|
@@ -4,10 +4,10 @@
|
|
| 4 |
|
| 5 |
title: blank # the website title (if blank, full name will be used instead)
|
| 6 |
first_name: Mark
|
| 7 |
-
middle_name:
|
| 8 |
last_name: Lovett
|
| 9 |
contact_note: >
|
| 10 |
-
Reach out via email
|
| 11 |
description: > # the ">" symbol means to ignore newlines until "footer_text:"
|
| 12 |
A simple, whitespace theme for academics. Based on [*folio](https://github.com/bogoli/-folio)
design.
|
| 13 |
footer_text: >
|
|
@@ -19,7 +19,7 @@ lang: en # the language of your site (for example: en, fr, cn, ru, etc.)
|
|
| 19 |
icon: ⚛️ # the emoji used as the favicon (alternatively, provide image name in /assets/img/)
|
| 20 |
|
| 21 |
url: https://MarksLovett.github.io # the base hostname & protocol for your site
|
| 22 |
-
baseurl:
|
| 23 |
last_updated: false # set to true if you want to display last updated in the footer
|
| 24 |
impressum_path: # set to path to include impressum link in the footer, use the same path as permalink in a page, helps to
conform with EU GDPR
|
| 25 |
back_to_top: true # set to false to disable the back to top button
|
|
@@ -73,7 +73,7 @@ og_image: # The site-wide (default for all links) Open Graph preview image
|
|
| 73 |
|
| 74 |
# For Google Analytics, see
https://support.google.com/analytics/answer/10447272?hl=en&ref_topic=14088998&sjid=5129943941510317771-SA#zippy=%2Cgoogle-sites
|
| 75 |
# and follow the instructions for Google Sites. You will need to create a Google Analytics property and copy the Google
tag ID.
|
| 76 |
-
google_analytics:
|
| 77 |
cronitor_analytics: # cronitor RUM analytics site ID (format: XXXXXXXXX)
|
| 78 |
pirsch_analytics: # your Pirsch analytics site ID (length 32 characters)
|
| 79 |
openpanel_analytics: # your Openpanel analytics client ID (format: XXXXXXXX-XXXX-XXXX-XXXX-XXXXXXXXXXXX)
|
|
|
|
| 4 |
|
| 5 |
title: blank # the website title (if blank, full name will be used instead)
|
| 6 |
first_name: Mark
|
| 7 |
+
middle_name:
|
| 8 |
last_name: Lovett
|
| 9 |
contact_note: >
|
| 10 |
+
Reach out via email
|
| 11 |
description: > # the ">" symbol means to ignore newlines until "footer_text:"
|
| 12 |
A simple, whitespace theme for academics. Based on [*folio](https://github.com/bogoli/-folio)
design.
|
| 13 |
footer_text: >
|
|
|
|
| 19 |
icon: ⚛️ # the emoji used as the favicon (alternatively, provide image name in /assets/img/)
|
| 20 |
|
| 21 |
url: https://MarksLovett.github.io # the base hostname & protocol for your site
|
| 22 |
+
baseurl: # the subpath of your site, e.g. /blog/. Leave blank for root
|
| 23 |
last_updated: false # set to true if you want to display last updated in the footer
|
| 24 |
impressum_path: # set to path to include impressum link in the footer, use the same path as permalink in a page, helps to
conform with EU GDPR
|
| 25 |
back_to_top: true # set to false to disable the back to top button
|
|
|
|
| 73 |
|
| 74 |
# For Google Analytics, see
https://support.google.com/analytics/answer/10447272?hl=en&ref_topic=14088998&sjid=5129943941510317771-SA#zippy=%2Cgoogle-sites
|
| 75 |
# and follow the instructions for Google Sites. You will need to create a Google Analytics property and copy the Google
tag ID.
|
| 76 |
+
google_analytics: "G-SS85HED103" # your Google Analytics measurement ID (format:
G-XXXXXXXXXX)
|
| 77 |
cronitor_analytics: # cronitor RUM analytics site ID (format: XXXXXXXXX)
|
| 78 |
pirsch_analytics: # your Pirsch analytics site ID (length 32 characters)
|
| 79 |
openpanel_analytics: # your Openpanel analytics client ID (format: XXXXXXXX-XXXX-XXXX-XXXX-XXXXXXXXXXXX)
|
|
@@ -28,15 +28,15 @@ latest_posts:
|
|
| 28 |
|
| 29 |
I am a Ph.D. candidate in applied mathematics at Dartmouth College, where I study game theory and artificial intelligence
(AI). In my game-theory research I model social interactions by reducing complex networks to concise, tractable
representations. These models help explain and predict phenomena across political science, AI, evolutionary ecology, and
business. My game-theory work focuses on several related themes:
|
| 30 |
|
| 31 |
-
- **Influence and allocation games.** Models of competition for limited resources in which players exert influence to
secure them. My thesis introduces the [
|
| 32 |
- **Evolutionary game theory.** The study of how strategies evolve in populations. By reducing complex population
dynamics to tractable mathematical models, evolutionary game theory yields accurate predictions of agent behavior and has
applications in ecology (for example, [Tilman et
al.](https://www.nature.com/articles/s41467-020-14531-6)), cooperation studies (for example, [Traulsen
et al.](https://royalsocietypublishing.org/doi/full/10.1098/rstb.2021.0508)), decision
making, and AI–human cooperation.
|
| 33 |
|
| 34 |
In AI, I study the incentive structures and learning environments that shape agent behavior: essentially, which
"games" drive learning and why. I investigate how formally specified objectives influence learning dynamics and
emergent strategies in both simulated and real-world settings, with most of my work in large language models and
multi-agent reinforcement learning. The reason to characterize these games is safety rather than performance. Failure
modes such as collusion, deception, and reward hacking are equilibrium properties rather than accidents of a particular
training run, so an incentive-level account tells us which failures a system is being pushed toward before we have to
catch them in evaluation.
|
| 35 |
|
| 36 |
This runs in two directions. On the reward-design side, I work with Elizaveta Tennant through the [MARS
program](https://caish.org/mars) on moral alignment for LLM agents, extending her framework of explicitly
specified intrinsic rewards ([paper](https://arxiv.org/abs/2410.01639)). On the oversight side, I work
with [Soroush Vosoughi](https://www.cs.dartmouth.edu/~soroush/) on ensemble and router dynamics for
language models, using the influencer's game as a lens on scalable oversight. <!-- TODO: link to the ensemble
project once public -->
|
| 37 |
|
| 38 |
-
Combining insights from AI and game theory has major applications for AI–AI, human–human, and especially human–AI
interactions. I think this matters urgently. As systems become more capable and more autonomous, misalignment becomes
harder to catch after the fact, and game theory is unusually well suited to the problem because it reasons about behavior
through incentives. That lets us construct the equilibria we want rather than only measure the dispositions a trained
model happens to arrive at. Two of my favorite cases for AI safety and game theory: Redwood Research on [the case for
ensuring that powerful AIs are
controlled](https://blog.redwoodresearch.org/p/the-case-for-ensuring-that-powerful) and the
Cooperative AI Foundation's [multi-agent risks report](https://arxiv.org/abs/2502.14143);..
|
| 39 |
|
| 40 |
Beyond research, I have thoroughly enjoyed teaching at Dartmouth. I design scalable, technology-enabled course frameworks
and curricula that improve delivery and student experience. My teaching emphasizes clear learning objectives, reproducible
assignments, and practical tools that make courses easier to run and extend. I have five years of TA experience across 12
courses at two institutions, and I have independently taught two courses on evolutionary game theory.
|
| 41 |
|
| 42 |
-
Outside of academics, I am a senior member of [Amifore Consulting and Management
LLC](https://www.amifore.com/), a startup delivering ethical consulting solutions to emerging businesses. I
work in AI and innovation consulting for clients internationally, including in Switzerland, implementing strategic project
management systems and building AI-assisted workflows. This work has sharpened my sense of how AI research translates into
practice, and of what responsible deployment actually asks of a small organization.
|
|
|
|
| 28 |
|
| 29 |
I am a Ph.D. candidate in applied mathematics at Dartmouth College, where I study game theory and artificial intelligence
(AI). In my game-theory research I model social interactions by reducing complex networks to concise, tractable
representations. These models help explain and predict phenomena across political science, AI, evolutionary ecology, and
business. My game-theory work focuses on several related themes:
|
| 30 |
|
| 31 |
+
- **Influence and allocation games.** Models of competition for limited resources in which players exert influence to
secure them. My thesis introduces the [_influencer's
game_](https://markslovett.github.io/InflGame/), in which each player's influence is
centered on its position in space, so position itself becomes a strategy. Examples include political campaigns (candidates
competing for votes) and platform competition for users; similar dynamics arise in machine-learning settings where agents
compete for attention or data.
|
| 32 |
- **Evolutionary game theory.** The study of how strategies evolve in populations. By reducing complex population
dynamics to tractable mathematical models, evolutionary game theory yields accurate predictions of agent behavior and has
applications in ecology (for example, [Tilman et
al.](https://www.nature.com/articles/s41467-020-14531-6)), cooperation studies (for example, [Traulsen
et al.](https://royalsocietypublishing.org/doi/full/10.1098/rstb.2021.0508)), decision
making, and AI–human cooperation.
|
| 33 |
|
| 34 |
In AI, I study the incentive structures and learning environments that shape agent behavior: essentially, which
"games" drive learning and why. I investigate how formally specified objectives influence learning dynamics and
emergent strategies in both simulated and real-world settings, with most of my work in large language models and
multi-agent reinforcement learning. The reason to characterize these games is safety rather than performance. Failure
modes such as collusion, deception, and reward hacking are equilibrium properties rather than accidents of a particular
training run, so an incentive-level account tells us which failures a system is being pushed toward before we have to
catch them in evaluation.
|
| 35 |
|
| 36 |
This runs in two directions. On the reward-design side, I work with Elizaveta Tennant through the [MARS
program](https://caish.org/mars) on moral alignment for LLM agents, extending her framework of explicitly
specified intrinsic rewards ([paper](https://arxiv.org/abs/2410.01639)). On the oversight side, I work
with [Soroush Vosoughi](https://www.cs.dartmouth.edu/~soroush/) on ensemble and router dynamics for
language models, using the influencer's game as a lens on scalable oversight. <!-- TODO: link to the ensemble
project once public -->
|
| 37 |
|
| 38 |
+
Combining insights from AI and game theory has major applications for AI–AI, human–human, and especially human–AI
interactions. I think this matters urgently. As systems become more capable and more autonomous, misalignment becomes
harder to catch after the fact, and game theory is unusually well suited to the problem because it reasons about behavior
through incentives. That lets us construct the equilibria we want rather than only measure the dispositions a trained
model happens to arrive at. Two of my favorite cases for AI safety and game theory: Redwood Research on [the case for
ensuring that powerful AIs are
controlled](https://blog.redwoodresearch.org/p/the-case-for-ensuring-that-powerful) and the
Cooperative AI Foundation's [multi-agent risks report](https://arxiv.org/abs/2502.14143);..
|
| 39 |
|
| 40 |
Beyond research, I have thoroughly enjoyed teaching at Dartmouth. I design scalable, technology-enabled course frameworks
and curricula that improve delivery and student experience. My teaching emphasizes clear learning objectives, reproducible
assignments, and practical tools that make courses easier to run and extend. I have five years of TA experience across 12
courses at two institutions, and I have independently taught two courses on evolutionary game theory.
|
| 41 |
|
| 42 |
+
Outside of academics, I am a senior member of [Amifore Consulting and Management
LLC](https://www.amifore.com/), a startup delivering ethical consulting solutions to emerging businesses. I
work in AI and innovation consulting for clients internationally, including in Switzerland, implementing strategic project
management systems and building AI-assisted workflows. This work has sharpened my sense of how AI research translates into
practice, and of what responsible deployment actually asks of a small organization.
|
|
@@ -1 +1 @@
|
|
| 1 |
-
google-site-verification: google414fd2a08fff7a9d.html
|
|
|
|
| 1 |
+
google-site-verification: google414fd2a08fff7a9d.html
|
|
@@ -2,7 +2,7 @@
|
|
| 2 |
layout: page
|
| 3 |
title: projects
|
| 4 |
permalink: /projects/
|
| 5 |
-
description: My current project and upcoming work.
|
| 6 |
nav: true
|
| 7 |
nav_order: 3
|
| 8 |
display_categories: [work, fun]
|
|
|
|
| 2 |
layout: page
|
| 3 |
title: projects
|
| 4 |
permalink: /projects/
|
| 5 |
+
description: My current project and upcoming work.
|
| 6 |
nav: true
|
| 7 |
nav_order: 3
|
| 8 |
display_categories: [work, fun]
|
|
@@ -2,7 +2,7 @@
|
|
| 2 |
layout: page
|
| 3 |
permalink: /repositories/
|
| 4 |
title: repositories
|
| 5 |
-
description: My Github repositories available to the public.
|
| 6 |
nav: true
|
| 7 |
nav_order: 4
|
| 8 |
---
|
|
|
|
| 2 |
layout: page
|
| 3 |
permalink: /repositories/
|
| 4 |
title: repositories
|
| 5 |
+
description: My Github repositories available to the public.
|
| 6 |
nav: true
|
| 7 |
nav_order: 4
|
| 8 |
---
|
|
@@ -2,13 +2,12 @@
|
|
| 2 |
layout: page
|
| 3 |
permalink: /teaching/
|
| 4 |
title: teaching
|
| 5 |
-
description: Course materials from the classes that I have taught.
|
| 6 |
nav: true
|
| 7 |
nav_order: 6
|
| 8 |
horizontal: false
|
| 9 |
---
|
| 10 |
|
| 11 |
-
|
| 12 |
<!-- pages/projects.md -->
|
| 13 |
<div class="teaching">
|
| 14 |
{% if site.enable_project_categories and page.display_categories %}
|
|
|
|
| 2 |
layout: page
|
| 3 |
permalink: /teaching/
|
| 4 |
title: teaching
|
| 5 |
+
description: Course materials from the classes that I have taught.
|
| 6 |
nav: true
|
| 7 |
nav_order: 6
|
| 8 |
horizontal: false
|
| 9 |
---
|
| 10 |
|
|
|
|
| 11 |
<!-- pages/projects.md -->
|
| 12 |
<div class="teaching">
|
| 13 |
{% if site.enable_project_categories and page.display_categories %}
|
|
@@ -9,7 +9,6 @@ category: work
|
|
| 9 |
related_publications: true
|
| 10 |
---
|
| 11 |
|
| 12 |
-
|
| 13 |
Every project has a feature showcase page.
|
| 14 |
This package implements influencer's game environments and tools for MARL and adaptive-dynamics experiments, making
it easy to probe opinion and influence dynamics across custom networked settings.
|
| 15 |
|
|
@@ -79,4 +78,4 @@ Here's the code for the last row of images above:
|
|
| 79 |
</div>
|
| 80 |
```
|
| 81 |
|
| 82 |
-
{% endraw %}
|
|
|
|
| 9 |
related_publications: true
|
| 10 |
---
|
| 11 |
|
|
|
|
| 12 |
Every project has a feature showcase page.
|
| 13 |
This package implements influencer's game environments and tools for MARL and adaptive-dynamics experiments, making
it easy to probe opinion and influence dynamics across custom networked settings.
|
| 14 |
|
|
|
|
| 78 |
</div>
|
| 79 |
```
|
| 80 |
|
| 81 |
+
{% endraw %}
|