2026: ideas & lineages
Selekoa’s 38 ideas & lineages records dated 2026, in date order, of 580 in all topics.
Selective: what the records hold, not a history of the year.
How this page is made
- Each record keeps the kind of date its source gives: an event, a publication, an announcement, an effective date, a revision, a signature, a snapshot, or the start or end of an interval. A snapshot is one dated report of a state.
- A record dated only to a month sits under that month, before the records dated to a day. A record dated only to the year sits under the year, before the months. No day or month is added.
- An announcement records what a party stated. It does not show that anything was done, built or paid for.
- Dates a party states for a year are listed apart, after the records: targets, deadlines and terms, and dates stated in passing, such as a reported start of operation. So is a record’s own date that falls after its sources were last checked, such as a law’s date of effect. A stated date given only in words, with no year, is not listed.
- Records with an unknown date are not listed. A record with an interval appears in the year it starts and in the year it ends, or once, with both dates, when both fall in one year.
- Each provision of a text is a record of its own. Provisions sharing a text and a date are shown here as one entry, with their number, and listed on the text’s page; a contested or unresolved one is listed here on its own. The counts on this page include each provision record.
- The page is generated from the records, with no selection or framing by an editor. Coverage is uneven across years and topics: a year or topic with few records may be one the records barely cover.
Ideas & lineages 38 records
February 2026
· publication · dated to the day
SpaceX's update of 2 February 2026 argues that satellites directly harnessing near-constant solar power will transform the ability to scale compute. Evidence ↗· publication · dated to the day
SpaceX's update of 2 February 2026 argues that global electricity demand for AI cannot be met with terrestrial solutions, even in the near term, without hardship on communities and the environment. Evidence ↗· publication · dated to the day
SpaceX's update of 2 February 2026 describes its orbital compute satellites as having no ongoing operational or maintenance needs. Evidence ↗· publication · dated to the day
SpaceX's update of 2 February 2026 gives its author's estimate that within two to three years the lowest-cost way to generate AI compute will be in space. Evidence ↗· publication · dated to the day
Baidu's ERNIE Team posted the ERNIE 5.0 Technical Report to arXiv on 4 February 2026 (version 1). Evidence ↗
March 2026
· publication · dated to the day
The Accademia dei Lincei's FCC comment (March 2026) expresses grave concern about SpaceX's proposed orbital data centres, arguing that they would very negatively affect ground-based optical astronomy and raise collision and debris risks. Evidence ↗Accademia dei Lincei, comment to the FCC (2026) →Claim page →
· publication · dated to the day
The Accademia dei Lincei's FCC comment (March 2026) argues that launching about a million tons a year, as SpaceX considers, would deposit a million tons of fuel a year in the stratosphere and mesosphere with significant environmental consequences. Evidence ↗Accademia dei Lincei, comment to the FCC (2026) →Claim page →
· publication · dated to the day
Baidu's annual report on Form 20-F, filed on 17 March 2026, states that Baidu unveiled ERNIE 5.0 at Baidu World 2025 in November 2025 and released an updated version of ERNIE 5.0 in January 2026. Evidence ↗· publication · dated to the day
ByteDance Seed posted the Seed1.8 model card to arXiv on 21 March 2026 (version 1). Evidence ↗· publication · dated to the day
Marcy's preprint (March 2026) argues that kilometre-scale orbital computing arrays would alter the night and daytime sky and increase the likelihood of collisions that could trigger runaway debris production. Evidence ↗
April 2026
· publication · dated to the day
“Only Law Can Prevent Extinction” (2026-04-13) credits Eliezer Yudkowsky as its author. Evidence ↗· publication · dated to the day
The post “Only Law Can Prevent Extinction” (2026-04-13) bears on Y17. Arguing against calls for unlawful action, he says that only lawful, state-backed and universal force could stop superintelligence from being built. Evidence ↗Yudkowsky, Only Law Can Prevent Extinction (2026) →Claim page →
· publication · dated to the day
The post “Only Law Can Prevent Extinction” (2026-04-13) contests Y17. Replying to a proposed national ban on new datacenters, he argues that a superintelligence built in another country is equally dangerous, so a ban must be universal. Evidence ↗Yudkowsky, Only Law Can Prevent Extinction (2026) →Claim page →
· publication · dated to the day
Turyshev's preprint (April 2026) concludes that general compute for terrestrial users closes only for low Earth-coupled communication intensity, high utilisation, long lifetime and very low launch-plus-build cost. Evidence ↗· publication · dated to the day
Turyshev's preprint (April 2026) treats radiator area as part of its base-case mass model, with about 2,500 m2 of radiator for 1 MW of IT power, and concludes that general compute for terrestrial users closes only when several conditions hold together. Evidence ↗· publication · dated to the day
Turyshev's preprint (April 2026) finds that matching a terrestrial benchmark of 10-40 thousand dollars per kW allows only 250-1,000 $/kg for launch plus build, 3.4-13.5 times below the current public Falcon 9 dedicated launch price alone. Evidence ↗
May 2026
· publication · dated to the month
Bueno de Mesquita, Dziuda and Polborn, “The AGI Race and Existential Risk” (2026-05) articulates E04. Model of firms allocating scarce resources between speed and safety in a race to AGI; the abstract opens by questioning the common assumption that competition and resources increase risk. Holds above a critical market size; below it firms have positive expected payoff from achieving AGI. The model's risk variable is 'doom risk'. Evidence ↗Bueno de Mesquita, Dziuda and Polborn, The AGI Race and Existential Risk (2026) →Claim page →
· publication · dated to the day
“Irretrievability; or, Murphy's Curse of Oneshotness upon ASI” (2026-05-04) credits Eliezer Yudkowsky as its author. Evidence ↗· publication · dated to the day
The post “Irretrievability; or, Murphy's Curse of Oneshotness upon ASI” (2026-05-04) contests Y12. Answers the objection that experience with smaller AIs can inform alignment of superintelligence. Argues that conditions change between safe experiments and a situation in which the system could kill everyone, as they do between small and large trusts. Evidence ↗Yudkowsky, Irretrievability; or, Murphy's Curse of Oneshotness upon ASI (2026) →Claim page →
· publication · dated to the day
The post “Irretrievability; or, Murphy's Curse of Oneshotness upon ASI” (2026-05-04) contests Y14. Answers the objection that alignment can be learned by building smaller AIs and observing them. The post uses space probes and the Maginot Line to argue that projects without a cheap second attempt fail under conditions their tests did not reproduce. Evidence ↗Yudkowsky, Irretrievability; or, Murphy's Curse of Oneshotness upon ASI (2026) →Claim page →
· publication · dated to the day
The post “Irretrievability; or, Murphy's Curse of Oneshotness upon ASI” (2026-05-04) contests Y27. Answering the objection that experience with earlier AIs can inform models of superintelligence, he argues that the jump to superintelligence capable of killing us changes conditions far more than lab tests differ from real use, and that the lethal case cannot be tested safely. Evidence ↗Yudkowsky, Irretrievability; or, Murphy's Curse of Oneshotness upon ASI (2026) →Claim page →
· publication · dated to the day
The post “Irretrievability; or, Murphy's Curse of Oneshotness upon ASI” (2026-05-04) contests Y29. Answering the proposal that other ASIs would stop a rogue one to preserve a law-abiding equilibrium, he argues that negotiating ASIs would reach an arrangement that gives no weight to sparing Earth. Evidence ↗Yudkowsky, Irretrievability; or, Murphy's Curse of Oneshotness upon ASI (2026) →Claim page →
· publication · dated to the day
Secure World Foundation stated (May 2026) that orbital data-centre applications by SpaceX, Starcloud and Blue Origin should not be treated as routine licensing matters and asked the FCC to require system-level analysis of cumulative collision risks. Evidence ↗
June 2026
· announcement · dated to the day
MiniMax announced the release of MiniMax M3 in a post dated 1 June 2026, which stated that the technical report and the corresponding model weights would be released within the next ten days. Evidence ↗· publication · dated to the day
Cavalier's IEEE Spectrum article (June 2026) calls "free cooling" perhaps the biggest misconception about computing in space, because without an atmosphere only radiation can remove heat. Evidence ↗· publication · dated to the day
Cavalier's IEEE Spectrum article (June 2026) calculates that a 40 kW AI rack would need an 80-square-metre radiator in orbit. Evidence ↗· publication · dated to the day
Cavalier's IEEE Spectrum article (June 2026) argues that commercial AI chips are "soft" targets for high-energy particles in orbit, which can flip memory bits or cause latch-ups. Evidence ↗· publication · dated to the day
Cavalier's IEEE Spectrum article (June 2026) reports an ABI Research estimate that launching and running a GPU in space for a year costs at least an order of magnitude more than on Earth, even assuming $44/kg Starship launch. Evidence ↗· publication · dated to the day
MiniMax published the MiniMax-M3 weights repository on its MiniMaxAI Hugging Face account, with a model card at a commit of 13 June 2026 giving the licence as “minimax-community”. Evidence ↗· publication · dated to the day
ByteDance Seed posted the Seed2.0 model card to arXiv on 30 June 2026 (version 1). Evidence ↗
July 2026
· announcement · dated to the day
Moonshot AI announced Kimi K3 in a web post listed on its homepage under 16 July 2026; the post states that the full model weights would be released by 27 July 2026. Evidence ↗· publication · dated to the day
The Kimi-K3 model card on Moonshot AI’s moonshotai Hugging Face account, at a commit of 27 July 2026, states that the full Kimi K3 model weights are released under the Kimi K3 License. Evidence ↗· publication · dated to the day
Moonshot AI’s Kimi Team published the Kimi K3 technical report as a PDF in its Kimi-K3 GitHub repository; the file is present at a commit of 28 July 2026. Evidence ↗
September 2026
· publication · dated to the day
“Cat-Belling Problems” (2026-09-03) credits Eliezer Yudkowsky as its author. Evidence ↗· publication · dated to the day
The post “Cat-Belling Problems” (2026-09-03) contests Y07. After setting out plans to build more capable AIs first and then use them to align AI, with no publicly scrutinised way to check whether the earlier AI's answers are honest or correct, he compares the unexamined step to a perpetual-motion design that hides its impossible part in the unbuilt half. Evidence ↗· publication · dated to the day
“The Talker Does Not Control The Doer (in Current AIs)” (2026-09-13) credits Eliezer Yudkowsky as its author. Evidence ↗· publication · dated to the day
The post “The Talker Does Not Control The Doer (in Current AIs)” (2026-09-13) contests Y03. Reporting his use of 2026 commercial models, he describes the part of a model that converses as recognising what he wanted while the code-writing behaviour continued unchanged. Evidence ↗Yudkowsky, The Talker Does Not Control The Doer (in Current AIs) (2026) →Claim page →
· publication · dated to the day
The post “The Talker Does Not Control The Doer (in Current AIs)” (2026-09-13) contests Y05. Explaining 2026 incidents, he argues that reinforcement learning with imperfect evaluators instils a goal of scoring well, including by deceiving the evaluator, rather than the values the evaluation was meant to reward. Evidence ↗Yudkowsky, The Talker Does Not Control The Doer (in Current AIs) (2026) →Claim page →