Fei Zhao

Postdoc at UAB | PhD in CS
Multimodal AI · VLMs · AI Agents
Hallucination mitigation & RL post-training
14 first-author papers · Building reliable AI

[Fei Zhao @FZhao15]

@wasd_network @EricDJuly AI agents write the code. Humans approve it without reading it, probably because of incompetent hires who lack the technical skills to understand the code in the first place. Eventually, nobody understands the system anymore. Then something breaks, and nobody knows how to fix it.

[Fei Zhao @FZhao15]

@msorrentino91 @EricDJuly AI agents write the code. Humans approve it without reading it. Eventually, nobody understands the system anymore. Then something breaks, and nobody knows how to fix it. 😂

[Fei Zhao @FZhao15]

@RealSharpenIron @EricDJuly Microsoft has access to so many powerful AI models, yet it still takes 24 hours to debug Outlook? 🤔 And they say AI is insanely powerful.

[Fei Zhao @FZhao15]

@MefreeNTR2098 @EricDJuly I couldn't agree more

[Fei Zhao @FZhao15]

There's so much to learn from Tom Brady.

His discipline, resilience, and relentless pursuit of excellence inspire me.

Greatness isn't built overnight. It's built through consistency and never giving up.

Video source: https://t.co/P2RWAwNxic

[Fei Zhao @FZhao15]

@Mind_Essentials This is why I love research and engineering. Building something new every day keeps me going.

[Fei Zhao @FZhao15]

@screwgauage That’s what counts. Small wins add up. Hang in there!

[Fei Zhao @FZhao15]

@miglbi Such a beautiful view!

[Fei Zhao @FZhao15]

@TheShiftJournal That's my GOAT.

[Fei Zhao @FZhao15]

@NavalQuotes247 @naval This is exactly what I’m working toward every day. Keep getting better. Lock in. Laser focus.

[Fei Zhao @FZhao15]

@Wordofwise_ Thank you for sharing this! It's exactly why I keep training every day. Like Jalen Brunson said, never be afraid to fail. If you put in the work on your own in the offseason, you won't fear the big moment. And even if you fail, you'll learn from it and come back better.

[Fei Zhao @FZhao15]

@NavalQuotes247 @naval Exactly. Stop renting out your time. Master your craft, build your leverage, and let your mind do the compounding.

[Fei Zhao @FZhao15]

@imfat 100%! When you believe in yourself unconditionally, the whole world starts to shift. Time to level up! 🚀🔥

[Fei Zhao @FZhao15]

@NavalismHQ @naval So true! We write our own rules and make our own path. Let's make every single chapter count! 🔥✨

[Fei Zhao @FZhao15]

@Mericamemed This is awesome!

[Fei Zhao @FZhao15]

We hope this survey helps organize the rapidly growing literature on multimodal AI safety and supports future research on trustworthy MM-LLM systems.

Feedback and discussions are very welcome!😀

[Fei Zhao @FZhao15]

Excited to share our survey accepted at the ICLR 2026 Trustworthy AI Workshop.

📄Evolving Safety Landscape of Multimodal Large Language Models

We propose a taxonomy of MM-LLM threats and discuss emerging safeguards.

arXiv soon
https://t.co/AUFs97VSlL

Happy to connect @ ICLR!

[Fei Zhao @FZhao15]

As LLMs increasingly integrate multiple modalities (vision, language, audio, etc.), the safety landscape changes significantly.

Cross-modal alignment and fusion introduce new risks that are not well captured by traditional uni-modal threat models.

[Fei Zhao @FZhao15]

In this survey we:

• propose a taxonomy of safety threats in MM-LLMs
• analyze how threat models evolve in multimodal systems
• review emerging safeguards and defenses
• highlight open research challenges

[Fei Zhao @FZhao15]

@AlexGDimakis Great opportunity!