Executive Code

Kirimgeray Kirimli & Rajiv Pant
  • 5,0 (1)
  • TECNOLOGIA

Welcome to Executive Code — where AI, engineering, and leadership collide. Hosted by the minds behind Snapshot AI and Flatiron Software, this podcast brings you inside conversations with the people shaping the future of technology. From deep dives into the latest AI research to unfiltered takes on engineering leadership, each episode is a front-row seat to the real talk happening behind the scenes in the tech world. Whether you're a CTO, a machine learning engineer, or just someone curious about where the future is heading — this is your space. Filmed across New York, London, and wherever the conversation takes us. Subscribe and join us every other week.

  1. The Future of Speech: Building Models That Listen, Think, and Talk | EP #21 Executive Code

    11/11/2025

    The Future of Speech: Building Models That Listen, Think, and Talk | EP #21 Executive Code

    In this episode of Executive Code, Kirim talks with Julian Salazar, Senior Research Scientist at Google DeepMind’s Foundational Research Unit, about the next frontier of AI — native speech models. They explore how systems that generate and understand audio directly could bridge the gap between text and speech, enabling agents that reason, respond, and even express emotion. From WaveNet to Speech SSM and the Sequence Layers library, Julian explains how these advances are redefining the limits of human-AI communication. For those who want to dive deeper into the research discussed in this episode: - Spectron: https://research.google/blog/spoken-question-answering-and-speech-continuation-using-a-spectrogram-powered-llm/ - SpeechSSM: https://google.github.io/tacotron/publications/speechssm/ - SequenceLayers: https://github.com/google/sequence-layers - Albert Gu’s blog post: https://goombalab.github.io/blog/2025/tradeoffs/ - Gemini Native Audio with Affective Dialog: https://aistudio.google.com/live?model=gemini-2.5-flash-native-audio-preview-09-2025, then select “Affective Dialog” in settings ------ Connect with ⁠Julian Salazar on LinkedIn⁠ https://www.linkedin.com/in/julianslzr/ Connect with⁠ A. Kirimgeray Kirimli on LinkedIn ⁠ https://www.linkedin.com/in/a-kirimgeray-kirimli-72bbb838/ ---- Flatiron Software: ⁠https://flatiron.software/⁠ Snapshot AI: ⁠https://www.snapshot.reviews/⁠

    1h1min
  2. The Limits of Marginalization: When Efficient Prediction Breaks the Rules | EP #20 Executive Code

    29/10/2025

    The Limits of Marginalization: When Efficient Prediction Breaks the Rules | EP #20 Executive Code

    In this episode of Executive Code, Kirim sits down with Sanyam Agarwal (Saarland University) and Oliver Broadrick (UCLA) to discuss their paper The Limits of Tractable Marginalization. The conversation dives deep into one of AI’s core problems — how models handle uncertainty through marginalization. It’s the process behind every probabilistic decision, from predicting the next token in ChatGPT to guiding self-driving cars. Sanyam and Oliver challenge a long-standing assumption in machine learning: that if a system can compute these probabilities efficiently, it must be representable in a simple, compact model. Their findings prove this isn’t always true, uncovering hidden limits in how we understand tractability — and opening new questions for the future of probabilistic AI. ------ Connect with Sanyam Agarwal: https://www.linkedin.com/in/sanyam-agarwal-97326114a/ Connect with Oliver Broadrick: https://www.linkedin.com/in/oliverbroadrick000/ Connect with A. Kirimgeray Kirimli: https://www.linkedin.com/in/a-kirimgeray-kirimli-72bbb838/ ---- Flatiron Software: https://flatiron.software/ Snapshot AI: https://www.snapshot.reviews/

    26min
  3. The Value of Prediction: When Algorithms Meet Public Policy | EP #19 Executive Code

    22/10/2025

    The Value of Prediction: When Algorithms Meet Public Policy | EP #19 Executive Code

    In this episode of Executive Code, Kirim sits down with Unai Fischer-Abaigar PhD researcher at the LMU Munich, to discuss his paper The Value of Prediction in Identifying the Worst-Off. They explore how governments use AI to allocate limited resources—and when it’s better to improve predictive models versus simply expanding access to public programs. Using real data from Germany’s employment offices, Unai’s research challenges the assumption that better prediction always means better outcomes in public decision-making. ------ Connect with Unai Fischer-Abaigar: https://www.linkedin.com/in/unai-fa/ Connect with A. Kirimgeray Kirimli: https://www.linkedin.com/in/a-kirimgeray-kirimli-72bbb838/' ---- Flatiron Software: https://flatiron.software/ Snapshot AI: https://www.snapshot.reviews/

    23min
  4. RAGs as Students: Measuring Model Ability with Item Response Theory | EP #18 Executive Code

    15/10/2025

    RAGs as Students: Measuring Model Ability with Item Response Theory | EP #18 Executive Code

    In this episode of Executive Code, host Kirim talks with Behrooz Omidvar-Tehrani, Principal Applied Scientist at Microsoft, about a new framework for evaluating retrieval-augmented generation (RAG) systems. Instead of relying on costly human benchmarks or shallow metrics, Behrooz and his co-authors propose treating models like students—using synthetic, domain-specific exams and Item Response Theory from psychometrics to measure real ability. The discussion dives into how this method uncovers what truly drives performance, from retrieval quality to model architecture, and why better evaluation might be the missing piece for the next generation of AI systems. ------ Connect with Behrooz Omidvar-Tehrani: / behroozomidvar Connect with A. Kirimgeray Kirimli: / a-kirimgeray-kirimli-72bbb838 ---- Flatiron Software: https://flatiron.software/ Snapshot AI: https://www.snapshot.reviews/

    27min
  5. Deeper Safety for AI: Beyond Token-Level Alignment | EP #17 Executive Code

    03/10/2025

    Deeper Safety for AI: Beyond Token-Level Alignment | EP #17 Executive Code

    In this episode of Executive Code, host Kirim sits down with Ahmad Beirami, Senior Researcher at Google DeepMind, to explore why most AI safety alignment is only “skin-deep”—and how his team is working to fix it. They unpack why jailbreaks are so effective, what it means when safety stops after just a few tokens, and the breakthrough methods that can make alignment resilient throughout a model’s behavior. Together, they cover: - How today’s alignment methods leave models vulnerable to simple attacks - Why adversarial examples expose deeper safety risks - New approaches to embedding safety reasoning into AI models This is a crucial conversation for anyone interested in building AI systems that are not just powerful, but trustworthy at scale ------ Connect with Ahmad Beirami: https://www.linkedin.com/in/abeirami/ Connect with A. Kirimgeray Kirimli: https://www.linkedin.com/in/a-kirimgeray-kirimli-72bbb838/ ---- Flatiron Software: https://flatiron.software/ Snapshot AI: https://www.snapshot.reviews/

    24min
  6. Teaching AI to Play It Safe: Risk-Aware Reinforcement Learning | Executive Code EP #16

    26/09/2025

    Teaching AI to Play It Safe: Risk-Aware Reinforcement Learning | Executive Code EP #16

    In this episode of Executive Code, Kirimgeray talks with Arnob Ghosh (NJIT) about his work on making reinforcement learning safer. Most AI agents learn by chasing rewards—but what if those rewards come with hidden dangers? Arnob explains how adding a “risk budget” can help AI avoid rare but catastrophic mistakes, from crashing cars to bad investments. We explore how this research could shape safer robots, drones, and even language models. ------ Connect with Arnob Ghosh: LinkedIn: in/arnob-ghosh-9191591b/ Connect with A. Kirimgeray Kirimli: LinkedIn: in/a-kirimgeray-k ---- Flatiron Software: https://flatiron.software/ Snapshot AI: https://www.snapshot.reviews/

    33min
  7. Kurucuların Gözünden: Yapay Zekânın Bugünü ve Yarını | Executive Code EP #15

    19/09/2025

    Kurucuların Gözünden: Yapay Zekânın Bugünü ve Yarını | Executive Code EP #15

    Executive Code’da genelde yapay zekâ araştırmalarını konuşuyoruz. Ama bu sefer farklı: İki kurucu olarak — Sezer ve Kirim — yapay zekânın genel gidişatını tartıştık. - AI’ın farklı sektörlerdeki uygulama alanları - Şirketlerin hızla AI’ı benimsemeye çalışması - Cursor ve Claude Code gibi yeni kodlama modellerinin etkisi - Uzun context window’lar, donanım sınırlamaları ve memory mosaics gibi yaklaşımlar - Gelecek için fırsatlar ve soru işaretleri Bu, araştırma odaklı değil, daha geniş perspektiften bir sohbet. Hem kurucu gözüyle hem de sahadan gelen deneyimlerle. Üstelik ilk kez Türkçe. ------ Kırım’la bağlantıda kalın: LinkedIn: / a-kirimgeray-k Sezer’le bağlantıda kalın: LinkedIn: /ovuncsezer/ ---- Flatiron Software: https://flatiron.software/ Snapshot AI: https://www.snapshot.reviews/

    26min
  8. From Star Trek to Reality: Capturing Photons in Motion | EP #14 Executive Code

    05/09/2025

    From Star Trek to Reality: Capturing Photons in Motion | EP #14 Executive Code

    Did you know that with LiDAR cameras, even walls can act like mirrors? In our first studio-recorded episode, we sit down with Anagh Malik (PhD candidate at University of Toronto, intern at Apple) to explore how high-speed cameras and lasers capture photons in motion—unlocking new ways to reconstruct 3D environments, analyze material properties, and even see around corners. ------ Connect with Anagh Malik: LinkedIn: linkedin.com/in/anaghmalik/ Website: https://anaghmalik.com/ Connect with A. Kirimgeray Kirimli: LinkedIn: linkedin.com/in/a-kirimgeray-kirimli-72bbb838/ ---- Flatiron Software: https://flatiron.software/ Snapshot AI: https://www.snapshot.reviews/

    39min
Ver tudo (21)

Sobre

Welcome to Executive Code — where AI, engineering, and leadership collide. Hosted by the minds behind Snapshot AI and Flatiron Software, this podcast brings you inside conversations with the people shaping the future of technology. From deep dives into the latest AI research to unfiltered takes on engineering leadership, each episode is a front-row seat to the real talk happening behind the scenes in the tech world. Whether you're a CTO, a machine learning engineer, or just someone curious about where the future is heading — this is your space. Filmed across New York, London, and wherever the conversation takes us. Subscribe and join us every other week.

Informações

  • Criado por
    Kirimgeray Kirimli & Rajiv Pant
  • Anos de atividade
    2 mil
  • Episódios
    21