Science and technology — C1

The vocabulary of science and technology has been the fastest-changing register in English over the past five years. The 2022 release of ChatGPT brought a wave of AI vocabulary into mainstream use; mRNA vaccines pulled molecular biology into kitchen-table conversation; and the climate vocabulary has continued to harden into a precise technical register. At C1 you should be able to read a Nature News feature, an Economist science section, or a Wired longread without footnotes — and write competently about the same topics.

This lesson clusters the vocabulary around: how scientific progress is described (paradigm shifts, breakthroughs, replication), the modern AI stack (LLMs, training, inference, alignment), the quantum-computing wave, biotech (CRISPR, mRNA, gene therapy), and 2026 climate science. Throughout, the focus is on register-aware terms: scientists talk about “results” and “evidence”; journalists talk about “breakthroughs” and “discoveries”. Knowing which to use, and when each is overblown, is the C1 marker.

How scientific progress is described

• paradigm — the foundational framework of a field (Kuhn)

• paradigm shift — a major reorganization of how a field thinks (overused; use with care)

• scientific revolution — the historical phrase for a paradigm shift

• normal science — Kuhn’s term for incremental work within a paradigm

• breakthrough — a major, sometimes hyped, advance

• major advance — calmer journalistic phrasing

• discovery — finding something new (often misused; many “discoveries” are observations)

• finding — a result from a study (more measured)

• result — a data point from an experiment

• proof of concept — early demonstration that something is possible

• breakthrough vs incremental progress — the contrast scientists draw against journalists’ hype

• the bleeding edge / the cutting edge / the leading edge — the frontier; bleeding edge implies higher risk

• state of the art / SOTA — the current best (very common in AI papers)

• emerging technology — early-stage, not yet mainstream

• mature technology — well-established

• replication — re-running a study to verify

• replication crisis — the problem that many published results don’t replicate

• reproducibility — broader concept (different researchers, different conditions)

• p-hacking — manipulating analysis to achieve statistical significance

• publication bias — the tendency to publish only positive results

• file-drawer problem — unpublished negative results sitting unpublished

A real-style sentence: The phrase “paradigm shift” appears in popular science writing far more often than Kuhn would have permitted; what most articles describe is not a paradigm shift but an empirical breakthrough within an existing paradigm — an important distinction that the philosophy-of-science community has tried, mostly in vain, to defend.

The AI vocabulary: from algorithms to AGI

The fastest-growing technical vocabulary of the 2020s. At C1 you need to be able to read both Wired features and AI research papers’ abstracts.

• AI (artificial intelligence) — the umbrella term

• machine learning (ML) — the dominant modern approach

• deep learning — ML using deep neural networks

• neural network — the layered computational architecture loosely inspired by the brain

• transformer — the 2017 architecture underlying modern LLMs

• attention / self-attention — the transformer mechanism

• LLM (large language model) — the technology behind ChatGPT, Claude, Gemini

• foundation model — a general-purpose model that can be adapted to many tasks

• frontier model — the largest and most capable models (used in the AI-safety literature)

• generative AI / GenAI — AI that produces text, images, audio, video

• discriminative AI — older AI that classifies rather than generates

• AGI (artificial general intelligence) — human-level AI across most tasks (definition contested)

• superintelligence — beyond-human AI

• ASI (artificial superintelligence) — same idea

• narrow AI — AI for specific tasks (most current AI)

• training — the process of teaching the model

• pretraining — the initial large-scale training

• fine-tuning — adapting a pretrained model for a specific task

• RLHF (reinforcement learning from human feedback) — the alignment technique behind ChatGPT

• inference — running the model to generate output

• prompt — the input to a generative model

• prompt engineering — crafting prompts for desired output

• system prompt — the model’s standing instructions

• context window — how much text the model can consider at once

• token — the unit a model processes (roughly a word fragment)

• embedding — a vector representation of meaning

• vector database — storage for embeddings

• RAG (retrieval-augmented generation) — augmenting a model with external knowledge retrieval

• agent / agentic AI — AI systems that take actions in the world, not just produce text

• chain of thought / CoT — making the model reason step by step

• multimodal — handling multiple input/output types (text + image + audio)

• hallucination — when a model generates false but plausible content

• alignment — making AI systems behave as intended

• alignment problem — the difficulty of ensuring AI does what we want

• AI safety — the field studying alignment and existential risk

• red-teaming — adversarial testing of AI systems

• jailbreak / jailbreaking — bypassing an AI’s safety measures

• guardrails — safety constraints on AI output

• bias — when models reflect biased training data

• fairness — formal notion of equitable treatment

• interpretability / mechanistic interpretability — understanding what’s happening inside models

• deepfake — AI-generated fake media (now a generic noun)

A real-style sentence: The release of GPT-4 in March 2023 catalyzed a rapid shift in how the AI alignment community talks about risk: discussions that had been dismissed as speculative — frontier models exhibiting deceptive behavior, agentic systems taking unauthorized actions — moved into the empirical literature, with red-teaming reports and interpretability papers from Anthropic, OpenAI, and DeepMind reshaping the policy conversation in Washington and Brussels.

The AI labs and the geography of AI

Knowing the institutional landscape is part of C1 fluency on this topic.

• OpenAI (San Francisco) — ChatGPT, GPT-4, GPT-5
• Anthropic (San Francisco) — Claude family
• Google DeepMind (London / Mountain View) — Gemini
• Meta AI / FAIR — Llama (open-weight family)
• xAI — Grok
• Mistral (Paris) — European frontier lab
• the Chinese labs — DeepSeek, Qwen (Alibaba), Baichuan, Zhipu
• AI Safety Institute (UK, US) — government-affiliated safety bodies
• open-weight vs closed-weight — whether model weights are public
• open-source AI — sometimes synonymous with open-weight, sometimes stricter
• frontier lab — the few labs training the largest models
• the compute frontier — the limit of available compute for training
• scaling laws — empirical relationships between compute, data, and performance
• the AI race — the geopolitical and commercial competition

Quantum computing

• quantum computing / QC — computing exploiting quantum mechanics
• qubit (quantum bit) — the quantum analog of a bit
• superposition — a qubit being in multiple states at once
• entanglement — correlation between qubits
• decoherence — the loss of quantum state through environmental interaction
• error correction / quantum error correction — managing decoherence with redundancy
• logical qubit — error-corrected qubit built from many physical qubits
• physical qubit — the raw hardware element
• quantum supremacy / quantum advantage — when a quantum machine outperforms classical for some task
• NISQ era (noisy intermediate-scale quantum) — the current near-term regime
• fault-tolerant quantum computing — the long-term goal
• post-quantum cryptography — encryption secure against quantum attacks
• Shor’s algorithm / Grover’s algorithm — the two foundational quantum algorithms
• Google Quantum AI, IBM Quantum, IonQ, Rigetti, PsiQuantum — the major US efforts

A real-style sentence: Google’s December 2024 Willow announcement, which demonstrated below-threshold error correction on a 105-qubit chip, is a meaningful milestone — not because the machine performed any commercially useful task, but because it crossed the threshold beyond which adding qubits should reduce, rather than amplify, logical errors.

Biotechnology: CRISPR, mRNA, and gene therapy

• biotech — the broad industry

• genomics — the study of genomes

• proteomics — the study of proteins

• DNA / RNA — the nucleic acids

• gene — a unit of heredity

• genome — the full genetic complement

• gene editing — modifying DNA at specific sites

• CRISPR (clustered regularly interspaced short palindromic repeats) — the gene-editing technology behind the 2020 Nobel

• CRISPR-Cas9 — the most-used CRISPR system

• base editing — newer, more precise editing without cutting both DNA strands

• prime editing — even newer, more precise still

• gene therapy — treating disease by inserting or modifying genes

• gene drive — a gene-editing tool to spread a modification through a population

• off-target effects — unintended edits elsewhere in the genome

• germline editing vs somatic editing — heritable vs non-heritable modifications

• mRNA (messenger RNA) — the molecule used in the COVID-19 vaccines

• mRNA vaccine — the Pfizer/BioNTech and Moderna platform

• spike protein — the SARS-CoV-2 protein targeted by the vaccines

• lipid nanoparticle — the delivery vehicle for mRNA

• adjuvant — substance that boosts immune response

• booster — additional vaccine dose

• immunity / herd immunity / breakthrough infection

• long COVID / post-acute sequelae of COVID-19 (PASC)

• monoclonal antibody — a lab-produced antibody (Keytruda, Humira)

• antibody-drug conjugate — antibody linked to a drug payload

• GLP-1 — the drug class behind Ozempic, Wegovy, Mounjaro

• biosimilar — a generic-equivalent of a biologic

• precision medicine / personalized medicine — treatment tailored to individual genetics

• theranostics — combined therapy and diagnostics

• synthetic biology / synbio — engineering biological systems

• xenotransplantation — transplanting organs across species

• stem cell therapy — using stem cells for treatment

Climate science vocabulary

• anthropogenic — caused by humans

• anthropogenic climate change — the technical term

• global warming — the temperature dimension specifically

• climate change — the broader phenomenon

• the climate crisis — the activist framing

• climate emergency — even more urgent framing

• greenhouse gas (GHG) — CO2, methane, nitrous oxide, fluorinated gases

• CO2 equivalent (CO2e) — common unit accounting for the different warming effects of different gases

• carbon dioxide / CO2 — the most-discussed GHG

• methane / CH4 — short-lived but potent GHG

• emissions — released GHGs

• scope 1 / 2 / 3 emissions — direct, indirect-energy, and value-chain emissions

• net zero — emissions balanced by removals

• carbon neutral — similar but vaguer

• decarbonization — eliminating CO2 emissions

• deep decarbonization — across the whole economy

• carbon offset — a credit for emissions reduced elsewhere (much-criticized)

• carbon market — the system for trading offsets

• carbon pricing — putting a price on CO2 emissions (tax or cap-and-trade)

• cap and trade — emissions trading scheme

• mitigation — reducing emissions

• adaptation — adjusting to climate impacts

• resilience — capacity to absorb and recover

• loss and damage — the climate-justice concept that vulnerable countries deserve compensation

• stranded assets — fossil fuel investments that become worthless

• renewable energy — solar, wind, hydro, geothermal

• clean energy — broader; includes nuclear in some definitions

• transition / energy transition — the shift away from fossil fuels

• grid / the electric grid — power transmission infrastructure

• base load — the minimum power demand

• intermittency — variability of solar and wind

• storage / grid-scale storage — batteries for the grid

• EV (electric vehicle) — battery-electric car

• range anxiety — fear of running out of charge

• heat pump — the central technology for heating decarbonization

• IPCC (Intergovernmental Panel on Climate Change) — the UN scientific body

• COP — Conference of the Parties (COP21 produced the Paris Agreement; COP28 was in 2023)

• the Paris Agreement — the 2015 global climate accord

• NDC (nationally determined contribution) — each country’s pledge under Paris

• 1.5 degrees / 2 degrees — the temperature goals (above pre-industrial)

• tipping point — climate threshold past which changes become self-reinforcing

• feedback loop — process that amplifies (positive feedback) or dampens (negative feedback) change

• anthropogenic vs natural — caused by humans vs caused naturally

• attribution science — the field of attributing extreme events to climate change

• extreme weather — events outside historical norms

• heat dome — stagnant high-pressure system producing extreme heat

• wildfire / wildland fire — uncontrolled fires (US standard term)

• megafire — a particularly large wildfire

• drought — prolonged water scarcity

• megadrought — multi-decadal drought (the current US Southwest one)

• flood / flooding — water excess

• storm surge — coastal flooding from storms

• geoengineering / climate engineering — deliberate intervention in the climate system

• solar geoengineering / SRM (solar radiation management) — reflecting sunlight (stratospheric aerosol injection, marine cloud brightening)

• carbon removal / CDR (carbon dioxide removal) — removing CO2 from the atmosphere

• direct air capture (DAC) — chemical extraction of CO2 from air

• BECCS (bioenergy with carbon capture and storage)

• CCS (carbon capture and storage) — capturing emissions at the source

AmE-specific science and tech vocabulary

Collocations and high-frequency phrases

• conduct / fund / publish research
• publish in / be published in Nature / Science / Cell
• make / report / replicate a finding
• prove / disprove / cast doubt on a hypothesis
• build / design / scale a system
• train / fine-tune / deploy a model
• run / execute / accelerate inference
• address / mitigate / solve the alignment problem
• edit / modify / silence / knock out a gene
• sequence / assemble / annotate a genome
• deploy / scale / phase out technology
• emit / capture / sequester / remove carbon
• hit / miss / exceed a target
• reach / breach / blow past 1.5 degrees
• the technology has matured / is still nascent / is at an inflection point

Common Russian-speaker mistakes

1. Scientist used too broadly. AmE scientist specifically means a researcher in the natural or social sciences. Russian учёный covers a broader range. A historian is a historian or scholar, not a scientist. Computer scientists are a specific kind; programmers and software engineers are not scientists in this sense.
2. Technology used too narrowly (calque of техника). Russian техника often means specific machines (appliances, equipment). English technology can mean (a) a field (information technology), (b) a specific innovation (mRNA technology), or (c) capability broadly. Saying I bought new technology for the kitchen sounds odd; I bought new appliances / new equipment is right.
3. Smart for intelligent in technical context. In AmE, smart has acquired a specific tech meaning (smart phone, smart speaker, smart contract) — connected, programmable. For умный applied to people, the right adjective is intelligent, bright, sharp, not smart-phone smart. He is a smart person is OK in casual speech; intelligent or brilliant is more formal.
4. Receive a discovery (calque of получить открытие). The English verb is make a discovery, publish a finding, report a result, or announce a breakthrough. We received a discovery is not English.
5. Experiments used loosely. In Russian эксперимент can mean any trial or attempt. In scientific AmE, experiment is a controlled procedure; for an informal trial, use trial, test, pilot, try out. We did an experiment with the new policy is loose; we ran a pilot or we tested the new policy is more precise.
6. Apparatus for device. Russian аппарат maps to several English words depending on context: device (a small electronic thing), machine (industrial), instrument (scientific), apparatus (a complex setup, often laboratory). I bought a new apparatus for measuring blood pressure is dated/odd in AmE; device or monitor is natural.
7. Innovation overused as a buzzword*. American business speech does overuse innovation, but at C1 academic register, prefer breakthrough, advance, finding, development, technique when describing a specific thing, and reserve innovation for the broader phenomenon. Our innovations include a new method… sounds corporate; We developed a new method… is more direct.

Summary

• Progress vocabulary: paradigm shift (sparingly), breakthrough, finding, replication, SOTA.
• The AI stack: LLM, transformer, attention, training/inference, RLHF, RAG, agentic AI, hallucination, alignment, red-teaming, guardrails.
• AI institutional map: OpenAI, Anthropic, Google DeepMind, Meta AI, Mistral; frontier labs; open-weight vs closed.
• Quantum: qubit, superposition, entanglement, decoherence, error correction, NISQ, post-quantum crypto.
• Biotech: CRISPR, base/prime editing, mRNA, gene therapy, GLP-1, precision medicine, synthetic biology.
• Climate: anthropogenic, GHG, CO2e, net zero, decarbonization, IPCC, COP, NDC, tipping point, geoengineering, CDR, DAC.
• Avoid: scientist too broadly, technology for appliances, received a discovery, experiment for trial, apparatus for device.

Next theme: Environment and sustainability — 2026 — anthropogenic change, biodiversity, ecosystem services, ESG, the circular economy, climate refugees, geoengineering, the IPCC framework.