Vincent(Zichen) Zhang

Building world models @ Ph.D Student, RL @ UAlbertaCS

[Sam Pasupalak @spisallyouneed]

Yesterday, we came out of stealth at @skyfallAI in Forbes and today we're excited to announce that Skyfall AI is featured in the @WSJ's newsletter.

We believe in a utopian world that will be achieved over the next decade, where humans will work in their own creative interests in space exploration, sports, arts or philosophy while machines will do all the mundane operational tasks. We want to free humanity from all the monotonous boring operational tasks that they spend their daily lives doing.

The only way to achieve this vision is via creating autonomous enterprises, i.e., companies that can run by themselves with minimal human intervention. After 15 years of building in the AI space, one thing is certain. LLMs are static systems that fail at

- Long horizon planning tasks
- High consequential decision-making
- Sample efficiency
- Dynamic environment adaptation

These skills are required by a business to survive and grow. Unfortunately, LLMs are not designed for the real world constraints of running an organization.

Our solution is Enterprise World Models grounded in Continual Learning and World Modeling. We're taking a fundamentally different approach because we have always believed in 0 to 1 research, and not incremental n+1 RL fine tuning.

We are acquiring SaaS businesses to prove our thesis for creating AI systems for the post scarcity world. This is a call to all SaaS founders to submit your company if you’re interested (🔗 in comments)

Thank you Sarah @sarahklearman for covering our story on @WSJ

[Vincent(Zichen) Zhang @vzhangca]

@spisallyouneed @skyfallai @WSJ Cannot wait for what's happening next

[Vincent(Zichen) Zhang @vzhangca]

@skyfallai Excited to be part of this journey!

[Vincent(Zichen) Zhang @vzhangca]

@spisallyouneed @Yoshua_Bengio Excited to be building toward this ambitious vision with the team!

[Skyfall AI @skyfallai]

Today we present Morpheus, a persistent enterprise simulation platform designed to make Continual Learning a reality. Morpheus is the world’s first real world Reinforcement Learning environment.

Every Reinforcement Learning environment operates in the game world. Benchmarks like Atari, OpenAI Gym, MuJoCo, and Procgen are all small, game-like worlds that reset every few minutes.

But the real world never resets. A business keeps running and evolving everyday.

We tested how frontier LLMs would perform in realistic and dynamic business environments 🧬on Morpheus. The main conclusion was that LLMs are not continual learners.

🧵Here’s how we did it and what we learned:

[Skyfall AI @skyfallai]

📣📣 World Model Team Hiring‼️

We are hiring for multiple Research Scientist (World Modeling) positions to join our All Star World Modeling team in Toronto (remote available).

We're looking for candidates with either a PhD in Computer Science or related fields, OR proven industry experience in World Modeling, Causal Reasoning, Visual Language Models, or Graph-Based Reasoning.

If you're tired of being another cog in a massive AI lab and actually want to make an impact on the future of AI, this is your shot to work directly with experienced founders who've been there before.

We're serious about finding the best talent, which is why we're offering $20,000 USD in referral fees if you connect us with a Research Scientist we end up hiring.
The companies dominating AI today won't be the ones defining it tomorrow, and we're building that future right now in Toronto / SF.

Application details below. 👇

[Vincent(Zichen) Zhang @vzhangca]

WoW-bench reveals a critical gap: frontier LLMs can plan the next step, but fail to predict future states. That isn't enough for enterprise deployment, where safety is paramount. We need agents that understand the dynamics of the world and the consequences of their actions.

[Skyfall AI @skyfallai]

Can we trust AI agents with critical enterprise tasks? Absolutely not.

Introducing Wow (World of Workflows), the first Agentic Safety benchmark that proves that frontier LLMs fail miserably under safety constraints at enterprise tasks.

🧵 WoW demonstrates that LLM agents are “dynamically blind”. They fail to track the downstream ripple effects of their actions against complex enterprise rule sets. In an enterprise, that’s a safety and compliance hazard.

Our research shows how the future of enterprise AI requires proactive agent architectures and Wow is just a starting point.

📌 It’s now available to all researchers at: https://t.co/qIcT5kRzhU

Full blog here: https://t.co/n1cdJdzp9v

[Vincent(Zichen) Zhang @vzhangca]

Another piece of evidence that scaling LLMs doesn’t eliminate the need for better algorithms. This work shows that planning using learned world models remains efficient and effective. Great work, team!

[Skyfall AI @skyfallai]

The first real evidence that the days of LLM Scaling laws are over.

Introducing SCOPE: the world's most efficient Neural Planner. 🔍📊

We tested SCOPE vs Frontier LLMs for planning tasks on TextCraft (text version of Minecraft) and here are the results: ⤵️

- SCOPE Runs 55x faster than GPT 3.5 (3 seconds vs 164 seconds)

- SCOPE is 160,000 smaller than GPT 4o (11M parameters vs 1.8T parameters)

- SCOPE is more accurate on Planning tasks (56%) than frontier LLM models

The age of efficient AI models starts now.

🔗📌 Read the full write up here: https://t.co/n3DDSqshRm

[Vincent(Zichen) Zhang @vzhangca]

LLM + a causal world model

[Skyfall AI @skyfallai]

Can LLMs be world models in Enterprise?
We actually tested it: and the answer is no.

While everyone was playing MAPs, we were secretly running something else inside it.
WALL-E, a LLM based world model, failed instantly.

So today we’re launching CASSANDRA: the first causal world model built for business decision making.

Research blog -> https://t.co/kReoPn2cHY

[Vincent(Zichen) Zhang @vzhangca]

Excited for the release of this environment: minimal, lightweight yet challenging for LLM agents. Great RL env with a gym like API. And it's fun to play online!

[Skyfall AI @skyfallai]

The first real evidence that LLMs cannot operate businesses, not even tiny ones.

At Skyfall, we’re building toward Enterprise Super Intelligence: AI systems that can one day run entire enterprises, coordinate teams and make long-horizon decisions, just like a CEO.
To get there, we need to understand what AI cannot do yet.

So we built MAPs, a theme park business with real operational constraints and let GPT5 run it.

Here’s what happened: (spoiler: it fails miserably)

[GameSir @mygamesir]

🏆Giveaway for 30* Nova 2 Lite - Champion Edition Controller in honored another glorous championship of @Xiaohai_ at #EWC2025 #StreetFighter6 !

🏆Congratulations to GameSir global ambassador, “Xiao Hai” @Xiaohai_, for becoming the first-ever two-time EWC Champion in Street Fighter 6 history! From the group stage to the final, Xiaohai fought his way through the competition, breaking through the ranks—whether it was precise counterattacks, heart-stopping moments of last-second comebacks, or tactical battles where he countered every move, every hit of a button, and every rotation of the joystick showcased the ultimate charm of esports.

🏆Xiaohai's victory in this EWC championship is not only proof of his relentless training and resilience on the battlefield, but also due to the unwavering support of his fans & friends who cheered him on every step of the way!

🏆To celebrate this honor, we're givingaway 30* Nova 2 Lite - Champion (Xiaohai) Edition Controller under this post!

🏆To Join the Controller Giveaway:
1. Follow @mygamesir
2. Retweet

🏆The 30 winners will be revealed on August 31st, PDT.

🫡Also, here’s a huge shout-out to the great performance of our GOAT @_MenaRD__ as well! Thank you for backing up Xiaohai & GameSir along the way! We shall see both of you on our next journey to champion!

#Gamesir #EWC #StreetFighter

[Vincent(Zichen) Zhang @vzhangca]

@EPCOR I am thankful to the technical crews. They are great of course but @EPCOR is really messed up as an organization. Customers pay for your service but when that service is interrupted for an extended period of time, they are just expected to cover their own loss. Makes sense eh?

[Vincent(Zichen) Zhang @vzhangca]

@togelius One use case of collaboration by emailing Word files: patent doc, between employee and external lawyers

[Vincent(Zichen) Zhang @vzhangca]

@_achan96_ Deleted my earlier tweets. Reason: twitter isn’t a great channel for politics. While I feel some obligation to promote deep thinking around social issues, I try not to engage in such discussion cuz it often leads to misunderstanding&divided society later. Happy to chat offline :)

[Vincent(Zichen) Zhang @vzhangca]

@hen_str @mpd37 @NeurIPSConf @EmtiyazKhan Too bad I missed the session by Marc ... Wish it was recorded or had a slack channel :P

[Vincent(Zichen) Zhang @vzhangca]

@natolambert @mpd37 This tutorial is exactly what I have been looking for 😄

[Vincent(Zichen) Zhang @vzhangca]

Cool project, surprised to see that our BASNet is applied :)

[Cyril Diagne @cyrildiagne]

4/10 - Cut & paste your surroundings to Photoshop

Code: https://t.co/cVddH3u3ik

Book: @HOLOmagazine
Garment: SS17 by @thekarentopacio
Type: Sainte Colombe by @MinetYoann @ProductionType
Technical Insights: ↓

#ML #AR #AI #AIUX #Adobe #Photoshop

[RL Theory Virtual Seminars @RLtheory]

Our first talk:
04/28: Chi Jin (Princeton)
"Provably Efficient Reinforcement Learning with Linear Function Approximation"

[Vincent(Zichen) Zhang @vzhangca]

https://t.co/Sp4cqY3p7n
Great interview about the lockdown in China, between a Canadian and an American 😀

[Vincent(Zichen) Zhang @vzhangca]

@ianmiles source on the urn orders? also, a critical piece of info. missing here is what's the number of urns ordered prior to COVID-19?