Jerry

PhD Student at UIUC | Prev Duke

[Arush Jain @Arushj01]

Why did we name our company @ModusAudit?

Audit is deeply process driven, but we believe the currently processes are archaic and flawed.

What makes Modus powerful is the combination of an AI-native process, 100+ human CPAs, and our agent Andi to drive outcomes in concert. Each leveraging the advantages of the full Modus Operandi playbook.

[Pengcheng Wang @PengchengWang19]

🚀 Wanna build your own customized agent with controllable workflow? Introducing AgentSPEX — a declarative DSL for building LLM agents.

- Customizable agentic workflow with GUI builder
- Reproducible State-of-the-Art in SWE-bench verified
- Controllable workflow with YAML specification + sandboxed VM + Lean4 verification

🌐 Demo: https://t.co/hZ7O8R4yKe
💻 Code: https://t.co/PniYZnOUqq
📄 Paper: https://t.co/qMGBhVG9Dj
💡 (One of the) Applications: https://t.co/KjmqqDJK6I
(1/n)

#LLM #AIAgents #AgenticAI #OpenSource #AIResearch #AgentHarness

[Rui Pan @rui4research]

🔥GUIDE: Automatic Machine Learning Advisor 🔥

⭐GUIDE-7B outperforms >600B general-purpose models in ML paper advising.
📊GUIDE-recommended papers have >90% probability of being accepted.
🌐Achieving human-level idea evaluation accuracy.
💡LLMs are good at significance and novelty evaluation, but fall short in quality assessment.

🔓Blog: https://t.co/5RXKprsO5N
🔓Paper: https://t.co/46HbeStylC
🔓Code (inference): https://t.co/W8WkcoJaSG

Amazing Collaborators: @Howard_Liu140 @howard05539786 @actuallyjerry @youjiaxuan

[Jerry @actuallyjerryh]

@YifeiZhou02 Cool work! Any reason you guys didn't end up using BFCL as benchmark?

[Jerry @actuallyjerryh]

@WenhuChen Can you provide the % overlap between the questions both methods get wrong versus correct? Thanks!

[Jerry @actuallyjerryh]

@mansourtarek_ kalshi binary options

[Jerry @actuallyjerryh]

@kshrivats Wow! Great work!

[Jerry @actuallyjerryh]

@nav_mtr Nice