Stripe

A Cheeky Pint with Google

457 summary words 2 min summary Watch video

Start with the signal

2 min read

Summary

A Cheeky Pint with Google - Summary

Main Topics

  • Google's Role in AI Development: Transformer invention and search integration
  • AI Product Development: Comparison between Google's internal projects and external commercialization
  • AGI Strategy and Investment: Google's capital scaling and AI ambitions
  • Technical Constraints: Infrastructure and memory limitations in product deployment
  • Future AI Agents: Timeline for agentic AI products at Google

Key Points

Transformer Technology & Search

  • Google invented Transformers but the technology was productized outside the company
  • Transformers were immediately integrated into Google Search to improve:
  • Language understanding
  • Web page comprehension
  • Query interpretation

Missed Opportunities

  • Google internally conceived of a product similar to ChatGPT (called Lambda)
  • The product was not shipped by Google but emerged successfully elsewhere
  • Suggests Google could have shipped a comparable product approximately 9 months earlier

AGI Perception Gap

  • External researchers perceive Google as less "AGI-pilled" (committed to AGI development)
  • There's a misconception that Google doesn't fully understand AGI
  • Despite this perception, Google assembled significant AGI talent (Demis, Jeff, Ilya, Dario)

Capital Investment & Commitment

  • Google scaled capital expenditure from $30 billion to $180 billion
  • This 6x increase signals serious commitment to AI development
  • Such spending decisions reflect confidence in scaling curves

Technical Engineering Excellence

  • Search infrastructure divided into four sub-teams with millisecond-level latency budgets
  • Shipping optimizations that save 3 milliseconds earn 1.5 milliseconds for budget reallocation
  • The remaining 1.5 milliseconds directly benefits end users
  • Demonstrates the precision required in Google's systems

Infrastructure Constraints

  • Memory capacity improvements from leading companies are limited in the short term
  • Hardware constraints pose practical challenges for near-term development

Notable Quotes

> "We built Transformers and used it immediately in search to improve language understanding, understanding web pages, understanding your queries."

> "You don't do it if you don't think about the curve a certain way." (regarding the $180B capex investment)

> "That's how much we think it matters." (referring to millisecond-level optimizations)

> "I expect 2027 to be a big year in which some of those shifts happen pretty profoundly." (on agentic AI deployment)

Takeaways

  • Google has AGI ambitions despite external perception—evidenced by massive capital allocation and talent recruitment
  • Execution gap exists: Google's innovation in foundational AI doesn't always translate to market-leading products
  • Timeline expectation: Fully agentic AI products from Google are anticipated around 2027
  • Infrastructure matters: Google's competitive advantage lies in precision engineering and optimization at scale
  • Current constraints are temporary: Memory and hardware limitations will be addressed, but are bottlenecks in 2025-2026
  • Early adoption in engineering: AI agents are being tested internally before broader deployment
Search this video Resume transcript at 0%
Full transcript 452 words · 3 min read
0:00

SPEAKER_00

Transformers were invented at Google, but then productized outside of Google. It's a bit misunderstood. We built Transformers and used it immediately in search to improve language understanding, understanding web pages, understanding your queries. We even conceived the product, which is ChatGPT, but it was Lambda. In the multiverse somewhere else, Google probably shipped that nine months later. When I talk to researchers at the other labs, they feel like Google is not as AGI-pilled. This notion that at Google we haven't understood what AGI is. At one point, Demis, Jeff, Ilya, Dario were all there. We probably have scaled our capex from 30 billion to 180 billion. You don't do it if you don't think about the curve a certain way. Search has four sub-teams with latency budgets in the milliseconds. If you ship something that shaves off three milliseconds, you earn 1.5 milliseconds for your latency budget, and 1.5 milliseconds gets passed on to the user. That's how much we think it matters. There is no way that the leading memory companies are going to dramatically improve their capacity. So you have those constraints in the short term. Eng is an early adopter, but outside of Eng. What quarter do you think Google's first fully agentic forecast is? I expect 2027 to be a big year in which some of those shifts happen pretty profoundly.

0:01

SPEAKER_00

but then productized outside of Google. It's a bit misunderstood. We built Transformers and used it immediately in search to improve language understanding, understanding web pages, understanding your queries. We exactly even conceived the product, which is like ChatGP need was Lambda. In the multiverse somewhere else, Google probably shipped that nine months later. When I talk to researchers at the other labs, they feel like Google is not as AGI-pilled. This notion that at Google we haven't understood what AGI is. At one point, Demis, Jeff, Ilya, Dario were all there. We probably have scaled our capex from 30 billion to 180 billion.

0:41

SPEAKER_00

You don't do it if you don't think about the curve a certain way. Search, they now have for sub-teams, like latency budgets, like in the milliseconds. If you ship something with shaves of three milliseconds, you earn 1.5 milliseconds for your latency budget, and 1.5 milliseconds gets passed on to the user. That's how much we think it matters. There is no way that the leading memory companies are going to dramatically improve their capacity. So you have those constraints in the short term. Eng is an early adopter, but kind of outside of Eng. What quarter do you think Google's first fully agentic forecast is?

1:19

SPEAKER_00

I expect 27 to be a big year in which some of those shifts happen pretty profoundly.

1:23

SPEAKER_00

So connect to them to connect to them to connect to them to connect to them to connect to them to connect to them to connect to them to connect to them to connect to them to connect to them to connect to them to connect to them to connect to them to connect to them to connect to them to connect to them to connect to them to connect to them to connect to them to connect to them to connect to them to connect to them to connect to them to connect to them to connect to them to connect to them to connect to them to connect to them to connect to them to connect to them to connect to them to connect to them to connect to them to connect to them to connect to them to connect to them to connect to them to connect to them to connect to them to connect to them to connect to them to connect to them to connect to them to connect to them to connect to them to connect to them to connect to them to connect to them to connect to them to connect to them to connect to them to connect to them to connect to them to connect to them to connect to them to

Reading tools

Type to find a passage

Appearance
Ask this transcript

Notebook

Export Markdown

Add a note