1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
|
# Product
<!-- impeccable:product-schema 1 -->
## Platform
web
## Stack
Static HTML, CSS and JS in this repo. The Makefile gzips every source file. The `test` branch deploys to test.waterlooagi.com. three.js from cdnjs is allowed.
## Users
University of Waterloo students who are technical (CS, SE, EE, math, anyone who ships) and curious about AGI research. Secondary readers are researchers at frontier labs who follow a link from a writeup or a member's profile and want to see who these students are and what they work on.
## Product Purpose
Waterloo AGI is a small group of UWaterloo students doing research on AGI and frontier AI, started 25 Sep 2026. The site shows what the group does, its research, and its members. Success is a lab researcher or a strong student coming away knowing the members by name and what each one is working on.
## Positioning
It is the only active student group at Waterloo focused on AGI research. It runs independently from WUSA. Members publish openly (LessWrong writeups, X threads, GitHub code) with the goal of doing work good enough that frontier labs notice them.
## Operating Context
Members run experiments, build and run evals, read papers weekly and discuss them, and follow up with their own research. Each member owns a project for the term. The proposed rhythm (a plan, label it as one) is weekly reading, weekly research check-ins, and a public writeup when a project has a result.
## Capabilities and Constraints
- No apply form or join call to action on the site for now.
- No hierarchy. Members have no roles or titles on the site, the founder included.
- WP-001 "Can an LLM dream?" is finished (October 2026). It lives at /experiments/llm-dream/ with an animated figure of the real setup and final recall from Table 2 of the paper, an in-page PDF.js reader and a PDF download. The paper (papers/WP-001-can-an-llm-dream.pdf) is the source for every number. The home page shows it as one teaser row right after the hero.
- WP-003 "Can a model explain its moves?" (Ahmed Osman, in progress) lives at /experiments/explain-moves/ with an animated chess analysis board. The position and explanation on it are an illustration of the setup, not a result. WP-002 (Eric Zou, relation vocabulary for agent memory, in progress) lives at /experiments/relation-vocab/ with an animated retrieval walk over a small memory graph (illustrative chat, no results).
- Member data lives in one editable data file (name, photo, facts about achievements, what they are building, research focus, hobbies).
## Brand Commitments
- Name: Waterloo AGI. Title text must be readable and short.
- Voice: plain declarative sentences, subject first. No em or en dashes, taglines, fragment pairs, mirrored sentences, triples, punchlines, drama adverbs. No "breathing dot" style AI design tells.
## Evidence on Hand
- Members: Eric Zou (Electrical Engineering), Brenda Li (Mathematics) and Ahmed Osman (Mathematics, from EE) have full profiles. No sample people.
- Socials: Instagram @waterlooagi is live. No club X account yet (Eric will sign one up), no club GitHub. Contact email is eric@waterlooagi.com.
- WP-001 results are the only results that exist. No speakers, partners or stats exist. Never invent them.
## Product Principles
- The members are the point. Every member is equal in the layout.
- Show real work, label plans as plans, and state small honest numbers plainly.
- The ML visuals must be recognizable to someone who works in ML.
|
