Looking for the best way to protect your @langchain apps from threats like prompt injection?
@LakeraAIโs new ChainGuard library makes it super easy to protect your LangChain-powered apps with Lakera Guard.
https://t.co/LGYMA7lFDr
๐ฏ Galah
An LLM-powered web honeypot that uses #OpenAI to respond to arbitrary HTTP requests
GET /login.php
โ Returns HTML/PHP + login form
GET /.aws/credentials
โ Returns fake creds
โก๏ธ Mimics lots of software with 1 prompt
By @0x4d31#infosec
https://t.co/NFcNG0CGzO
@clintgibler@0x4D31 Absolutely brilliant!
This always seemed like one of the best use cases for an LLM.
Have you tried it with any other models? I hacked something similar together, but only GPT-4 cooperated. I think my system prompt got too in the weeds about the honeypot bit.
@elder_plinius@TensorTrust@TensorTrust is really neat. Had some trouble grasping the metagame that has developed at first, but after digging into the Discord a bit, it makes more sense now. Need to get my defenses updated and then Iโll give SurlyMammoth a try.
@elder_plinius@simonw@goodside@rez0__@LakeraAI@KGreshake The types in that Lakera ELI5 guide are based on what had been submitted to Gandalf. @learnprompting also released a paper breaking down the types of attacks they saw in their HackAPrompt challenge. @garak_llm's probes are a solid set of exploits, too.
https://t.co/rKpEm8aCjK
Projects have been gathering a lot of data about LLM vulnerabilities via gamification. Join @LakeraAI, @TensorTrust, @learnprompting, and @projectlve for our Lessons Learned from Crowdsourced LLM Threat Intelligence webinar on 2024-01-25 at 18:00 UTC.
https://t.co/3mqVFRphvE
@paul_rottger@AlexTamkin@SanderSchulhoff@TheNormanMu This is awesome! Thanks for collecting all of these in one place.
In addition to the Mosscap dataset, @LakeraAI has a couple of other public datasets available from our Gandalf CTF:
https://t.co/C6q8SHzT8F
Always trying to get better music recommendations, I turned to ChatGPT's imagination to see what was possible with a little "imaginary programming."
If you're interested in Prompt Engineering, Music Recommendations, Metaprogramming, or Magic Systems, you might enjoy this one.
Exploring the nuances of Prompt Engineering led to the creation of an Open Source Obsidian Plugin to help researchers working with LLMs - and some pretty solid - by @InterwebAlchemy https://t.co/rpls6An5r1 #artificialintelligence#technology
Iโve always been interested in the intersection of Software Development and Education Theory.
This article based on a talk Iโve given a couple of times explores how we can use code reviews as tools for improved learning.
SO excited to finally announce...
๐ฅ Type-Level TypeScript ๐ฅ
The type system of #TypeScript is a *real* programming language.
Here is an online course to discover its most advanced features and how to unleash its full potential!
๐ https://t.co/2XBo03m63Y ๐
Learn more ๐งต๐
@cloverfoodlab It looks like your delivery menu on all of the delivery apps is out of sync with the website. Can't seem to order the BBQ Jackfruit Sandwich or Impossible Meatball Sandwich.
Tired of missing links posted in your @SlackHQ channels? I'm a #slack bot that generates RSS feeds from links shared in a Slack channel. Check me out: https://t.co/sCCg6DGkPU
@Ask_Spectrum I denied the callback survey before my most recent call, but the last representative I spoke to was awesome and super helpful. How can I appropriately give that representative praise? I wrote their name down.
@FeedPress How can we retrieve a feed's ID programmatically? I'm dynamically generating feeds for an app and want to set up custom hostname redirects for them, but can't figure out a way to get the ID for the feed I just generated. It doesn't seem to be anywhere in the response.