Скачать или смотреть Why Do LLMs Hallucinate? Understanding AI Failures | AIAI Boston

What causes Large Language Models (LLMs) to hallucinate, and why does it matter? In this insightful discussion at the Generative AI Summit in Boston, Jeff Chapman (Field CTO, Vectara) explains the fundamental reason behind LLM hallucinations, from extreme data compression to real-world failures—including a high-profile lawsuit involving Air Canada’s AI chatbot.

Learn why hallucination risk is a serious challenge for enterprises and how companies can mitigate these risks using Responsible AI and RAG (Retrieval-Augmented Generation).

Start building today!

** Introduction to Vectara **

Vectara's AI Assistant/Agent solution is enterprise-ready, focusing on accuracy, explainability of results, and secure access control.

** Try Vectara **

Start your Account at vectara.com. Or visit:

https://console.vectara.com/signup

*The Hughes Hallucination Evaluation Model (HHEM) has reached 2 Million downloads on Hugging Face!**
https://vectara.com/blog/celebrating-...

*Deep Dive into Mockingbird*
https://vectara.com/blog/mockingbird-...

** Hughes Hallucination Evaluation Model (HHEM) **

(3/26/2024 Release!) - Vectara now includes a Factual Consistency Score (FCS) to evaluate Hallucinations:

https://vectara.com/blog/automating-h...

Technical Deep Dive of the HHEM:
https://vectara.com/blog/cut-the-bull...

Measuring Hallucinations in RAG Systems (layperson HHEM intro):
https://vectara.com/blog/measuring-ha...

Get the HHEM Open Source code from GitHub
https://github.com/vectara/hallucinat...

Setting up Your Own HEM Leaderboard
https://huggingface.co/blog/leaderboa...

** Try HHEM **

Vectara has not only focused on retrieval but on mitigating hallucinations in LLMs, the leading risk factor preventing companies from adopting this monumental technology. As such, Vectara has developed an open-source model for evaluating hallucinations in LLMs, called the HHEM, and has also developed a complementary leaderboard that lists the top LLMs and ranks their proclivity for hallucinating. The Hughes Hallucination Evaluation Model (HHEM) is available as an open-source model on Huggingface:

https://huggingface.co/vectara/halluc...

See the HHEM Leaderboard:
https://huggingface.co/spaces/vectara...

*Vectara Use-Cases*
https://docs.vectara.com/docs/use-cas...

*Vectara x Incorta Nexus*
https://vectara.com/blog/power-precis...

Check out this article for more insight into integrating Vectara into LangChain!
https://blog.langchain.dev/langchain-...

** DEVELOPER Resources **

Join our Discord
  / discord  

Read our API documentation
https://docs.vectara.com/

Join the Vectara Developer Community
https://discuss.vectara.com/

** Social Channels **

  / vectara  
  / mycompany  
   / @vectara