About
Clark Whitehead
Independent ML researcher based in San Francisco, CA.
My interests center on continual learning, reinforcement learning, and world models. Over time, I hope to grow this work into an independent AI research lab.
The future I want to help build
- Cures for cancer and all disease
- Responsible genetic medicine
- End aging; pursue immortality
- Interstellar exploration and discovery
- A stable, healthy climate
- Abundance for everyone
- A future free from suffering
Long-term aspirations. Today’s support funds foundational AI research, with no promised breakthroughs or timeline.
More about this vision
Eliminate disease
Contribute to cures for cancer and, ultimately, all disease.
Advance genetic medicine
Enable precise, responsible DNA editing to prevent and treat disease.
Overcome biological aging
Extend healthy life by solving aging, with immortality as the ultimate long-term aspiration.
Explore beyond our solar system
Make interstellar travel possible and advance the search for other civilizations in the universe.
Restore climate stability
Help stop and reverse global warming and support a healthy, resilient planet.
Create abundance for all
Make essential resources, health care and opportunity widely accessible, so the benefits of progress reach everyone.
Reduce and ultimately end suffering
Direct scientific and technological progress toward lasting well-being and a future free from suffering.
Education & experience
Education
- –
Bachelor of Science
University of Hawaiʻi at Mānoa
Computer Science and Machine Learning
Focus in Data Science
ML experience
- Current
Independent ML research
Clark Whitehead Research
Exploratory continual-learning research.
- –
Machine Learning Engineer
TheCRM Corporation
Work focus
Fine-tuning Llama 405B on RV trim data.
- –
Machine Learning Engineer
Opt-Intelligence
Work focus
Adapting and fine-tuning Llama 70B to estimate opt-in probability for an advertising network.
- –
Machine Learning Engineer
University of Hawaiʻi at Mānoa
Work focus
Building LLMs to extract metadata from PDFs.
Work you can explore
A third of forgetting is the arbitrary choice in the example
Rough draft of the paper. An exact solver makes every demonstration provably optimal, so only the arbitrary tie-break varies; that tie-break is a third of what the next task destroys.
Rough draft v0.1 · measured result · not peer reviewed · PDF, 10 pages
Read the paperThe Muscle Memory benchmark: does an AI that learns beat one that takes notes?
A written-up plan for the Muscle Memory benchmark: two copies of one AI, one with a notebook, one that trains, the same levels, until one falls behind.
Preregistered design, draft v0 · not run
Read the paperWhen more self-play makes an agent worse
An AI that trains by playing itself got better, then worse, even though its training numbers kept improving.
Working manuscript · preliminary, not peer reviewed · PDF, 5 pages
Read the paperTarget drift in deep Q-learning
An interactive MountainCar teaching project on the moving-target problem in deep Q-learning and the role of a slowly updated target network.
It uses constructed data for teaching clarity. It is separate from the new research direction and is not experimental evidence for it.
Open the interactive explainer