Research desk
Signal, not spectacle.
Frontier labs, journals, and deep essays — eight pillars, primary sources linked out.
Superintelligence
View pillar →Systems that outpace human cognition across domains — and what that does to power, labor, and meaning.
Economic futures under AI automation
OpenAI’s public research agenda on how transformative AI reshapes labor, policy, and preparedness — not a takeoff prophecy, an instrumentation problem.
DeepMind on AGI safety and security
Google DeepMind’s framing of AGI-level risk: misuse, misalignment, and the institutional stack required before capability runs ahead of control.
Situational Awareness — Leopold Aschenbrenner
A widely circulated essay on short AI timelines, cluster geopolitics, and what ‘superintelligence’ looks like if scaling laws keep paying rent.
Consciousness
View pillar →What it is like to be — and whether machines can ever share that interior.
Being You — Anil Seth on the controlled hallucination
Seth’s predictive-processing account of consciousness: the self as a model the brain maintains, not a ghost in the machine.
Integrated Information Theory — Tononi & collaborators
IIT’s formal bet: consciousness tracks irreducible cause–effect structure (Φ). Adored and attacked in equal measure.
Global workspace — Dehaene’s conscious access
Stanislas Dehaene’s cognitive neuroscience of when information becomes globally available — ignition, report, and the signatures of awareness.
Dreams
View pillar →Nocturnal simulation, memory replay, and the brain’s private generative models.
Dreams as offline model fitting
Contemporary neuroscience frames REM and NREM dreaming as active inference / memory consolidation — the brain training on generated worlds.
Lucid dreaming as a consciousness switch
Lucidity research shows metacognition can re-enter the dream state — a rare window into volitional control of an endogenous VR.
Sleep replay and the overnight rewrite
Sharp-wave ripples and systems consolidation: how the sleeping brain reorganizes the day’s traces into durable structure.
Artificial Alignment & Safety
View pillar →Making powerful systems do what we intend — under distribution shift, deception, and scale.
Constitutional AI — Anthropic
Training assistants with a written constitution and AI critique loops — a public blueprint for scalable oversight without pure RLHF dependence.
Superalignment — OpenAI’s framing
OpenAI’s (evolving) public stance on aligning systems smarter than their operators — recursive oversight, interpretability, and empirical bets.
Frontier Safety Framework — DeepMind
Risk thresholds, capability evals, and escalation policies for frontier models — safety as an operational protocol, not a slogan.
Many-shot jailbreaking — Anthropic
A concrete failure mode: long-context prompting that overrides refusals. Alignment that works in short chats can fail at scale.
BCIs & Neural Interfaces
View pillar →Reading and writing cortex at bandwidth — implants, decoding, and neural rights.
High-performance speech neuroprostheses
Peer-reviewed leaps in decoding intended speech from cortex — word rates approaching conversation, with hard translational caveats.
Neuralink’s public clinical updates
Company demos and PR are not papers — but they shape the BCI narrative. Track claims against independent clinical reporting.
Synchron Stentrode — endovascular BCI path
A less invasive route to neural interfaces via blood vessels — different risk profile, different bandwidth ceiling.
Humanoid Robotics
View pillar →Bodies that walk our world — embodiment as the missing half of intelligence.
Figure — humanoids meeting foundation models
Figure’s public demos fuse bipedal hardware with multimodal AI — the bet that general-purpose robots need general-purpose models.
Boston Dynamics Atlas — athletic embodiment
Atlas remains the kinetic proof that dynamic balance and whole-body control can look eerily alive — control theory as spectacle and science.
Tesla Optimus — mass-manufacture bet
Optimus pitches humanoids as a scalable product line. The interesting claim is industrial: can autonomy ride the same learning stack as cars?
BioTech & Neurotech
View pillar →Organoids, gene tools, neuromodulation — biology as programmable substrate.
Brain organoids and the sentience debate
Lab-grown neural tissue that exhibits complex activity — and an ethics literature asking when a dish deserves moral consideration.
Optogenetics — writing into circuits with light
The method that gave systems neuroscience a high-speed write channel — and the template for every later neuromodulation modality.
Allen Institute — cell types and brain maps
Public atlases of cell types, connectivity, and multimodal brain data — infrastructure for the next decade of neurotech.
Theoretical Philosophies
View pillar →Mind, identity, value, and the long argument about what we owe future minds.
The hard problem of consciousness — David Chalmers
Why explaining function may leave experience unexplained — the philosophical wedge still driving lab and AI debates.
Superintelligence — Nick Bostrom
The book that mainstreamed control problems, orthogality, and decisive strategic advantage for a generation of SI thinkers.
The extended mind — Clark & Chalmers
If a notebook can be part of cognition, what happens when the notebook is a cortex implant or a cloud model?
AGI ruin arguments — Eliezer Yudkowsky
A maximalist alignment pessimism that shaped the rationalist SI community — sharp, contested, impossible to ignore historically.
Full index
26 entriesEconomic futures under AI automation
OpenAI’s public research agenda on how transformative AI reshapes labor, policy, and preparedness — not a takeoff prophecy, an instrumentation problem.
DeepMind on AGI safety and security
Google DeepMind’s framing of AGI-level risk: misuse, misalignment, and the institutional stack required before capability runs ahead of control.
Situational Awareness — Leopold Aschenbrenner
A widely circulated essay on short AI timelines, cluster geopolitics, and what ‘superintelligence’ looks like if scaling laws keep paying rent.
Being You — Anil Seth on the controlled hallucination
Seth’s predictive-processing account of consciousness: the self as a model the brain maintains, not a ghost in the machine.
Integrated Information Theory — Tononi & collaborators
IIT’s formal bet: consciousness tracks irreducible cause–effect structure (Φ). Adored and attacked in equal measure.
Global workspace — Dehaene’s conscious access
Stanislas Dehaene’s cognitive neuroscience of when information becomes globally available — ignition, report, and the signatures of awareness.
Dreams as offline model fitting
Contemporary neuroscience frames REM and NREM dreaming as active inference / memory consolidation — the brain training on generated worlds.
Lucid dreaming as a consciousness switch
Lucidity research shows metacognition can re-enter the dream state — a rare window into volitional control of an endogenous VR.
Sleep replay and the overnight rewrite
Sharp-wave ripples and systems consolidation: how the sleeping brain reorganizes the day’s traces into durable structure.
Constitutional AI — Anthropic
Training assistants with a written constitution and AI critique loops — a public blueprint for scalable oversight without pure RLHF dependence.
Superalignment — OpenAI’s framing
OpenAI’s (evolving) public stance on aligning systems smarter than their operators — recursive oversight, interpretability, and empirical bets.
Frontier Safety Framework — DeepMind
Risk thresholds, capability evals, and escalation policies for frontier models — safety as an operational protocol, not a slogan.
Many-shot jailbreaking — Anthropic
A concrete failure mode: long-context prompting that overrides refusals. Alignment that works in short chats can fail at scale.
High-performance speech neuroprostheses
Peer-reviewed leaps in decoding intended speech from cortex — word rates approaching conversation, with hard translational caveats.
Neuralink’s public clinical updates
Company demos and PR are not papers — but they shape the BCI narrative. Track claims against independent clinical reporting.
Synchron Stentrode — endovascular BCI path
A less invasive route to neural interfaces via blood vessels — different risk profile, different bandwidth ceiling.
Figure — humanoids meeting foundation models
Figure’s public demos fuse bipedal hardware with multimodal AI — the bet that general-purpose robots need general-purpose models.
Boston Dynamics Atlas — athletic embodiment
Atlas remains the kinetic proof that dynamic balance and whole-body control can look eerily alive — control theory as spectacle and science.
Tesla Optimus — mass-manufacture bet
Optimus pitches humanoids as a scalable product line. The interesting claim is industrial: can autonomy ride the same learning stack as cars?
Brain organoids and the sentience debate
Lab-grown neural tissue that exhibits complex activity — and an ethics literature asking when a dish deserves moral consideration.
Optogenetics — writing into circuits with light
The method that gave systems neuroscience a high-speed write channel — and the template for every later neuromodulation modality.
Allen Institute — cell types and brain maps
Public atlases of cell types, connectivity, and multimodal brain data — infrastructure for the next decade of neurotech.
The hard problem of consciousness — David Chalmers
Why explaining function may leave experience unexplained — the philosophical wedge still driving lab and AI debates.
Superintelligence — Nick Bostrom
The book that mainstreamed control problems, orthogality, and decisive strategic advantage for a generation of SI thinkers.
The extended mind — Clark & Chalmers
If a notebook can be part of cognition, what happens when the notebook is a cortex implant or a cloud model?
AGI ruin arguments — Eliezer Yudkowsky
A maximalist alignment pessimism that shaped the rationalist SI community — sharp, contested, impossible to ignore historically.