Sep 6, 2023 · 49m · big-technology
Google’s Weird Year + Neeva Goes to Snowflake — With Sridhar Ramaswamy
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
In this episode of the Big Technology Podcast, former Google executive and Snowflake SVP Sridhar Ramaswamy examines Google's turbulent AI reset, the structural disruption of consumer search, and the rapid deployment of generative AI across enterprise data platforms.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Alex holds 19.4% of the talking time here. How this is scored →
Alex as informed peer 5.5 Guest teaching 6.3 Guest disagreement 2.1 Alex pushing back 2.3
speaking balance: gold is Alex, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 27:23 Rejection of Google leaked open source memo
Ramaswamy directly calls the viral Google engineer memo a 'misinformed opinion' and argues enterprise relationships and product defaults matter far more than raw model availability.
Hardest push from Alex ▶ 23:28 Challenging Google's decision to open source TransformersKantrowitz presses Ramaswamy on whether Google made a critical strategic error by open-sourcing the transformer architecture rather than building an exclusive defensive moat.
Biggest teaching moment ▶ 3:00 Contrasting legacy voice engines with transformer modelsRamaswamy details the underlying architectural difference between early voice assistants like Alexa/Google Assistant and modern transformer-based generative models.
Alex holds their own ▶ 20:31 Detailed interrogation of Google's ad unit mechanicsKantrowitz demonstrates sharp industry expertise by breaking down how generative AI search summaries cannibalize traditional blue link real estate and ad revenue.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Alex as informed peer | Guest teaching | Guest disagreement | Alex pushing back | Why |
|---|---|---|---|---|---|---|
| Evaluating Google's Turbulent 2023 and Strategic AI Reset | 6 | 7 | 2 | 3 | Kantrowitz challenges the narrative around Google Assistant being the right conversational product with poor execution. Ramaswamy educates him on the technical divergence between pre-transformer voice assistants and modern generative models. | |
| How Generative AI Dramatically Expands User Query Behavior | 5 | 6 | 2 | 2 | Kantrowitz asks whether chat interfaces truly expand queries beyond traditional search. Ramaswamy reframes query behavior using Neeva's backend observations on subjective multi-hop questions. | |
| Hallucinations, Trust, and Information Authenticity in the AI Era | 5 | 7 | 1 | 1 | Kantrowitz probes user trust and hallucinations in generative search results. Ramaswamy explains the architectural requirement of source citations and Google's internal 'flight to quality' routing. | |
| The Changing Search Paradigm and Open Web Content Consolidation | 6 | 6 | 2 | 3 | Kantrowitz pushes on how zero-click AI summaries disrupt web publisher economics. Ramaswamy agrees and outlines likely downstream effects, including publisher consolidation and site-specific chatbots. | |
| Reimagining Search Advertising and Monetization Models | 6 | 7 | 2 | 2 | Kantrowitz questions how Google monetizes full-page generative answers when blue links lose impressions. Ramaswamy draws on his tenure running Google Ads to explain unified commercial intent and API-driven purchasing. | |
| Google's Research Openness, Innovator's Dilemma, and Corporate Caution | 5 | 7 | 3 | 4 | Kantrowitz asks whether open-sourcing the Transformer paper was an unforced strategic blunder for Google. Ramaswamy reframes the research culture in 2017, explaining talent retention dynamics and corporate hesitation over chatbot safety. | |
| Analyzing Open Source Versus Frontier Foundation Models | 6 | 7 | 4 | 3 | Kantrowitz cites the leaked Google memo claiming neither Google nor OpenAI has a moat against open source. Ramaswamy flatly rejects the memo's premise, asserting products and frontier reasoning quality trump raw model weights. | |
| Multimodal AI Capabilities and Future User Interfaces | 5 | 6 | 2 | 2 | Kantrowitz asks about the transformative potential of multimodal models. Ramaswamy moderates consumer search hype, emphasizing enterprise document analysis and tool use over visual search. | |
| Mid-Roll Break and the Neeva-to-Snowflake Acquisition | 5 | 6 | 1 | 2 | Kantrowitz inquires about Neeva's pivot away from consumer search to an acquisition by Snowflake. Ramaswamy details macro interest rate shifts, valuation multiples, and distribution barriers on platforms like iOS. | |
| Applying Generative AI to Enterprise Data at Snowflake | 6 | 6 | 1 | 1 | Kantrowitz outlines enterprise data conversational workflows. Ramaswamy explains how Snowflake integrates text-to-SQL, document retrieval, and security perimeters for proprietary data. | |
| Lightning Round: AI Impact on White-Collar Employment | 5 | 5 | 2 | 2 | Kantrowitz kicks off the lightning round asking about white-collar job displacement and Musk's xAI venture. Ramaswamy predicts slow enterprise change alongside aggressive automation in customer support and simple info tasks. | |
| Lightning Round: AI Moratoriums, Governance, and Existential Risk | 6 | 6 | 3 | 2 | Kantrowitz notes the paradox of tech leaders warning about extinction while launching frontier labs. Ramaswamy dismisses calls for a six-month moratorium as absurd and advocates enforcing existing statutes. |
Statements from this episode (29)
Ramaswamy: AI boom exposed deep gaps in Google leadership
“I think at one level, it exposed fairly deep gaps in the ability of Google leadership, both to visualize the future, but also execute towards it.”
Ramaswamy: Rapid rollout of ChatGPT and AI products surprised Google
“It's very clear that the crazy popularity of ChatGPT and the speed at which others, including Neva, rolled out Pretty credible AI powered products. I think was pretty surprising for them.”
Ramaswamy: Google's competitive position improved by mid-2023 due to Bing's slow progress
“There's now a pretty credible integration into a regular search. And Bard is getting better by the day. And this combined with the fact that you know, Sydney, Bing's chatbot has not made as much progress. I would say actually puts them in a better position in …”
Ramaswamy: Google assigned thousands of employees to voice assistant development
“And Google actually put a multi-thousand person team to work on this, both in search, but also within my team. The shopping team, for example, had an assistant. We had partnerships with companies like Walmart. We took it seriously”
Ramaswamy: Generative AI surpasses voice assistants by synthesizing multi-page summaries
“The power of generative AI now, I think, comes from the fact that you can understand multiple pages, and you can write a fluid answer for way more queries than what the Assistant could ever do. Remember, with the Assistant, whether it's on Alexa or on Google p…”
Ramaswamy: Chatbots like Pi and Character.ai rely on gimmicks for long-term memory
“You know, they don't really have things like long-term memory. Like there's a bunch of technical gimmicks That people can use to have these bots pretend like they have long-term state.”
Ramaswamy: Conversational search sharply increased comparison queries at Neeva
“We don't look at sort of individual queries. That's one of the no-no's of any search engine. And so, like, we would do things like analysis on the length of queries, what sort of quality that that, that we would serve. But as I said, comparison queries increas…”
Ramaswamy: Neeva required every sentence in AI search to include citations
“And this is why we were very persnickety about making sure that every sentence that we provided came with a citation.”
Ramaswamy: Google delayed AI search because users inherently trust its text output
“Similarly any text that is generated by a chatbot is, and this is part of the reason why Google has been hesitant is if they put up some text, even if there's a citation, people are going to say, oh, Google said, so this must be true.”
Ramaswamy: Synthetic media barrage will severely challenge human perception of reality
“I think, like, our notion of reality is going to be subject to such a barrage of, like, fake and real signals that I think it's going to be a real problem for us to keep our heads straight.”
Ramaswamy: Google routed trending topics to trusted sites to avoid conspiracy theories
“Google had this system that was roughly called flight to quality. Whenever there would be a new and trending topic the search algorithm would basically say, I'm going to go to trustworthy sites because if it is a brand new topic, the likelihood that someone sp…”
Ramaswamy: AI language models trained on AI-generated content eventually produce trash
“If these language models learn on content that other language models have generated beyond a point, they just generate trash.”
Ramaswamy: NYT and WSJ are the only successful open-web newspaper businesses
“The sort of two credible pure information businesses, newspapers that have come out of that are the New York Times and the Wall Street Journal. And everybody else is a little bit off and also ran.”
Ramaswamy: On-site chatbots will become the standard for publisher content discovery
“Meaning that chatbots are going to be the norm for how information is discovered on a on a site once you get that person to come to that site.”
Ramaswamy: Google merged organic and paid shopping to eliminate duplicate search engines
“Google used to have organic shopping. I famously made combined organic shopping and paid shopping because I was like, all of this is commercial content. I can't have four search engines on one search page. With organic shopping, paid shopping paid text adverti…”
Ramaswamy: Consumers will chat directly with websites to execute purchases
“I think you're going to see experiences where you can again chat with a website and be able to drive purchases off of it.”
Ramaswamy: Google open-sourced Transformers in 2017 to prevent top researchers from leaving
“If you had told these people at that time that they could not publish, they would have gone and worked for universities, they would have gone and worked for Microsoft, they would have gone and worked for other people.”
Ramaswamy: Google's delay in generative AI was an internal choice, not technical
“And remember Google's lack of progress in generative AI like, you know, that's internally driven. Nothing stopped them from creating ChatGPT. They chose not to.”
Ramaswamy: GPT-4 and Claude are a clear step ahead of open source models
“The blunt truth is that the very best of the models out there, whether it's GPT-IV or Claude's biggest model, are a clear step ahead of the pack when it comes to quality. When it comes to reasoning, when it comes to the quality of the text that they produce th…”
Ramaswamy: ChatGPT's user growth has flattened out
“ChatGPT, you know, has had growth, but it has sort of flattened out.”
Ramaswamy: Open source AI does not substantially threaten incumbent businesses
“I don't think the mere presence of open source models threaten the existing businesses as much.”
Ramaswamy: Multimodal AI models will see strong adoption for parsing PDFs
“I think multimodal models will have a lot more business use cases where you're looking at PDFs.”
Ramaswamy: Multimodal AI will not significantly impact consumer search beyond image generation
“I have a harder time you know, other than for things like image generation, how multimodal is going to make a humongous difference. For something like consumer search.”
Ramaswamy: Safari restricts default search engines to five preselected options
“Outside of the prescribed five, You can't change the search engine on Safari to anything else, even today.”
Ramaswamy: Great search retrieval is critical for believable generative AI
“In my mind, one of the key ingredients for believable AI, for referenceable AI, is a great search retrieval system that sets the context for how a language model is going to generate answers.”
Ramaswamy: Snowflake bets generative AI will 10x users and 100x queries
“Our bet is that we can 10 X the number of users that can use the platform, hundred X the number of queries that are going to be done on the platform.”
Ramaswamy: Calls for a six-month AI moratorium are absurd
“I found things like the call for a moratorium for six months to be absurd.”
Ramaswamy: Individuals can fine-tune an AI model for $500 in one evening
“You can fine tune a model for 500 bucks in one evening without a whole lot of technical skill.”
Ramaswamy: Nvidia's Graphics Architecture Perfectly Positioned It for AI Dominance
“The same things that made them really good for doing graphics processing, which is a lot of You know, fairly simple operations done at massive scale. You know, graphics processing is like drawing a lot of triangles on your screen. But the same thing and matrix…”