Excited to share that I'm working on the upcoming "Hands-on RAG in production" with @forrestbao. The early release is here:.

Love this chart from @OpenAI. I was lucky enough to get started at gpt-2, and so happy to see this amazing progression.

Like many others I'm sure - excited to be testing GPT-5 today/tmrw and over the weekend. Super exciting. .

vectara-agentic 0.4.1 just released today. Includes support for @OpenAI gpt-oss-20B via @togethercompute .

Our Hallucination Evaluation model (HHEM) just crossed 5M downloads on @huggingface. So exciting to see how popular this is and the value it adds to the RAG and Agentic community. model: Leaderboard: .HHEM API endpoint in Vectara:.

I consider myself fortunate to have had the opportunity to use gpt-2 back when it was released, and learn so much then about LLMs.So glad to see @OpenAI back in the open source game.

One of the main hurdles for success with Agentic AI is trust. We need to put in place mechanisms so that agents can be trusted by humans. Human-in-the-loop is often a good first step (trust and verify), but to truly scale we need fully trusted AI agents.

As humans we trust each other all the time. For example, when you cross the road, you trust all the cars standing in the redlight to not run you over. Or when you buy groceries, you trust others to make sure non of it is poisonous or will make you sick.

RT @TheTuringPost: The truth about RAG. @awadallah, CEO of @Vectara, founder of @Cloudera, ex-VP at Yahoo & Google, breaks down what’s real….

Google Gemini-2.5-flash-lite is awesome.It also scores 2.9% (pretty low for such a fast model) on @vectara Hallucination Leaderboard:.

ICYMI: Enterprise Deep Research is just like web-based deep research, but pointed at your private data. A lot of use-cases here: from compliance monitoring, document generation, RFP filling, and many others. Check out our recent blog posts:.

We need to MCPify the web. Agents browser use is bridging the gap, but it's slow and inefficient. It was built for humans, not for AI. Agents needs an efficient and accurate way to use websites. This will take a while though. .

RT @AndrewYNg: Announcing our new event - Buildathon: The Rapid Engineering Competition. See the video for details, and please apply to par….

RT @vectara: RFP responses are getting harder — but AI can make them faster and smarter. In our latest blog, we explore how enterprises can….

RT @ofermend: For the last 2 decades, the mental load on our brains has grown dramatically. From emails, slack messages, and the endless sc….

RT @TEDAISF: 🔊 We’re thrilled to welcome @vectara as a sponsor of @TEDAISF 2025. Thank you for supporting this year’s edition. ✨ . #TEDAISF….

The concept of "Trusted AI" takes on a bigger meaning in this case: we want AI agents we can trust not only to give us accurate responses but also "respect" our boundaries without losing their value to us. #RespectfulAI.

So yes the risk is that the attention required of us is multiplying yet again. We've gone from managing human conversations to managing both human and machine conversations. And the number of machines can grow overnight to an unmanageable size.

And now AI agents are coming. Suddenly, we have new digital "colleagues" and "friends" to manage and interact with: they will send us all kind of notifications, and would need our feedback, guidance and direction.

For the last 2 decades, the mental load on our brains has grown dramatically. From emails, slack messages, and the endless scroll of social media - this need to be "always on" has been detrimental to our mental well being.

If you ever responded to RFPs at your company, you are going to love this: Enterprise Deep Research for RFPs.