# How Classical Thinking Shaped Modern AI | Peter Danenberg | TEDxBoston Source: https://www.youtube.com/watch?v=1BMMUTjp3hk Recap page: https://rapidrecap.app/video/1BMMUTjp3hk Generated: 2026-01-21T17:15:07.952+00:00 --- ## Quick Overview The speaker argues that the rapid advancements in AI, particularly Large Language Models (LLMs) like those from Google, are largely due to an accelerationist mindset rooted in classical thinking, drawing a parallel to the hypothetical scenario of sending modern AI knowledge back to the 1970s, which would have dramatically accelerated progress compared to the slow, incremental development seen historically. **Key Points:** - The speaker attributes the rapid progress in modern AI, especially LLMs, to an underlying accelerationist philosophy derived from classical thinking. - He references the concept of sending knowledge about modern AI (like LLMs) back to the 1970s, suggesting progress would have been much faster than the historical timeline. - The speaker notes that Google's early approach to AI was perceived as 'weird' because they prioritized pushing boundaries, which he views as positive, even if it led to PR issues like the initial chatbot issues (e.g., Bard's early performance). - He contrasts current LLMs with previous models like LSTMs, suggesting that the new architecture (Transformers) is not inherently a step function change but rather a scaling issue, though he admits this is debatable. - The speaker mentions that he is currently working on 'Ambiant LLM' and suggests that if the current rapid pace continues, we might need new thinking rather than just scaling existing models to avoid a 'creepy' outcome. - The discussion concludes with a hypothetical question: if you sent modern AI knowledge back 40 years, would we have achieved AGI sooner, suggesting that the acceleration is the key factor, not necessarily a single architectural breakthrough. ![Screenshot at 00:09: Peter Danenberg speaking during the Q&A session, gesturing while discussing the influence of classical thinking on modern AI development.](https://ss.rapidrecap.app/screens/1BMMUTjp3hk/00-00-09.jpg) **Context:** This segment is part of a TEDxBoston talk featuring Peter Danenberg, who discusses the role of classical philosophical thinking and an 'accelerationist' mindset in driving modern advancements in Artificial Intelligence, specifically Large Language Models (LLMs). The conversation involves an unnamed interviewer (John) who prompts Danenberg to reflect on his past experiences at Google and the trajectory of AI development relative to historical progress. ## Detailed Analysis Peter Danenberg discusses how an underlying accelerationist mindset, which he traces back to classical thinking, is responsible for the current rapid advancements in AI, exemplified by the success of Large Language Models (LLMs). He recalls his time at Google, noting that while the company sometimes faced PR challenges (like early issues with Bard), their willingness to pursue unconventional paths was crucial. He contrasts the current Transformer architecture with older models like LSTMs, suggesting the difference is less about a fundamental architectural leap and more about the massive scaling achieved. He expresses concern about a potential 'AI bubble' and the possibility that unchecked acceleration could lead to negative outcomes, suggesting caution is needed in how we deploy this technology relative to humanity's capacity to adapt. He references the idea that if knowledge about current AI capabilities (like GPT-4) were available in the 1970s, progress would have been exponentially faster. The discussion ends with the interviewer posing a hypothetical question about what one would change if sent back in time to accelerate AI development, implying that accelerating the timeline itself is a key element of the current AI boom. ### Early AI Philosophy at Google - Google hired 'weird people' who pushed boundaries, even if it caused PR issues like Bard's initial performance - Danenberg took the 'weird' label as a compliment, contrasting it with the more cautious approach that led to him not getting into 2025 Google. ### LLM Architecture vs. Scaling - The leap from LSTMs to Transformers might not be a true step function but rather a result of scaling, as evidenced by the 'hockey stick' growth in performance coinciding with increased token training. ### Future Concerns and Ethics - Danenberg expresses concern about an 'AI bubble' and the potential for superintelligence to punish those who fail to keep up; he advocates for careful, measured deployment rather than pure accelerationism. ### Hypothetical Time Travel Question - The interviewer asks what the listener (Danenberg) would do if sent back 40 years with current AI knowledge to accelerate AGI, suggesting the acceleration itself is a powerful variable. ![Screenshot at 00:04: Introduction slide displaying "IDEAS IN ACTION PRESENTS" followed by the TEDxBoston logo.](https://ss.rapidrecap.app/screens/1BMMUTjp3hk/00-00-04.jpg) ![Screenshot at 00:06: Peter Danenberg \(left, in suit\) and the interviewer \(right, holding a mic\) seated on stage during the discussion.](https://ss.rapidrecap.app/screens/1BMMUTjp3hk/00-00-06.jpg) ![Screenshot at 00:29: Danenberg speaking animatedly while describing the early days of AI development at Google.](https://ss.rapidrecap.app/screens/1BMMUTjp3hk/00-00-29.jpg) ![Screenshot at 01:01: The interviewer asking a question while holding up a small device \(likely a phone or remote\).](https://ss.rapidrecap.app/screens/1BMMUTjp3hk/00-01-01.jpg) ![Screenshot at 01:47: Danenberg holding the microphone and gesturing while discussing the rapid pace of AI development.](https://ss.rapidrecap.app/screens/1BMMUTjp3hk/00-01-47.jpg)