People & Blogs

Databricks CEO: AGI Is Already Here, Why Companies Are Behind, and the Superintelligence Debate

by All-In Podcast

Share:

📚 Main topics

  • Databricks’ origins and strategyAli Ghodsi says Databricks grew out of UC Berkeley’s AMPLab, where researchers saw potential in combining data processing with machine learning. The company committed early to cloud computing, AI, and open source—bets that were considered risky at the time. 2:04
  • Why enterprise AI adoption is slowMany organizations still use AI mainly for search and coding. Ghodsi points to barriers including missing or inaccessible business context, security concerns, rising token costs, and the risk of giving away proprietary information. 12:24
  • Databricks’ Genie demoGenie builds an organizational knowledge graph from company information, such as documents, meetings, and reporting lines. Ghodsi demonstrates using it to analyze employee survey results, organizational structure, engineering output, and customer feedback. 23:18
  • AI, AGI, and recursive self-improvementGhodsi argues that AI already meets a practical definition of AGI, but says there is no evidence of the rapid recursive self-improvement required for the superintelligence scenario described by Nick Bostrom. 9:19
  • Going public and staying privateGhodsi says public markets could bring capital, visibility, and easier liquidity, but stock-price swings could distract employees and management during a period of rapid AI change. 38:48
  • Cybersecurity and AI’s broader effectsGhodsi sees cyber risk as a serious concern, while arguing that AI can help find and fix vulnerabilities. He also discusses AI’s potential in drug discovery and crisis support. 50:40

✨ Key takeaways

  • Enterprise context is essentialAI can only help with organizational work if it has relevant information about decisions, people, and processes. In many companies, privacy and liability concerns make collecting that context difficult. 13:57
  • Agents must be secure and cost-effectiveDeploying agents at scale requires controlling what they can access, protecting company data, and managing inference costs so that AI use makes economic sense. 14:27
  • AI can reveal organizational patternsGenie’s analyses can help leaders find coordination bottlenecks, surface customer needs, and compare how teams work—all with the aim of reducing delays and human handoffs. 29:32
  • The RSI threshold is demandingGhodsi lays out four conditions for recursive self-improvement: training must get much faster, use fewer resources, improve accuracy broadly, and repeat in a continuous loop. He says those conditions are not currently being met. 46:34
  • Cybersecurity risk is asymmetricAttackers need only one successful breach, while defenders must protect against every attempt. Ghodsi argues that AI-assisted scanning and remediation can help address vulnerabilities in new and legacy code. 51:13

🧠 Lessons learned

  • Contrarian bets can create an advantageDatabricks’ early commitment to cloud, AI, and open source helped it build a lead while others doubted those directions. 6:14
  • Technology alone doesn’t change organizationsEven capable AI tools face entrenched software, regulation, security requirements, and internal processes. Enterprise adoption is likely to take years rather than months. 4:07
  • Use data to investigate, not just to confirmThe demo illustrates how leaders can test assumptions against evidence, identify differences across teams, and spot gaps between stated company values and actual behavior. 25:23
  • AI can accelerate work without eliminating human judgmentAgents can prepare analyses and move information across teams, but human oversight remains important for decisions, security, and areas where context is incomplete. 32:34

📚 Conclusion/next steps

  • Databricks’ focusGhodsi says the company is working to make AI useful inside organizations by connecting it to business context while addressing governance, security, cost, and data ownership. 17:02
  • A gradual transitionGhodsi expects enterprise AI adoption to take roughly a decade or longer, partly because many organizations are not yet ready to capture the information AI needs. 59:58
  • Balance opportunity with cautionHe sees substantial potential in AI, including healthcare applications, while emphasizing that cybersecurity deserves serious attention and that claims about imminent superintelligence remain unsupported by the evidence he sees. 55:50

Transcript excerpt

0:00 So it was kind of out of despair that we started Databricks because no one else is listening to it. At that time there is zero interest in AI. Now I think we have all the intelligence we need to completely transform the planet. >> It's not possible to have a thousand agents doing stuff realistically. Am I correct? >> They're just using AI in very rudimentary ways. They're just using it as a glorified Google search. We're actually getting into like management money ball, aren't we? >> So if the token cost is going up exponentially, is your revenue at your company going up exponentially? But I actually think GDP will move very slowly

0:30 and it's going to take decades. >> Are they a little drunk on their own success? >> Just because 98% of the code is being written by AI, we don't have RSI. If it automates the last 2%, who cares? >> When are you going public? >> I'm going. >> If you were building a global financial system from first principles today, you wouldn't build it on 50-year-old legacy rails. You'd build Airwallex, one AI native platform for global accounts, cards, and payments. is designed to make the entire world feel like a local

🔒 The full, searchable transcript is available with Pro.

🔒 Unlock Premium Features

This is a premium feature. Upgrade to unlock unlimited Q&A, transcripts, mindmaps, and translations.

Questions & Answers

Common questions about this video

Why did Ali Ghodsi and his colleagues start Databricks?

They were researchers at UC Berkeley’s AMPLab, where they worked on combining data processing with AI. Because the industry showed little interest in the technology at the time, they started Databricks to help organizations adopt it. 2:34

What early bets helped Databricks succeed?

Databricks bet on cloud computing, AI, and open source when those choices were controversial. Ghodsi said those contrarian bets gave the company a head start. 6:14

Why does Ghodsi think enterprises are behind in adopting AI?

Many organizations use AI mainly as a glorified search engine or for coding, rather than deploying agents to do core work. He cited challenges including limited access to organizational context, security risks, rising token costs, and concerns about retaining control of company data and intellectual property. 12:24

How does Databricks’ Genie help Ghodsi manage the company?

Genie builds an ontology—a graph of company information, including people, reporting relationships, documents, communications, meetings, and decisions. Ghodsi said it can use that context to answer questions and analyze issues that might otherwise take his team days to investigate. 23:18

Why is Databricks staying private for now?

Ghodsi said being public offers benefits such as access to capital and visibility, but he wants to avoid stock-price swings distracting employees and management during the AI transition. He said Databricks will go public, but not yet. 42:25

What four conditions does Ghodsi say would be needed for recursive self-improvement?

A future model would need to be trained much faster, use fewer resources, improve its accuracy broadly, and then repeat that cycle continuously so each successive model becomes faster, cheaper, and smarter. Ghodsi said he does not see that flywheel happening now. 46:34

What does Ghodsi see as the most immediate AI-related risk?

He sees cyber risk as the most immediate concern, particularly because attackers only need to find one weakness while defenders have to protect against all of them. He also said AI can help scan code for vulnerabilities and that legacy code remains a concern. 51:13

How quickly does Ghodsi think AI will affect the broader economy?

He expects the economic impact to unfold slowly, over roughly 10 to 15 years, because many organizations still have barriers such as restrictions on recording meetings and difficulty sharing the context AI would need to help with their work. 59:58

🔒 Unlock Premium Features

Access to Chat is a premium feature. Upgrade now to unlock unlimited studying tools.

🔒 Unlock Premium Features

Access to Mindmap is a premium feature. Upgrade now to unlock unlimited studying tools.

🔒 Unlock Premium Features

Access to Translation is a premium feature. Upgrade now to unlock unlimited studying tools.

Get unlimited summaries, Q&A, transcripts and more with Pro

Upgrade to Pro

Suggestions

🔒 Unlock Premium Features

Access to AI Suggestions is a premium feature. Upgrade now to unlock unlimited studying tools.