https://www.youtube.com/watch?v=e-gwvmhyU7A
#!/bin/bash
randomtext=$(head /dev/urandom | tr -dc A-Za-z | head -c 12 ; echo '')
url=$1
model="claude-3-opus-20240229"
/home/dez/bin/youtubeimage $url
echo "Getting you the deets on the veed!"
yt --transcript https://youtu.be/$url | fabric --model $model -sp extract_wisdom > $url-$randomtext-$model.md
echo "Alrighty! We're DONE! Check out the markdown file here: $url-$randomtext-$model.md"
Made with Fabric
FULL TRANSCRIPT:
https://lexfridman.com/aravind-srinivas-transcript
PODCAST EPISODE #434
https://podcasts.apple.com/gb/podcast/lex-fridman-podcast/id1434243584?i=1000659563748
SUMMARY:
Arvind Srinivasan, CEO of Perplexity, discusses how Perplexity combines search and large language models to provide answers backed by sources, reducing hallucinations. He shares insights on AI, search, startups, and the future of knowledge discovery.
DIGGING DEEPER:
This conversation between Arvin Sookias (CEO of Perplexity) and Lex explores the relationship between curiosity, purpose, and human nature. They discuss how people's curiosity can lead them to build true connections with others and pursue their interests, which can bring fulfillment and actualization.
However, they also touch on the darker aspects of human nature, such as biases and manipulation. Arvin shares his concerns about building AI-powered relationships that mimic emotional connections, which could potentially lead to dystopian outcomes if not done thoughtfully.
The conversation highlights the importance of curiosity, knowledge, and understanding in preserving human consciousness and intelligence. They believe that people are naturally curious and want to learn, and that this desire can be harnessed to improve our understanding of each other and the world.
Arvin hopes that AI can help humans reduce their biases and increase their understanding of each other's perspectives, leading to a more peaceful and loving world. He sees Perplexity as a tool for facilitating this kind of understanding and encouraging people to think critically about complex issues.
Throughout the conversation, Lex encourages Arvin to consider the potential risks and challenges of developing AI-powered relationships that simulate emotional connections. Arvin acknowledges these concerns but emphasizes his commitment to using technology in a way that promotes human flourishing and actualization.
The conversation concludes with a quote from Albert Einstein, emphasizing the importance of curiosity and awe in exploring the mysteries of life and the universe.
IDEAS:
- Perplexity provides answers backed by sources, combining search and LLMs
- Perplexity aims to be a knowledge discovery engine, not just search or answers
- Open source models are important for enabling more players in AI
- Retrieval augmented generation is key - LLMs should only say things they can cite
- Search requires a lot of domain knowledge - indexing, ranking, UX are all critical
- Inference compute, not just model size, will be key for reasoning breakthroughs
- AI systems that can research a topic deeply and come back with insights will be powerful
- Curiosity is a fundamental human trait that AI should cater to
- AI has potential to help humans be more truth-seeking and reduce biases
- The future is about AI-assisted knowledge discovery, not just search
INSIGHTS:
- Combining search, LLMs, and citations enables powerful knowledge discovery
- Open source democratizes AI progress and prevents concentration of power
- Search quality depends on indexing, retrieval, ranking, UX - not just bigger models
- Iterative inference compute could lead to AI systems with deep reasoning abilities
- AI that acts as a thought partner for humans, enhancing curiosity, will be transformative
- The end goal is AI-assisted knowledge discovery to help humans understand the world better
QUOTES:
"The journey doesn't end once you get an answer. The journey begins after you get an answer."
"I think open source is important because it gives you a good base model to start with and try different experiments in the post-training phase."
"Search is a hard problem. There's a lot of domain knowledge involved."
"If we can achieve that amount of inference compute where it leads to a dramatically better answer as you apply more inference compute, I think that would be the beginning of real reasoning breakthroughs."
"I think curiosity makes humans special and we want to cater to that. That's the mission of the company."
HABITS:
- Thinking long-term and having a clear mission for your company
- Constantly testing your product, even in suboptimal conditions, to improve it
- Hiring people who are genuinely passionate about the problems you're solving
- Optimizing for the user experience above all else
- Staying up to date on the latest research and being willing to question assumptions
FACTS:
- Netflix uses over 100,000 server instances on AWS for its computing and storage needs
- Google's search advertising model was inspired by a company called Overture
- Nvidia's next generation GPUs are expected to be 30x more efficient than current ones
- Humans are more likely to search for themselves than anyone else on social media
- BM25, a ranking function, often still outperforms embedding-based retrieval methods
REFERENCES:
- Google's research on latency and the importance of fast responses
- The book "Invent and Wander" about Jeff Bezos and Amazon's principles
- Yann LeCun's research on unsupervised learning and critique of pure autoregressive models
- The paper "Chain of Thought Prompting" on improving reasoning in language models
- David Deutsch's book "The Beginning of Infinity" on the power of knowledge creation
ONE-SENTENCE TAKEAWAY:
Perplexity aims to enhance human knowledge discovery and flourishing through AI-powered search and answers.
RECOMMENDATIONS:
- Experiment with different ways of combining search, retrieval, LLMs for knowledge discovery
- Invest in open source models to democratize progress and prevent concentration of power
- Focus on optimizing the end-to-end user experience, not just model size
- Explore techniques for imbuing AI systems with genuine curiosity to assist human discovery
- Keep the ultimate goal of expanding human understanding and reducing biases in mind
QUESTIONS TO ASK
-
Can you have a conversation with an AI where it feels like you talk to Einstein or Feynman where you ask them a hard question they don't know and then after a week they do a lot of research and come back and just blow your mind?
-
If we can achieve that amount of inference compute where it leads to a dramatically better answer as you apply more inference compute, do you think that would be the beginning of real reasoning breakthroughs?