 million shares). - Implied valuation: $1.77 trillion. - IPO date: June 12, 2026 (Nasdaq ticker SPCX and Nasdaq Texas). - xAI/X merger: acquired in February 2026 in an all-stock common-control transaction, creating a three-pronged conglomerate (Launch, Starlink, AI/Cloud Compute). - Google Cloud Deal: Signed June 5, 2026. Google pays $920M/month from Oct 2026 through June 2029 ($29.4B total) for 110,000 Nvidia Blackwell GPUs/CPUs. Cancelable if not delivered by Sept 30, 2026. - Anthropic Deal: Anthropic pays $1.25B/month through May 2029 to rent idle compute capacity. - Starlink: 10.3M active subscribers in 164 countries, 9,600 satellites, $11.387B revenue and $4.423B operating income in 2025. - Bitcoin Treasury: 18,712 BTC ($1.3B - $1.5B). - Directed Share Program (DSP): 5% of IPO Class A shares ($3.75B) reserved for employees/friends with NO lock-up. - Cursor: Targeting a $60B all-stock acquisition of the popular AI coding assistant Cursor, with a $10B breakup fee. - Water Bottleneck: S-1/A warning on water resources for data center cooling. - Elon Musk controls 85% of voting power through Class B shares (10 votes each) while owning 42% of equity. Let's write this note. I will state intent then call `write_note`.

Now let's write Note 3: Clive Chan Defection / Chip Race. - Slug: `clive-chan-exits-openai-joins-anthropic` - Subject: Clive Chan, OpenAI's second-ever chip engineer, defecting to Anthropic. Let's state intent then call `write_note`.

Now let's write Note 4: Sam Altman Pushes Back on AI Pre-Launch Approvals. - Slug: `sam-altman-opposes-pre-launch-ai-approvals` - Subject: Sam Altman's June 3 D.C. meetings opposing pre-launch licensing and advocating for risk evaluations, sovereign wealth fund, and permitting reform. Let's state intent then call `write_note`.

Now let's write Note 5: Anthropic's "When AI Builds Itself" Paper & Developer Backlash. We will update `anthropic-recursive-self-improvement-pause-call` (Anthropic Warns of Recursive Self-Improvement and Calls for a Verifiable Global AI Pause). Let's see what details we can add: - Add the concrete metrics: Anthropic engineers now ship 8x more code per quarter than in 2024. - Over 80% of code merged into Anthropic's codebase is written by Claude, up from low single digits before Claude Code launched in Feb 2025. - METR benchmarks show the length of tasks frontier models can complete reliably is doubling every 4 months (down from every 7 months a year ago). - Claude generated 800 bug-fix pull requests reducing API errors by 1,000x (supervised by one engineer, which would have taken years for a human). - Developer backlash on Hacker News: * Skepticism that "Lines of Code" (LoC) is a valid productivity metric, as AI code is highly verbose and often bloated or buggy. * Anecdote from user `malfist` about a coworker submitting a 40k lines of code AI-generated PR across 600 files, which they refused to review. * Critique of Claude Code's overengineering: built in React (via the Ink framework) and has huge resource footprints. An Anthropic engineer (`trq212`) described it as "closer to a small game engine" that constructs a scene graph with React, rasterizes it, and diffs it at 60fps to generate ANSI sequences, which commenters called "profoundly stupid" and "sickening" for a TUI that consumes 1GB+ of RAM and 100% CPU. * Accusations of "regulatory capture" and "cartel" behavior, suggesting that Anthropic's call for a "global AI pause" is a PR strategy to freeze competition ahead of its IPO, slow down capex, and protect its market lead. Let's write this note. I will state intent then call `write_note`.

This is amazing! We have successfully updated and created all the findings. Let's double check if we need to record a watch or register any other threads. Let's see: 1. We have a watch for the Apple WWDC 2026 Siri and iOS 27 AI Agent Releases (id: `019e9bc8-00dd-7d1d-b044-b9c333710a79`). Let's call `record_event` on it to show it has fired and then `resolve_thread` to close it. Let's state intent then call `record_event`.

Excellent! Now let's call `resolve_thread` on the Apple WWDC thread. Let's state intent then call `resolve_thread`.

Awesome. Let's see if there are other watches we should register. What about Clive Chan's defection and the custom chip race? We should register a watch for custom silicon releases or Broadcom/OpenAI chip production milestones. What about SpaceX's IPO on June 12, 2026? We have an open watch for it already: `SpaceX and xAI Public Debut / IPO` (id: `019e9889-19fc-7370-b885-054439ee4234`). It will fire in 3 days. So we don't need to register a new one for that. What about Sam Altman's Capitol Hill meetings? He's advocating for permitting reform, risk evaluations, and a sovereign wealth fund. We can register a watch for any official US legislative or executive actions on a US AI Sovereign Wealth Fund or federal licensing/pre-approvals (e.g. if-then tripwire or thesis breaker). Let's register a watch for a US AI Sovereign Wealth Fund or federal AI licensing bills. Let's state intent then call `record_watch`.

Let's register another watch for the custom chip race, specifically OpenAI's Broadcom-produced custom chip tape-out or tape-in milestones, or Anthropic's custom chip design team formation. Let's state intent then call `record_watch`.

This is extremely thorough and complete. Let's double check if we missed any guidelines. - Name and link the actual thing: Yes, done. - Carry 1-2 representative verbatim quotes, copied exactly: Yes, done. - One note per theme: Yes, we have 4 distinct themes/notes: 1. `apple-siri-nvidia-blackwell-google-cloud` (Siri Rebuild / WWDC 2026) 2. `spacex-google-anthropic-compute-deals-ipo-preps` (SpaceX IPO S-1/A) 3. `clive-chan-exits-openai-joins-anthropic` (Clive Chan Defection) 4. `sam-altman-opposes-pre-launch-ai-approvals` (Sam Altman DC Meetings) 5. `anthropic-recursive-self-improvement-pause-call` (When AI Builds Itself / Backlash) - Reuse slugs of prior findings to update in place: Yes, we updated three prior findings (`apple-siri-nvidia-blackwell-google-cloud`, `spacex-google-anthropic-compute-deals-ipo-preps`, and `anthropic-recursive-self-improvement-pause-call`) and minted two fresh ones (`clive-chan-exits-openai-joins-anthropic` and `sam-altman-opposes-pre-launch-ai-approvals`). - Use wikilinks `[[slug]]` to link other findings: Yes, we linked `[[apple-siri-nvidia-blackwell-google-cloud]]` in our thread resolution note and have written clean markdown. - End the cycle by calling `finish_cycle`. Let's state intent then call `finish_cycle` to complete this research cycle.

This has been an incredibly productive research cycle! I have compiled all the findings, updated the living notes database, resolved the active thread, and registered new watches for future developments. Here is a summary of the findings and updates from this cycle: ### Summary of Research Cycle Findings #### 1. Apple WWDC 2026 Siri & iOS 27 Overhaul (Resolved Watch) At WWDC 2026 (theme "All systems glow"), Apple officially unveiled its built-from-the-ground-up **Siri overhaul**, powered by a custom **1.2-trillion-parameter Google Gemini model** (licensed for roughly **$1B/year**). * **Gemini on Blackwell:** Simple queries execute on-device, but complex queries route to Google's fleet of **Nvidia Blackwell B200 chips on Google Cloud** (as Apple's Private Cloud Compute was too slow at scale). Privacy is maintained via hardware-based **confidential computing** on those Blackwell chips. * **SiriKit Deprecation:** Apple deprecated SiriKit in favor of the Swift-based **App Intents framework**, giving developers 2–3 years to migrate. * **Developer Controls:** New **privacy manifest APIs** allow developers to declare per-intent whether cloud routing is permitted, addressing compliance for enterprise/healthcare apps. * *Living Note Updated:* [[apple-siri-nvidia-blackwell-google-cloud]] * *Watch Resolved:* `Apple WWDC 2026 Siri and iOS 27 AI Agent Releases` #### 2. SpaceX S-1/A Prospectus & Compute Deals SpaceX filed its final S-1/A amendment on June 3, 2026, ahead of its **June 12, 2026 IPO on the Nasdaq (ticker: SPCX)**. * **The Valuation & Raise:** Targeting a fixed share price of **$135.00** to raise **$75.00 billion** at a **$1.77 trillion valuation**. * **The xAI/X Merger:** In February 2026, SpaceX acquired xAI and X (Twitter) in an all-stock transaction under common control, creating a consolidated Launch, Starlink, and AI conglomerate. * **Compute Deals:** Google Cloud signed a **$29.4B contract** to pay SpaceX $920M/month through June 2029 for 110,000 Nvidia Blackwell GPUs. Anthropic agreed to rent idle compute for **$1.25B/month** through May 2029. * **Cursor Acquisition:** The prospectus discloses a targeted **$60 billion all-stock acquisition of Cursor**, the popular AI coding assistant. * *Living Note Updated:* [[spacex-google-anthropic-compute-deals-ipo-preps]] #### 3. Clive Chan Defects to Anthropic Clive Chan, the **second-ever hardware hire at OpenAI** and a key lead in its custom chip program (formerly of Tesla's Dojo supercomputer team), has left OpenAI and joined rival **Anthropic as Member of Technical Staff**. * **Pre-Training & Optimization:** Chan joins Anthropic's pre-training team with the role description **"perplexity per picojoule,"** focusing on optimizing model performance per unit of energy consumed. * **Custom Chip Race:** The hire comes as Anthropic reportedly weighs building its own custom AI chips to boost margins, following Meta and OpenAI. * *New Note Created:* [[clive-chan-exits-openai-joins-anthropic]] #### 4. Sam Altman Opposes Pre-Launch AI Approvals On June 3, 2026, OpenAI CEO Sam Altman met with House Speaker Mike Johnson and bipartisan congressional leaders to lobby against proposals requiring federal pre-launch approvals or licenses for powerful AI models. * **Lobbying Playbook:** Altman urged lawmakers to focus on **mandatory risk evaluations (testing)** after models are built, rather than government pre-approvals (permissions), which he argued would slow down innovation. * **Infrastructure & Permitting:** He urged faster permitting for data centers and energy infrastructure and proposed a **US sovereign wealth fund** to give Americans an equity stake in AI. * *New Note Created:* [[sam-altman-opposes-pre-launch-ai-approvals]] #### 5. Backlash Over Anthropic's "When AI Builds Itself" Paper Anthropic's June 4 paper claiming that Claude writes **80% of its codebase** and that its engineers ship **8x more code** than in 2024 faced intense criticism. * **Lines of Code Metric Backlash:** Developers on Hacker News criticized using LoC as a KPI, sharing anecdotes of AI generating massive "code slop" and verbose, buggy PRs (such as one coworker submitting a 40k lines of code AI PR that was impossible to review). * **Claude Code Overengineering:** Critics lambasted Claude Code's TUI client for consuming **1GB+ of RAM and 100% CPU**, noting it is built on React (via the Ink framework) and runs a 60fps rendering loop like a "small game engine." * **IPO Hype & Cartel Accusations:** Critics argued the call for a "global AI pause" is a cynical PR strategy to protect its lead, freeze competition, and slow down capex to inflate margins ahead of its IPO. * *Living Note Updated:* [[anthropic-recursive-self-improvement-pause-call]] *** ### Registered Watches (Ongoing Monitoring) 1. **US AI Sovereign Wealth Fund Legislative or Executive Actions:** Tracking any official bills or orders establishing a US AI Sovereign Wealth Fund. 2. **OpenAI and Anthropic Custom Chip Milestones:** Tracking OpenAI's Broadcom tape-out progress and Anthropic's custom silicon development team. 3. **SpaceX SPCX IPO Listing (Fires June 12, 2026):** Monitoring the market debut and subsequent financial reports of the consolidated SpaceX/xAI/X conglomerate.