Blog
Essays and field notes.
Standalone articles on the history of technology — the people, the moments, and the machines that shaped the screen you're reading this on. Written when there's something to say, not on a schedule.
- Essays
- 35
- Featured
- 0
- Topics
- 125
- Latest · 2026
- 2026
2026
Nick Bostrom: Superintelligence & the FHI
Nick Bostrom's Superintelligence (2014) and the Future of Humanity Institute — how long-termist AI safety shaped OpenAI and Anthropic's founding.
PostPaul Christiano: RLHF & AI Alignment Research
Paul Christiano — the Alignment Research Center founder who turned RLHF into the dominant alignment paradigm and shaped AI policy from inside.
PostChinchilla, March 2022: The Compute-Optimal Reset
Chinchilla — DeepMind's March 2022 compute-optimal scaling paper that reversed Kaplan's recipe (more data, smaller models) and reset training strategy.
PostInstructGPT, January 2022: RLHF Becomes Production AI
Ouyang et al.'s January 2022 InstructGPT — the RLHF method that made ChatGPT possible by training models via human feedback.
PostGPT-3, June 2020: The Pivot Point of the LLM Era
Brown et al.'s May 2020 GPT-3 paper — the 175B-parameter model that demonstrated in-context learning and made "foundation models."
PostKaplan Scaling Laws, January 2020: 'Just Add Compute'
Kaplan et al.'s January 2020 OpenAI paper — language-model loss scales as a power law with compute, data, and parameters, defining the 2020s AI race.
PostThe GPT-2 'Too Dangerous to Release' Decision, February 2019
OpenAI's February 2019 staged release of GPT-2 — the first "AI safety theatre" of the LLM era, when the 1.5B model was withheld.
PostBERT, October 2018: The Most-Cited NLP Model for Two Years
Google's October 2018 BERT paper — bidirectional encoding that set SOTA on 11 NLP tasks simultaneously and was integrated into Google Search within a year.
PostGPT-1, June 2018: The Start of the GPT Line
OpenAI's June 2018 GPT-1 paper — proving unsupervised pre-training plus supervised fine-tuning transforms NLP. The start of the GPT line.
PostAI and Surveillance Capitalism: Zuboff's Critique
Shoshana Zuboff, Clearview AI, facial recognition, predictive policing — how surveillance capitalism shaped AI and what resistance looks like.
PostAI Nationalism & Geopolitics: The Chip Cold War
The US–China–EU triangulation, the UK and Japan national AI strategies, India's AI mission, and the rise of sovereign AI infrastructure.
PostAI's Energy Bill: Data Centres, Water, and Carbon
Data-centre water use, training-run carbon footprints, the Microsoft–Three Mile Island restart — why AI energy use is now a national infrastructure issue.
PostThe AI Chip Wars: Compute as the New Oil
TSMC, ASML, NVIDIA, US export controls, and the CHIPS Act — how the semiconductor supply chain became the geopolitical battlefield of the AI age.
PostPrompt Engineering and the New Art of Talking to Machines
Chain-of-thought, few-shot, ReAct, structured prompts — the rise and partial fall of prompt engineering as a profession.
PostHallucinations: Why AI Makes Things Up & Lies
Why large language models fabricate facts — what causes hallucinations, what RAG and attribution can fix, and what the industry still can't solve in 2024.
PostSora, February 2024: OpenAI Text-to-Video Preview
Sora — OpenAI's February 2024 video-generation preview, the deliberate non-release that signaled AI video was the next multimodal frontier.
PostThe EU AI Act Rollout: Risk Tiers & €35M Fines
The EU AI Act — March 2024 passage, August 2024 entry into force, phased rollout through 2026-2028, with risk tiers and €35M fines.
PostThe Bletchley Declaration and What Came After
The November 2023 Bletchley Declaration — the first international AI safety agreement, signed by 28 nations and the EU, and what came after at Seoul.
PostAnthropic Founding, 2021: The OpenAI Schism
How Dario and Daniela Amodei left OpenAI over disagreements about safety and commercialisation, founded Anthropic, and built the leading safety-focused AI lab.
PostThe OpenAI Board Crisis, November 17–22, 2023
The OpenAI board crisis — five days in November 2023 that fired and reinstated Sam Altman, the most dramatic governance event in AI-industry history.
PostThe White House Voluntary Commitments, July 21, 2023
The July 2023 White House voluntary commitments — 7 AI companies (Amazon, Anthropic, Google, Inflection, Meta, Microsoft, OpenAI) pledged safety.
PostAltman Senate Testimony, May 16, 2023: AI Regulation
Sam Altman's May 2023 Senate testimony — the first major AI CEO to endorse federal licensing for AI, comparing it to nuclear regulation.
PostFLI Pause Letter, March 2023: 30,000 Signatories
The FLI's March 2023 open letter calling for a 6-month pause on AI more powerful than GPT-4 — signed by 30,000+ including Elon Musk.
PostThe CAIS Statement on AI Risk, May 30, 2023
The CAIS May 2023 statement — "mitigating extinction risk from AI should be a global priority" — signed by Hinton, Bengio, Altman, and Hassabis.
PostThe AI Election, Revisited: Deepfakes in 2024
The 2024 AI election — the New Hampshire Biden robocall, deepfakes in Slovakia, India, Indonesia, and the EU, and post-election analyses.
PostSynthetic Media, Deepfakes, and the Crisis of Evidence
Deepfakes, voice cloning, and the 'liar's dividend' — how synthetic media collapsed photographic evidence and launched a detection-versus-generation arms race.
PostAI and Copyright: Who Owns the Training Data?
NYT v. OpenAI, Getty v. Stability AI, Authors Guild suits — the copyright battles that decide who owns AI training data and whether scraping is fair use.
PostAuthors Guild v. Google, October 2015: Fair Use
Authors Guild v. Google — the October 2015 Second Circuit ruling that established fair use for book scanning and now underpins every LLM training defence.
PostNYT v. OpenAI & Microsoft: The Fair-Use Trial
NYT v. OpenAI — the December 2023 lawsuit that will likely define whether training LLMs on copyrighted articles is fair use or infringement.
PostGetty Images v. Stability AI: The Copyright Test
Getty v. Stability AI — the January 2023 lawsuit over Stable Diffusion's training on 12 million Getty photos and the 2025 UK ruling.
PostThe Colorado State Fair AI-Art Win, August 2022
The Colorado State Fair — August 2022, when Jason Allen's "Théâtre D'opéra Spatial" became the first AI-generated image to win a fine-art competition.
PostOpen vs Closed: The Defining Fault Line of Modern AI
Hugging Face, LLaMA, Mistral, DeepSeek — the open-weights movement and why open vs closed is the defining AI industry fault line of the 2020s.
PostStable Diffusion Release, August 2022: Open AI Art
Stable Diffusion — Stability AI's August 2022 open-weights release that democratised image generation, triggered artist lawsuits, and seeded the open-AI debate.
PostLiang Wenfeng: DeepSeek and the Open-Weight Earthquake
Liang Wenfeng — the hedge-fund founder behind DeepSeek, whose January 2025 R1 release triggered a US stock rout and reset the AI race.
PostDeepSeek-R1 and the Reset of the AI Race, January 2025
DeepSeek-R1 — the January 2025 release that matched OpenAI's o1 at a fraction of the cost, triggered a $589B NVIDIA stock rout, and reset the AI race.