1 hr 4 min
AI-powered Search for the Enterprise — with Deedy Das of Glean Latent Space: The AI Engineer Podcast — Practitioners talking LLMs, CodeGen, Agents, Multimodality, AI UX, GPU Infra and al
-
- Technology
The most recent YCombinator W23 batch graduated 59 companies building with Generative AI for everything from sales, support, engineering, data, and more:
Many of these B2B startups will be seeking to establish an AI foothold in the enterprise. As they look to recent success, they will find Glean, started in 2019 by a group of ex-Googlers to finally solve AI-enabled enterprise search. In 2022 Sequoia led their Series C at a $1b valuation and Glean have just refreshed their website touting new logos across Databricks, Canva, Confluent, Duolingo, Samsara, and more in the Fortune 50 and announcing Enterprise-ready AI features including AI answers, Expert detection, and In-context recommendations.
We talked to Deedy Das, Founding Engineer at Glean and a former Tech Lead on Google Search, on why he thinks many of these startups are solutions looking for problems, and how Glean’s holistic approach to enterprise probllem solving has brought so much success.
Deedy is also just a fascinating commentator on AI current events, being both extremely qualified and great at distilling insights, so we also went over his many viral tweets diving into Google’s competitive threats, AI Startup investing, and his exposure of Indian University Exam Fraud!
Show Notes
* Deedy on LinkedIn and Twitter and Personal Site
* Glean
* Glean and Google Moma
* Golinks.io
* Deedy on Google vs ChatGPT
* Deedy on Google Ad Revenue
* Deedy on How much does it cost to train a state-of-the-art foundational LLM?
* Deedy on Google LaMDA cost
* Deedy’s Indian Exam Fraud Story
* Lightning Round
* Favorite Products: (covered in segment)
* Favorite AI People: AI Pub
* Predictions: Models will get faster for the same quality
* Request for Products: Hybrid Email Autoresponder
* Parting Takeaway: Read the research!
Timestamps
* [00:00:21] Introducing Deedy
* [00:02:27] Introducing Glean
* [00:05:41] From Syntactic to Semantic Search
* [00:09:39] Why Employee Portals
* [00:12:01] The Requirements of Good Enterprise Search
* [00:15:26] Glean Chat?
* [00:15:53] Google vs ChatGPT
* [00:19:47] Search Issues: Freshness
* [00:20:49] Search Issues: Ad Revenue
* [00:23:17] Search Issues: Latency
* [00:24:42] Search Issues: Accuracy
* [00:26:24] Search Issues: Tool Use
* [00:28:52] Other AI Search takes: Perplexity and Neeva
* [00:30:05] Why Document QA will Struggle
* [00:33:18] Investing in AI Startups
* [00:35:21] Actually Interesting Ideas in AI
* [00:38:13] Harry Potter IRL
* [00:39:23] AI Infra Cost Math
* [00:43:04] Open Source LLMs
* [00:46:45] Other Modalities
* [00:48:09] Exam Fraud and Generated Text Detection
* [00:58:01] Lightning Round
Transcript
[00:00:00] Hey everyone. Welcome to the Latent Space Podcast. This is Alessio, partner and CTO and residence at Decibel Partners. I'm joined by my, cohost swyx, writer and editor of
[00:00:19] Latent Space. Yeah. Awesome.
[00:00:21] Introducing Deedy
[00:00:21] And today we have a special guest. It's Deedy Das from Glean. Uh, do you go by Deedy or Debarghya? I go by Deedy. Okay.
[00:00:30] Uh, it's, it's a little bit easier for the rest of us to, uh, to, to spell out. And so what we typically do is I'll introduce you based on your LinkedIn profile, and then you can fill in what's not on your LinkedIn. So, uh, you graduated your bachelor's and masters in CS from Cornell. Then you worked at Facebook and then Google on search, specifically search, uh, and also leading a sports team focusing on cricket.
[00:00:50] That's something that we, we can dive into. Um, and then you moved over to Glean, which is now a search unicorn in building intelligent search for the workplace. What's not on your LinkedIn that people should know about you? Firstly,
[00:01:01] guys, it's a pleasure. Pleasure to be here. Thank you so much for having me.
[00:01:04] What's not on my LinkedIn is probably everything that's non-professional. I think the biggest ones are I'm a huge movie buff and I love reading, so I think I get through, usually I li
The most recent YCombinator W23 batch graduated 59 companies building with Generative AI for everything from sales, support, engineering, data, and more:
Many of these B2B startups will be seeking to establish an AI foothold in the enterprise. As they look to recent success, they will find Glean, started in 2019 by a group of ex-Googlers to finally solve AI-enabled enterprise search. In 2022 Sequoia led their Series C at a $1b valuation and Glean have just refreshed their website touting new logos across Databricks, Canva, Confluent, Duolingo, Samsara, and more in the Fortune 50 and announcing Enterprise-ready AI features including AI answers, Expert detection, and In-context recommendations.
We talked to Deedy Das, Founding Engineer at Glean and a former Tech Lead on Google Search, on why he thinks many of these startups are solutions looking for problems, and how Glean’s holistic approach to enterprise probllem solving has brought so much success.
Deedy is also just a fascinating commentator on AI current events, being both extremely qualified and great at distilling insights, so we also went over his many viral tweets diving into Google’s competitive threats, AI Startup investing, and his exposure of Indian University Exam Fraud!
Show Notes
* Deedy on LinkedIn and Twitter and Personal Site
* Glean
* Glean and Google Moma
* Golinks.io
* Deedy on Google vs ChatGPT
* Deedy on Google Ad Revenue
* Deedy on How much does it cost to train a state-of-the-art foundational LLM?
* Deedy on Google LaMDA cost
* Deedy’s Indian Exam Fraud Story
* Lightning Round
* Favorite Products: (covered in segment)
* Favorite AI People: AI Pub
* Predictions: Models will get faster for the same quality
* Request for Products: Hybrid Email Autoresponder
* Parting Takeaway: Read the research!
Timestamps
* [00:00:21] Introducing Deedy
* [00:02:27] Introducing Glean
* [00:05:41] From Syntactic to Semantic Search
* [00:09:39] Why Employee Portals
* [00:12:01] The Requirements of Good Enterprise Search
* [00:15:26] Glean Chat?
* [00:15:53] Google vs ChatGPT
* [00:19:47] Search Issues: Freshness
* [00:20:49] Search Issues: Ad Revenue
* [00:23:17] Search Issues: Latency
* [00:24:42] Search Issues: Accuracy
* [00:26:24] Search Issues: Tool Use
* [00:28:52] Other AI Search takes: Perplexity and Neeva
* [00:30:05] Why Document QA will Struggle
* [00:33:18] Investing in AI Startups
* [00:35:21] Actually Interesting Ideas in AI
* [00:38:13] Harry Potter IRL
* [00:39:23] AI Infra Cost Math
* [00:43:04] Open Source LLMs
* [00:46:45] Other Modalities
* [00:48:09] Exam Fraud and Generated Text Detection
* [00:58:01] Lightning Round
Transcript
[00:00:00] Hey everyone. Welcome to the Latent Space Podcast. This is Alessio, partner and CTO and residence at Decibel Partners. I'm joined by my, cohost swyx, writer and editor of
[00:00:19] Latent Space. Yeah. Awesome.
[00:00:21] Introducing Deedy
[00:00:21] And today we have a special guest. It's Deedy Das from Glean. Uh, do you go by Deedy or Debarghya? I go by Deedy. Okay.
[00:00:30] Uh, it's, it's a little bit easier for the rest of us to, uh, to, to spell out. And so what we typically do is I'll introduce you based on your LinkedIn profile, and then you can fill in what's not on your LinkedIn. So, uh, you graduated your bachelor's and masters in CS from Cornell. Then you worked at Facebook and then Google on search, specifically search, uh, and also leading a sports team focusing on cricket.
[00:00:50] That's something that we, we can dive into. Um, and then you moved over to Glean, which is now a search unicorn in building intelligent search for the workplace. What's not on your LinkedIn that people should know about you? Firstly,
[00:01:01] guys, it's a pleasure. Pleasure to be here. Thank you so much for having me.
[00:01:04] What's not on my LinkedIn is probably everything that's non-professional. I think the biggest ones are I'm a huge movie buff and I love reading, so I think I get through, usually I li
1 hr 4 min