Hacker News new | past | comments | ask | show | jobs | submit | wintonzheng's comments login

Curious: what use cases do you use to test the spacial reasoning ability of these models?

Autonomous vehicles went through the same phases. The reliability part of autonomous agent has to become really really reliable first. The iterations in software is much faster than hardware though

we're a pretty small team and don't have a plan for it in the near future :(

I would love to know the reason you're interested in react native though if you don't mind sharing! pls email me or suchintan at shu@skyvern.com / suchintan@skyvern.com or join our discord


https://github.com/Skyvern-AI/skyvern/issues/76 we're planning to introduce a llm router in a week and you should be able to call your local llama after that.

We're prioritizing on cloude 3, as its performance seems to be good. That said, please join our discord and bring more thoughts/requests to us. code contribution is also more than welcome


I'm Shu, also cofounder of skyvern. first of all, you are more than welcome to join our coummunity. one big reason of open sourcing skyvern is to serve the individuals. This project was inspired by problems we learnt from tlking to corporates but it doesn't have to always serve those use case. problems like boring form filling are pretty common in real life.

Second, llms definitely can help bridge the gap. My 58y old mom who grew up in the rural area of China doesn't know much about internet and doesn't know how to order takeout on her phone. She only knows the basic usage of wechat, the whatsapp in China and text messages. I've been a coder for 10+ years and I still find it so darn hard to keep up with tools and information out there. I do hope skyvern becomes what you're saying and help people get access to more in the world.


Shu, I thank you for communicating your personal ambitions for Skyvern, and the touching personal anecdote. Making computers easier to use for my aging parents is also one of my goals.

I will be reaching out to the project with an analysis of its security model against prompt injection attacks.

I'll also be taking on a project for a KeepassXC plug-in that automates the process of rotating password in online services, integrating Skyvern as the underlying system. At that time I'll need support and understanding Skyvern's gaps against the projects requirements, and I'll ask for mentorship on helping fill those gaps.

This use case I believe has potential to help both individual and corporate users achieve best practice and policy driven password management - currently a very difficult thing because of the propensity of users do not reset their passwords on time, which is ultimately caused by the difficulty and variety in password resetting mechanisms. Our plug-in will aim to solve that problem for KeepassXC users.

I believe this work could result in a valuable contribution to Skyvern's security model, since an llm driven password reset workflow is uniquely vulnerable to attack or to attacker controlled texts. This provides a great benchmark for Skyvern's overall security model, and a great point to explore both classical and llm-based mitigation techniques.

How can best reach you and the team when the initial letter is ready? Once that's ready, I would like to have a video chat and plan a collaboration. I expect something to hatch in the next 2 weeks. I think you will enjoy our approach to the problem!


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: