Skip to content
Astrohacker
Day 31Singularity Watch

SpaceX Catches Starship From Orbit as the Fireworks AI Breach Reaches Uber and Dwarkesh Patel Moves Superintelligence to 2030

Starship Ship 42 became the first Starship upper stage caught by its launch tower, the Fireworks AI breach behind Cursor's incident reached Uber drivers and Notion, and Dwarkesh Patel wrote that he now expects superintelligence by 2030.

Two arms extend from a lattice launch tower to hold Starship Ship 42 upright above the ground.

SpaceX Catches Starship From Orbit as the Fireworks AI Breach Reaches Uber and Dwarkesh Patel Moves Superintelligence to 2030

Space: The First Ship Catch

SpaceX caught Ship 42 with the arms of its Starbase launch tower at 3:20 PM CST on November 14, about 22.5 hours after Starship Flight 15 lifted off. It was the first catch of a Starship upper stage; SpaceX had caught the Super Heavy booster on three earlier flights but had never tried with the ship. [1][2]

SpaceX said Starship reached orbit, deployed 26 Starlink V3 satellites, coasted for more than 20 hours, relit a single Raptor engine for its deorbit burn, and lit three Raptors for a landing flip into the tower’s arms. [2]

The booster did not make it back. SpaceX said its boostback burn ended early due to low fuel mass, with the root cause under investigation; its guidance triggered an automated abort to an offshore landing, and its flight termination system fired after the landing burn over the Gulf. [2]

“That was the missing link,” SpaceX’s Dan Huot said on the webcast. “We caught a Starship, and now everything changes.” Space.com reported that SpaceX calls the catch key to the launch cadence needed for larger constellations, including a planned AI data-center constellation of a million spacecraft that Elon Musk calls Starmind. [1]

Ryan wrote at 3:25 PM CST, “Amazing! Congrats @elonmusk and @SpaceX team!” He surfaced posts from NASA Administrator Jared Isaacman, from Xander, who wrote that he had spent six years working toward the moment and that the heat shield his team built helped bring the ship home, and from others who noted the ship flew 14 orbits before its return. [3]

Security: The Fireworks Breach Spreads

Cursor published an update on November 13 on its October security incident. It said an attacker obtained an access key Cursor had issued to Fireworks AI, which Fireworks had exposed, and used it to copy coding agent traces, mostly from 2025, from sessions not in Privacy Mode; a small number of the traces held credentials that had not been redacted. [4]

From those credentials the attacker reached SaaS tools used by Cursor, then archived source code, support data and some credentials users had sent to support. On October 29 the attacker used a stolen credential to place malicious copies of Anthropic’s Claude Code extension on the cache serving Cursor’s extension marketplace, where at most 5,000 users may have received it during about 11 hours. [4]

Cursor said Anthropic did not publish the copy and that the Open VSX source never had the malware. It said the last attacker activity it saw was on November 1, that it believes the incident is contained, and that it has moved all inference off Fireworks. [4]

TechCrunch reported on November 13 that Uber is notifying drivers that their personal and driver’s license information was stolen in an October breach at a tech provider, which TechCrunch identified as Fireworks AI. State filings list 2,567 people in Massachusetts, 3,430 in South Carolina and 1,873 in Rhode Island. Notion also confirmed its data was taken in the Fireworks hack. [5]

Ryan surfaced Haris Nadeem’s post that one leaked set of credentials at Fireworks had turned into incidents at Cursor, Notion and Uber, a summary of the attack chain by Prasad Pilla, and 1Password’s breakdown of how the attack began with an exposed vendor key. [6]

Research: Superintelligence by 2030

Dwarkesh Patel published an essay on November 13, “The avalanche that hit math is coming for AI development.” He wrote that earlier this year he would have put superintelligence at least the better part of a decade away, and that he now expects AIs as good as the best humans at AI research within a couple of years, possibly a couple of months. [7]

He expects automated AI research to compress so much progress that by 2030, plausibly as early as next year, AIs will be at least as good as the best human experts in every field. He cited METR’s finding that Anthropic’s rate of progress on well-scoped tasks is 2x what it was a year ago, and Anthropic’s report that Claude led more than 26% of its AI R&D work as of August. [7]

A footnote imagines “SlothAI,” a lab with OpenAI’s compute whose researchers work one hour a week, and argues that compared to a lab of automated researchers, OpenAI is SlothAI. Dan Robinson wrote that the intuition “hit me like a sledgehammer.” Steve Hou called it a string of words; Ryan replied at 5:47 AM CST: “Not only human intelligence not the ceiling, it is the floor of what’s possible with AI.” [8]

Patel added that even without a full singularity, the intelligence explosion might compress four years of progress into one. Noah Smith agreed. Ryan quoted a post contrasting Patel’s earlier timelines with his new ones at 5:15 AM CST: “Accelerating timelines.” [9]

Compute: Nvidia After Huang

A screenshot posted by Yuchen Jin on November 13 shows Dario Amodei, chief executive of Anthropic, listed as a public signer on the Stand with Jensen letter, dated November 13. Jin wrote that it “was not on my bingo card,” and Beff Jezos wrote that it “violated my priors as well.” Ryan surfaced the post. [10]

Eric S. Raymond wrote that he had signed the petition for Huang’s reinstatement, mainly because Huang is “one of the most powerful people speaking in favor of open-weight and open-source models.” Nina Schick criticized commentary on Huang’s fall, citing a Bloomberg opinion piece; the Bloomberg article could not be read. Ryan surfaced both posts. [11]

Jukan wrote that industry experts told him Nvidia’s senior engineers, many already financially free, may leave without Huang, and that competitors are already calling them. That is an account of unnamed sources. [12]

Labs: Google’s Week

Sundar Pichai listed what Google shipped in one week: two eighth-generation TPUs, Gemini 4 Carbon, an AlphaProof Code verification of the Signal protocol, Gemini Robotics 2.5, AlphaGenome 2 for disease research, a Gemini Decisions API, Gemini 4 Hydrogen and pre-orders for intelligent eyewear. Beff Jezos wrote, “Big G is back.” Ryan surfaced it. [13]

A post Ryan surfaced said Google had moved into second place, ahead of OpenAI, without naming the ranking. [14]

Media: An Official Trailer

An account called TUPACABRA posted “NUREMBERG - Official Trailer” on November 13. Jason wrote that it was the first trailer to cross the uncanny valley for him, and that it is both a blockbuster and AI slop. [15]

Robert Scoble called it a step change in AI video, and a studio head told Michael Davidson that Hollywood is in denial. Another poster argued instead that such tools democratize cinema on a scale comparable to the printing press. Ryan surfaced all of these. [16]

At 10:45 AM CST Ryan posted, “so long, legacy media / i am the news now,” with a screenshot of the Astrohacker News front page, subtitled “Watching the Singularity,” listing Singularity Watch Days 27 through 30. [17]

Agents: Software Whose User Is an Agent

Ryan published an essay, “Software for Agents: A New Market Nobody Is Counting,” dated November 13 and posted at 12:24 PM CST on November 14. Its thesis is that a category of software humans buy but never touch is forming, and that no market report counts it. [18]

Of nine products he checked, including E2B, Browserbase and AgentMail, seven bill by what the agent consumes and none charges per agent. He counted 13 YC companies from 2025 and 2026 batches whose user is an agent, and four 2026 payment launches built for agents as payers, from Stripe, Tempo, Coinbase and the Linux Foundation. He discloses that he is building one such product, Botpenny. [18]

The market reports he read size the agents themselves, at 30% to 47% annual growth, and none has a segment for tools agents use. He calls the size of that market unmeasured. [18]

At 11:53 AM CST he showed his new personal home page, which he wrote was created by Claude, ChatGPT and Grok “working in harmony.” [19]

Work: Agentic Engineering

Amir Salihefendić wrote that he had merged 434 pull requests across 28 repositories with zero reverts in 14 days, after Brian Armstrong reported 46 across 10 repositories, part time, “by talking to agents.” Ryan wrote at 5:45 AM CST: “Agentic Engineering is the new Software Engineering.” [20]

Responding to a solo developer who wrote that agents let her make hundreds of decisions a day across several projects, Ryan wrote at 6:08 AM CST, “AI-first is hands-down faster.” He also wrote that the prompt “is really just whatever context you give it access to,” and, “You can manage a team of 1 million agents.” [21]

Blake Scholl of Boom wrote that his team found a design mistake in its first turbine and had a replacement part made overnight, at a cost of about 24 hours. Ryan surfaced the post. [22]

In a clip Rohan Paul posted, Elon Musk said agentic AI traffic will be 1,000 times human internet traffic within five years and that, per Cloudflare, agent traffic has already passed human traffic. Ryan surfaced a reply arguing the forecast is low. [23]

Tools: Subagents and Harnesses

Vercel released fx v0.0.17 on November 14, adding subagents that work alongside the main agent and tool definitions 48.0% smaller. Guillermo Rauch wrote that all his projects are built by planning with the smartest model and delegating to fast-model subagents. Ryan surfaced it. [24]

lauren, of the Grok Bot team, posted a conference talk on how the team ships quickly without compromising quality, and a user reported the team said in a Q&A that someone is working on Apple Health support. Ryan surfaced both. [25]

Society: A Papal Warning on Transhumanism

Pope Leo XIV wrote on November 13 that the promises of transhumanism and posthumanism are multiplying, that scientific progress must not be confused with a privilege for “millionaires seeking immortality,” and that in Christian anthropology a limitation “is not a manufacturing defect.” Ryan quoted the post twice, the second time at 6:31 AM CST: “Follow this man.” [26]

Astrohacker

@AstrohackerLabs published Singularity Watch, Day 30 at 4:21 AM CST on November 14, and reposted Ryan’s posts through the day. [27]

Sources:

← Back to News