Mapping superintelligence, AI safety, and advanced model systems through papers, simplified research guides, interactive models, and public technical artifacts.
Research console
Superintelligence discovery map
From singular-event narratives to slower, messier complex-system takeoff models.
Featured paper
Control problems, honesty, value alignment, reward models, and safety evaluation frames.
Core lane
Hardware, inference, kernels, and model craft constraints that determine what research can run.
Applied lane
Tools and models for explaining behavior inside increasingly opaque neural systems.
Research map
Featured path
The first paper is the front door: theory history, modern critiques, alternative models, and interactive diagrams.
Applied proof
6
OpenAI PRs
<16 MB
Model cap
Publications
This is the main page narrative: discovery, theory, safety, and research navigation. Technical competitions sit underneath as evidence, not as the whole identity.
A comprehensive analysis synthesizing 60 years of intelligence explosion theory, from I.J. Good through modern critiques and alternative models.
A working map of theoretical gaps, interdisciplinary links, and accessible research opportunities in superintelligence research.
Practical safety evaluation frameworks for systems approaching human-level capabilities across multiple domains.
Systematic evaluation of reasoning capabilities and emergent behaviors in current large language models.
Explore research topics
Theory
From singular-event narratives to slower, messier complex-system takeoff models.
Safety
Control problems, honesty, value alignment, reward models, and safety evaluation frames.
Systems
Hardware, inference, kernels, and model craft constraints that determine what research can run.
Visibility
Tools and models for explaining behavior inside increasingly opaque neural systems.
Pre-training, scaling laws, post-training, reinforcement learning, and the practical bottlenecks behind capability growth.
Distributed systems, accelerators, networking, and operational constraints behind serious model experiments.
Cross-modal learning and specialized systems for biology, scientific discovery, and frontier research tools.
Model craft case study
One applied research artifact: OpenAI's challenge to train the strongest language model inside a 16 MB artifact and a 10 minute 8xH100 window. It supports the broader research story without replacing it.
NewyorkDev
The GitHub trail, submission media, and visual mark sit inside the case study as proof of the applied research lane.
Artifact board
Parameter Golf v9
Public PRs
6
NewyorkDev submissions to openai/parameter-golf
Best preserved run
1.12336724
final_int6 exact BPB
Frontier PR mean
0.94175270
v13 CaseOps PPM
Artifact cap
<16 MB
Code plus compressed model
Submission media
Videos loop silently. Click a video to hear one pass at low volume; it pauses after that pass.
Research collaboration
For AI safety, superintelligence theory, model systems, or citation requests, send the specific paper, artifact, or topic you want to discuss. Concrete references make the exchange useful.