2024: ideas & lineages
Selekoa’s 48 ideas & lineages records dated 2024, in date order, of 407 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 48 records
January 2024
· publication · dated to the day
RAND's report on the operational risks of AI in large-scale biological attacks (RR-A2977-2, 25 January 2024) states that it found no statistically significant difference in the overall viability of biological weapon attack plans generated with or without the aid of the large language models it tested, and that such planning is beyond the capability frontier of the models available in summer 2023. Evidence ↗The Operational Risks of AI in Large-Scale Biological Attacks: Results of a Red-Team Study →Claim page →
February 2024
· publication · dated to the day
"On the Societal Impact of Open Foundation Models" (arXiv v1, 27 February 2024) states that across several misuse vectors, biological weapons among them, current research is insufficient to characterise the marginal risk of open foundation models relative to pre-existing technologies. Evidence ↗On the Societal Impact of Open Foundation Models →Claim page →
March 2024
· publication · dated to the month
“Scenarios for the Transition to AGI” (2024-03) credits Anton Korinek as one of its 2 authors. Evidence ↗· publication · dated to the month
“Scenarios for the Transition to AGI” (2024-03) credits Donghyun Suh as one of its 2 authors. Evidence ↗· publication · dated to the month
Korinek and Suh, “Scenarios for the Transition to AGI” (2024-03) bears on E01. Following Aghion et al. (2019) and Trammell and Korinek (2023), the model shows a 'type II singularity' (infinite output in finite time) once automation passes a threshold. Evidence ↗Korinek and Suh, Scenarios for the Transition to AGI (2024) →Claim page →
· publication · dated to the month
Korinek and Suh, “Scenarios for the Transition to AGI” (2024-03) bears on E02. The paper models wages and output under scenarios that 'may culminate' in AGI. Evidence ↗Korinek and Suh, Scenarios for the Transition to AGI (2024) →Claim page →
· publication · dated to the month
Korinek and Suh, “Scenarios for the Transition to AGI” (2024-03) articulates E05. Final list of next steps after a 'compute-centric' model of the transition to AGI. Evidence ↗Korinek and Suh, Scenarios for the Transition to AGI (2024) →Claim page →
· publication · dated to the month
Korinek and Suh, “Scenarios for the Transition to AGI” (2024-03) bears on E05. Followed by Good (1965) on an intelligence explosion, Bostrom (2014) on superintelligence and Yudkowsky (2013) on 'several of the economic implications of the transition to AGI'. Literature review entry on Jones's existential-risk paper. Evidence ↗Korinek and Suh, Scenarios for the Transition to AGI (2024) →Claim page →
April 2024
· publication · dated to the day
Acemoglu, “The Simple Macroeconomics of AI” (2024-04-05) articulates G08. A task-based estimate using Hulten's theorem: the share of tasks exposed to AI times the average task-level cost saving. The GDP effect after an assumed investment response; under the investment assumptions of earlier automation the upper bound rises to 1.6–1.8%. Evidence ↗Acemoglu, The Simple Macroeconomics of AI (2024) →Claim page →
· publication · dated to the day
Acemoglu, “The Simple Macroeconomics of AI” (2024-04-05) bears on G09. Compares his ten-year estimate with larger forecasts. A footnote cites Davidson (2021) among those who expect "explosive" growth. Evidence ↗Acemoglu, The Simple Macroeconomics of AI (2024) →Claim page →
· publication · dated to the day
Acemoglu, “The Simple Macroeconomics of AI” (2024-04-05) articulates M08. Acemoglu applies a Hulten's-theorem task framework to exposure and cost-saving estimates to bound AI's ten-year TFP effect, against much larger forecasts (Goldman Sachs, McKinsey). Allowing for hard-to-learn tasks where AI gains are smaller lowers the ten-year TFP bound. Acemoglu converts the TFP bounds to GDP effects under two investment-response assumptions. The author-hosted version of 5 April 2024 gives no more than 0.71% over ten years, and less than 0.55% allowing for hard tasks; the NBER version of May 2024 gives 0.66% and 0.53%. Evidence ↗Acemoglu, The Simple Macroeconomics of AI (2024) →Claim page →
· publication · dated to the day
An ITIF commentary of 30 April 2024 states that experts, having studied the issue more closely, have concluded that large language models do not introduce a new set of capabilities or significantly enhance existing ones that could lead to a biological attack. Evidence ↗The Shift in Rhetoric on AI and Biothreats Is a Lesson on the Risks of Premature Regulation →Claim page →
May 2024
· publication · dated to the month
“The Simple Macroeconomics of AI” (2024-05) credits Daron Acemoglu as its author. Evidence ↗· publication · dated to the month
Acemoglu, “The Simple Macroeconomics of AI” (2024-05) bears on E01. Explains the paper's scope; the example given is DeepMind's protein-folding and crystal-structure work. Evidence ↗Acemoglu, The Simple Macroeconomics of AI (2024) →Claim page →
· publication · dated to the month
Acemoglu, “The Simple Macroeconomics of AI” (2024-05) bears on E02. Footnote 1 cites Korinek and Suh (2024) and those who see a confirmation of Kurzweil's (2005) singularity forecast; Acemoglu contrasts 'more grounded' forecasters. Evidence ↗Acemoglu, The Simple Macroeconomics of AI (2024) →Claim page →
· publication · dated to the month
Acemoglu, “The Simple Macroeconomics of AI” (2024-05) bears on E05. First of four possible medium-run macro effects of AI; the paper analyses the more modest task-level effects. Evidence ↗Acemoglu, The Simple Macroeconomics of AI (2024) →Claim page →
· publication · dated to the month
Acemoglu, “The Simple Macroeconomics of AI” (2024-05) articulates M02. Acemoglu models AI in a task framework where automation expands the set of tasks allocated to capital, alongside task complementarity, deepening and new tasks. Evidence ↗Acemoglu, The Simple Macroeconomics of AI (2024) →Claim page →
· publication · dated to the month
Acemoglu, “The Simple Macroeconomics of AI” (2024-05) bears on M05. Acemoglu lists a revolutionary, near-singularity scenario among four possible macro effects, says it cannot be ruled out, and excludes it from his analysis. Acemoglu argues that better wage and productivity outcomes depend on AI creating new tasks for workers, which he says is feasible in theory but not the industry's current focus. Evidence ↗Acemoglu, The Simple Macroeconomics of AI (2024) →Claim page →
· publication · dated to the month
Acemoglu, “The Simple Macroeconomics of AI” (2024-05) articulates M06a. Using Acemoglu and Restrepo (2022) general-equilibrium estimates, Acemoglu finds AI's wage effects more evenly spread than earlier automation but negative for low-education women and widening the capital–labour income gap. Evidence ↗Acemoglu, The Simple Macroeconomics of AI (2024) →Claim page →
· publication · dated to the month
Acemoglu, “The Simple Macroeconomics of AI” (2024-05) bears on M06b. Acemoglu argues that new-task creation with better wage outcomes may need new institutions and regulation and a reorientation of the AI industry. Evidence ↗Acemoglu, The Simple Macroeconomics of AI (2024) →Claim page →
· publication · dated to the month
Acemoglu, “The Simple Macroeconomics of AI” (2024-05) bears on M07. Acemoglu converts Eloundou et al.'s task exposure measure into a wage-bill-weighted share of US labour tasks exposed to AI. Evidence ↗Acemoglu, The Simple Macroeconomics of AI (2024) →Claim page →
June 2024
· publication · dated to the day
The ChatGLM report describes instruction tuning followed by SFT and RLHF, with ChatGPT explicitly named as a motivation. Evidence ↗· publication · dated to the day
Jie Tang appears in the ChatGLM report’s Team GLM author list. Evidence ↗· publication · dated to the day
Tsinghua University is the other named institutional affiliation of the ChatGLM report. Evidence ↗· publication · dated to the day
Zhipu AI is a named institutional affiliation of the joint ChatGLM report. Evidence ↗· publication · dated to the day
Thales Alenia Space announced the results of the ASCEND feasibility study on 27 June 2024. Evidence ↗· publication · dated to the day
The ASCEND results release (June 2024) states that space data centres would not require water to cool them. Evidence ↗· publication · dated to the day
The ASCEND results release (June 2024) states that, to significantly reduce the CO2 emissions of data processing and storage, space data centres would require a launcher ten times less emissive over its lifecycle. Evidence ↗
July 2024
· publication · dated to the day
Mark Zuckerberg authored Meta’s July 2024 open-AI strategy essay. Evidence ↗
September 2024
· publication · dated to the month
Lumen Orbit (now Starcloud) published the white paper "Why we should train AI in space", whose version 1.01 is dated September 2024. Evidence ↗· publication · dated to the month
The Lumen Orbit white paper (September 2024) argues that data centres in orbit can use solar power around the clock, unhindered by night, weather and the atmosphere. Evidence ↗· publication · dated to the month
The Lumen Orbit white paper (September 2024) argues that, if best practices are followed, terrestrial astronomy is unlikely to be affected by orbital data centres in its chosen dawn-dusk orbit, while naming debris mitigation as a key hurdle. Evidence ↗· publication · dated to the month
The Lumen Orbit white paper (September 2024) argues that orbital data centres can lower cooling costs by passive radiative cooling to space. Evidence ↗· publication · dated to the month
The Lumen Orbit white paper (September 2024) argues that orbital data centres can be scaled almost indefinitely without the physical or permitting constraints faced on Earth. Evidence ↗· publication · dated to the month
The Lumen Orbit white paper (September 2024) assumes a long-term launch price of about $30 per kilogram and argues that at such prices launch is no longer a primary cost driver. Evidence ↗· publication · dated to the month
The Lumen Orbit white paper (September 2024) states that a gigawatt-scale orbital data centre would need radiators dissipating gigawatts of heat, by far the largest deployed in space, while estimating them at under half the solar-array size. Evidence ↗· publication · dated to the month
The Lumen Orbit white paper (September 2024) states that shielding is required against radiation-induced latch-up, transients and total-dose effects in sensitive components, while arguing that shielding mass per unit of compute falls with container size. Evidence ↗· publication · dated to the month
The Lumen Orbit white paper (September 2024) states that its design lets compute containers and other modules be swapped out to replace old or faulty equipment. Evidence ↗· publication · dated to the day
METR’s o1-preview report explicitly reuses the general-autonomy test suite from its GPT-4o report. Evidence ↗· publication · dated to the day
Qwen Team published the Qwen2.5 announcement later cited in DeepSeek-R1. Evidence ↗· publication · dated to the day
“The Sun is big, but superintelligences will not spare Earth a little sunlight” (2024-09-23) credits Eliezer Yudkowsky as its author. Evidence ↗· publication · dated to the day
The post “The Sun is big, but superintelligences will not spare Earth a little sunlight” (2024-09-23) contests Y18. Concludes his reply to the claim that, because the solar system is large, superintelligences will leave Earth alone: a tiny cost is still a cost to minds that do not value us. Answering the claim that a superintelligence would leave Earth alone because the Solar System is big, he argues by analogy that great wealth does not imply giving a small share to any particular party. Replies to the suggestion that humanity could trade with superintelligences for sunlight: Ricardo's law assumes each party's labour goes on existing and does not show that keeping humans is more profitable than using their share. Evidence ↗Yudkowsky, The Sun is big, but superintelligences will not spare Earth a little sunlight (2024) →Claim page →
December 2024
· publication · dated to the month
“The AI Dilemma: Growth versus Existential Risk” (2024-12) credits Charles I. Jones as its author. Evidence ↗· publication · dated to the month
Jones, “The AI Dilemma: Growth versus Existential Risk” (2024-12) articulates E03. Published version of Jones's working paper; the abstract attributes existential risk from a misaligned superintelligence to 'many experts'. Evidence ↗· publication · dated to the day
The preprint "The Reality of AI and Biorisk" (arXiv v1, 2 December 2024) states that the available literature does not support the notion that access to biological information and planning through current, publicly available large language models can significantly increase biorisk. Evidence ↗· publication · dated to the day
The preprint "The Reality of AI and Biorisk" (arXiv v1, 2 December 2024) states that existing studies of AI-related biorisk are nascent, often speculative, or limited in methodological maturity and transparency. Evidence ↗· publication · dated to the day
DeepSeek published its V3 training report. Evidence ↗· publication · dated to the day
Wenfeng Liang is listed among the V3 report’s contributors. Evidence ↗