Research Library
Landmark research across intelligence science
Original, educational summaries of the most influential papers in artificial intelligence, AGI, AI safety, cognitive neuroscience, neurotechnology, neurodivergence, and human intelligence. Each summary covers objective, methodology, findings, strengths, limitations, and practical implications.
Categories
Artificial Intelligence
Architectures, training paradigms, and capabilities of modern machine-learning systems.
3 papers
Artificial General Intelligence
Theoretical foundations and empirical progress toward systems with broad cognitive competence.
3 papers
AI Safety & Alignment
Specification, robustness, interpretability, and governance of advanced AI systems.
3 papers
Cognitive Neuroscience
How the brain encodes perception, memory, prediction, and decision-making.
3 papers
Neurotechnology
Brain-computer interfaces, neural decoding, and devices that read or modulate the nervous system.
3 papers
Neurodivergence
Cognitive variation, atypical processing styles, and the neurobiology of difference.
3 papers
Human Intelligence
Psychometrics, learning, expertise, and the cognitive architecture of human thought.
3 papers
All papers
Attention Is All You Need
2017 · NeurIPSVaswani et al.
Introduced the Transformer, replacing recurrence with self-attention and reshaping modern AI.
Deep Residual Learning for Image Recognition
2015 · CVPRHe, Zhang, Ren, Sun
Residual connections enabled training of very deep networks, winning ImageNet 2015.
Mastering the Game of Go with Deep Neural Networks and Tree Search
2016 · NatureSilver et al.
AlphaGo combined deep networks and Monte Carlo tree search to defeat world-class Go players.
Scaling Laws for Neural Language Models
2020 · OpenAI Technical ReportKaplan et al.
Showed that LLM performance follows smooth, predictable power-law relationships with compute, data, and parameters.
Training Compute-Optimal Large Language Models
2022 · DeepMindHoffmann et al.
Demonstrated that for a fixed compute budget, model size and training tokens should scale roughly equally.
Emergent Abilities of Large Language Models
2022 · TMLRWei et al.
Argued that certain capabilities appear abruptly above a scale threshold rather than improving smoothly.
Concrete Problems in AI Safety
2016 · arXivAmodei, Olah, Steinhardt, Christiano, Schulman, Mané
Foundational taxonomy of practical safety problems in modern ML systems.
Training Language Models to Follow Instructions with Human Feedback
2022 · OpenAI / NeurIPSOuyang et al.
Introduced InstructGPT and the now-standard RLHF pipeline for aligning LLMs with human intent.
Constitutional AI: Harmlessness from AI Feedback
2022 · AnthropicBai et al.
Trained a helpful, harmless assistant using AI-generated critiques guided by a written constitution.
The Free-Energy Principle: A Unified Brain Theory?
2010 · Nature Reviews NeuroscienceKarl Friston
Proposed that the brain minimizes a single quantity - variational free energy - across perception, action, and learning.
The Hippocampus as a Cognitive Map
1978 · Oxford University PressJohn O'Keefe, Lynn Nadel
Established the hippocampus as a spatial-mapping system, foundational to memory research.
A Default Mode of Brain Function
2001 · PNASRaichle et al.
Identified the default mode network - brain regions most active during rest and internal thought.
High-Performance Brain-to-Text Communication via Handwriting
2021 · NatureWillett et al.
An intracortical BCI decoded imagined handwriting into text at 90 characters per minute.
Speech Synthesis from Neural Decoding of Spoken Sentences
2019 · NatureAnumanchipalli, Chartier, Chang
Synthesized intelligible speech directly from cortical activity in the vocal motor area.
Millisecond-Timescale, Genetically Targeted Optical Control of Neural Activity
2005 · Nature NeuroscienceBoyden, Zhang, Bamberg, Nagel, Deisseroth
Founding paper of optogenetics - using light to control genetically targeted neurons with millisecond precision.
Weak Central Coherence and Cognition in Autism
1994 · CognitionUta Frith, Francesca Happé
Proposed that autistic cognition favors local detail processing over global integration.
Default-Mode Network Activity in ADHD
2007 · Neuroscience & Biobehavioral ReviewsSonuga-Barke, Castellanos
Proposed that ADHD attention lapses arise from failure to suppress default-mode network activity.
The Phonological Deficit Hypothesis of Dyslexia
1995 · Journal of Research in ReadingSnowling, Stanovich
Established phonological processing impairments as the core cognitive feature of developmental dyslexia.
'General Intelligence,' Objectively Determined and Measured
1904 · American Journal of PsychologyCharles Spearman
Proposed the existence of a general intelligence factor (g) underlying performance across cognitive tasks.
Thinking, Fast and Slow (Dual-Process Theory)
2011 · Farrar, Straus and GirouxDaniel Kahneman
Synthesized decades of research into a dual-process model: fast intuitive System 1 and slow deliberative System 2.
Theory of Fluid and Crystallized Intelligence
1963 · Journal of Educational PsychologyRaymond Cattell
Distinguished fluid intelligence (novel reasoning) from crystallized intelligence (accumulated knowledge).