Tim Scarfe
Creator & Host, Machine Learning Street Talk
關於
Tim Scarfe is the creator and host of Machine Learning Street Talk (MLST), an in-depth technical AI podcast and YouTube channel he started at the beginning of the COVID-19 pandemic in 2020 alongside Yannic Kilcher, Keith Duggar and Connor Shorten. He holds a PhD in machine learning and a first-class degree in computer science, started the software company Dot Net Solutions in 2003, and has since done stints as a Principal Engineer at Microsoft and as Chief Data Scientist at bp. Keith Duggar, who obtained his PhD from MIT and has worked at IBM Research, on Wall Street and at Microsoft, is often his co-interviewer. Scarfe states the show's ethos as "intellectual curiosity, depth, diversity, forward-thinking and production quality", covering symbolic AI, deep learning research, evolutionary methods, AI safety, AI philosophy and highly skeptical positions, and says he deliberately chose not to interview top tech CEOs. His own GitHub profile also lists him as CTO of the augmented-reality company XRAI.
主要貢獻
- Created Machine Learning Street Talk in 2020 and has hosted it since, publishing long-form technical interviews with AI researchers
- Deliberately spans the field's disagreements — symbolic AI, deep learning research, evolutionary methods, AI safety, AI philosophy and highly skeptical positions — rather than one camp
- Says the show covered LLMs, test-time compute, active inference, neurosymbolic models and open-endedness years before they became mainstream
- MLST's conversations in this collection carry Karl Friston on the free energy principle, Jürgen Schmidhuber, Jeff Clune and Ben Goertzel
- Before MLST: founded Dot Net Solutions in 2003, was a Principal Engineer at Microsoft and Chief Data Scientist at bp, and completed a PhD in machine learning
主持過的 4 場對話
思想連結
傑夫·克盧恩
曾經對談英屬哥倫比亞大學電腦科學教授
兩個小時，談的是貫穿 Clune 整個生涯的那個論證：要求人類去設計智能，等於是在要更快的馬；替代方案則是開放式的達爾文式搜尋，在豐富到足以持續生成自身挑戰的空間裡進行。這場對話一路談到他共同創辦 Recursive 的起點——把 AI 研究交給 AI，意味著什麼，又冒了什麼險。
youtube.com
班·格策爾
曾經對談創辦人暨執行長・SingularityNET
Goertzel 在這一集把他與 Ray Kurzweil 的分歧講得很精確：Kurzweil 把人類水準的 AGI 放在 2029 年、奇點放在 2045 年，而 Goertzel 認為中間那十六年的落差，只是把曲線硬套在「由人類發明」這件事上的產物。一旦發明的是 AGI 自己——自己設計晶片、自己設計網路——指數就變了，落差會塌縮成幾年。這是本站兩位人物之間一場具體而且可標上日期的爭論。
youtube.com
尤爾根·施密特胡伯
曾經對談KAUST AI 計畫主任、IDSIA 科學總監
這一集的標題毫不遮掩：《那個發明了現代 AI 的人（比所有人都早）》——這既是 Schmidhuber 自己講了多年的主張，也是多數訪問者不願意大聲說出口的話。把場子讓給他，讓他完整地把「優先權」這件事說一遍，正是重點：深度學習的歷史比它的獎項引用所暗示的更久、也更不盎格魯美利堅，而他是這段記憶最不屈不撓的守護者。
youtube.com
Karl Friston
曾經對談神經科學家與理論生物學家
自由能原理出了名地難以獲得一個乾淨的陳述——單一個想法宣稱能涵蓋知覺、行動、學習，以及生命為何得以持存；而人們通常不是從方程式、就是從把它弄丟了的轉述裡遇見它。一集節目什麼都不做，只是讓 Friston 從頭把它鋪陳完整，本身就是一項真正的服務；這也是為什麼這個頻道會一再出現在本站的引用裡。
youtube.com
