StarZero All shows Try the podcast clipper →

May 18, 2018 · 25m · mad

Make AI Less Mysterious // Serkan Piantino, Spell (FirstMark's Data Driven)

Serkan Piantino · 21m spoken Matt Turck · 47s spoken
0:00 / 0:00
▶ Watch on YouTube →

gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions

At Data Driven NYC, Spell Founder and CEO Serkan Piantino presents his platform designed to democratize machine learning by streamlining remote GPU execution, ensuring scientific experiment reproducibility, and enabling automated AI workflows.

How this conversation actually went

Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Matt holds 3.5% of the talking time here. How this is scored →

Matt as informed peer 0.7 Guest teaching 0.5 Guest disagreement 0.2 Matt pushing back 0.5
05100:0010:0020:000:09–3:55 · Matt as informed peer 0/10 Serkan Piantino Background and the Mission of Spell Serkan delivers an introductory monologue outlining his background at Facebook AI Research and the mission of Spell. The host does not speak during this monologue segment.3:55–7:07 · Matt as informed peer 0/10 Defining New Capabilities and the Machine Learning Expertise Bottleneck Serkan presents the core drivers of new AI capabilities and identifies human machine learning expertise as the primary bottleneck. The host is inactive during this presentation segment.7:07–11:20 · Matt as informed peer 0/10 Spell Product Demonstration: Simple Remote GPU Execution Serkan demonstrates Spell's command-line interface for running remote GPU jobs seamlessly. The segment consists entirely of a product video demonstration without host interaction.11:20–15:30 · Matt as informed peer 0/10 Scientific Rigor, Empirical Engineering, and Collaboration Tools Serkan details how Spell brings scientific rigor and reproducible experiment tracking to empirical deep learning. The host does not participate in this monologue segment.15:30–19:12 · Matt as informed peer 0/10 Meta-Learning, Automated Machine Learning, and Beta Announcement Serkan discusses meta-learning, automated hyperparameter optimization, and opens the beta signup. This segment is a continuous monologue with no host involvement.19:12–25:34 · Matt as informed peer 4/10 Fireside Q&A on Infrastructure, Datasets, and Roadmap Matt opens the fireside Q&A, asking targeted questions about cloud hardware and challenging initial impressions of Spell's target market. Serkan clarifies product positioning while maintaining an agreeable tone.0:09–3:55 · Guest teaching 0/10 Serkan Piantino Background and the Mission of Spell Serkan delivers an introductory monologue outlining his background at Facebook AI Research and the mission of Spell. The host does not speak during this monologue segment.3:55–7:07 · Guest teaching 0/10 Defining New Capabilities and the Machine Learning Expertise Bottleneck Serkan presents the core drivers of new AI capabilities and identifies human machine learning expertise as the primary bottleneck. The host is inactive during this presentation segment.7:07–11:20 · Guest teaching 0/10 Spell Product Demonstration: Simple Remote GPU Execution Serkan demonstrates Spell's command-line interface for running remote GPU jobs seamlessly. The segment consists entirely of a product video demonstration without host interaction.11:20–15:30 · Guest teaching 0/10 Scientific Rigor, Empirical Engineering, and Collaboration Tools Serkan details how Spell brings scientific rigor and reproducible experiment tracking to empirical deep learning. The host does not participate in this monologue segment.15:30–19:12 · Guest teaching 0/10 Meta-Learning, Automated Machine Learning, and Beta Announcement Serkan discusses meta-learning, automated hyperparameter optimization, and opens the beta signup. This segment is a continuous monologue with no host involvement.19:12–25:34 · Guest teaching 3/10 Fireside Q&A on Infrastructure, Datasets, and Roadmap Matt opens the fireside Q&A, asking targeted questions about cloud hardware and challenging initial impressions of Spell's target market. Serkan clarifies product positioning while maintaining an agreeable tone.0:09–3:55 · Guest disagreement 0/10 Serkan Piantino Background and the Mission of Spell Serkan delivers an introductory monologue outlining his background at Facebook AI Research and the mission of Spell. The host does not speak during this monologue segment.3:55–7:07 · Guest disagreement 0/10 Defining New Capabilities and the Machine Learning Expertise Bottleneck Serkan presents the core drivers of new AI capabilities and identifies human machine learning expertise as the primary bottleneck. The host is inactive during this presentation segment.7:07–11:20 · Guest disagreement 0/10 Spell Product Demonstration: Simple Remote GPU Execution Serkan demonstrates Spell's command-line interface for running remote GPU jobs seamlessly. The segment consists entirely of a product video demonstration without host interaction.11:20–15:30 · Guest disagreement 0/10 Scientific Rigor, Empirical Engineering, and Collaboration Tools Serkan details how Spell brings scientific rigor and reproducible experiment tracking to empirical deep learning. The host does not participate in this monologue segment.15:30–19:12 · Guest disagreement 0/10 Meta-Learning, Automated Machine Learning, and Beta Announcement Serkan discusses meta-learning, automated hyperparameter optimization, and opens the beta signup. This segment is a continuous monologue with no host involvement.19:12–25:34 · Guest disagreement 1/10 Fireside Q&A on Infrastructure, Datasets, and Roadmap Matt opens the fireside Q&A, asking targeted questions about cloud hardware and challenging initial impressions of Spell's target market. Serkan clarifies product positioning while maintaining an agreeable tone.0:09–3:55 · Matt pushing back 0/10 Serkan Piantino Background and the Mission of Spell Serkan delivers an introductory monologue outlining his background at Facebook AI Research and the mission of Spell. The host does not speak during this monologue segment.3:55–7:07 · Matt pushing back 0/10 Defining New Capabilities and the Machine Learning Expertise Bottleneck Serkan presents the core drivers of new AI capabilities and identifies human machine learning expertise as the primary bottleneck. The host is inactive during this presentation segment.7:07–11:20 · Matt pushing back 0/10 Spell Product Demonstration: Simple Remote GPU Execution Serkan demonstrates Spell's command-line interface for running remote GPU jobs seamlessly. The segment consists entirely of a product video demonstration without host interaction.11:20–15:30 · Matt pushing back 0/10 Scientific Rigor, Empirical Engineering, and Collaboration Tools Serkan details how Spell brings scientific rigor and reproducible experiment tracking to empirical deep learning. The host does not participate in this monologue segment.15:30–19:12 · Matt pushing back 0/10 Meta-Learning, Automated Machine Learning, and Beta Announcement Serkan discusses meta-learning, automated hyperparameter optimization, and opens the beta signup. This segment is a continuous monologue with no host involvement.19:12–25:34 · Matt pushing back 3/10 Fireside Q&A on Infrastructure, Datasets, and Roadmap Matt opens the fireside Q&A, asking targeted questions about cloud hardware and challenging initial impressions of Spell's target market. Serkan clarifies product positioning while maintaining an agreeable tone.

speaking balance: gold is Matt, purple is the guest (3 minute bins)

0:00 · Matt 0% · guest 100%0:00 · Matt 0% · guest 100%3:00 · Matt 0% · guest 100%3:00 · Matt 0% · guest 100%6:00 · Matt 0% · guest 100%6:00 · Matt 0% · guest 100%9:00 · Matt 0% · guest 100%9:00 · Matt 0% · guest 100%12:00 · Matt 0% · guest 100%12:00 · Matt 0% · guest 100%15:00 · Matt 0% · guest 100%15:00 · Matt 0% · guest 100%18:00 · Matt 19.2% · guest 80.8%18:00 · Matt 19.2% · guest 80.8%21:00 · Matt 9.2% · guest 90.8%21:00 · Matt 9.2% · guest 90.8%24:00 · Matt 5.4% · guest 94.6%24:00 · Matt 5.4% · guest 94.6%
Sharpest disagreement ▶ 21:21 Reframing Target Audience Assumption

Serkan gently pushes back on Matt's assumption that Spell is primarily built for high-end quants, explaining broader use cases across non-traditional industries.

 Hardest push from Matt ▶ 21:05 Host Challenges Target Audience Scope

Matt questions the product positioning, pointing out his original impression that Spell was aimed strictly at supercomputer-level quantitative researchers.

 Biggest teaching moment ▶ 19:28 Explaining Custom Hardware and TPU Integration

Serkan explains to Matt how Spell abstracts underlying infrastructure across proprietary hardware, AWS instances, and Google TPUs.

 Matt holds his own ▶ 19:49 Host Demonstrates Knowledge of TPU Architecture

Matt displays specific industry expertise by proactively asking whether the platform integrates with Google TPUs as they hit the market.
the scores for every segment, with the reasoning behind each
ChapterTopicMatt as informed peerGuest teachingGuest disagreementMatt pushing backWhy
Serkan Piantino Background and the Mission of Spell 0000 Serkan delivers an introductory monologue outlining his background at Facebook AI Research and the mission of Spell. The host does not speak during this monologue segment.
Defining New Capabilities and the Machine Learning Expertise Bottleneck 0000 Serkan presents the core drivers of new AI capabilities and identifies human machine learning expertise as the primary bottleneck. The host is inactive during this presentation segment.
Spell Product Demonstration: Simple Remote GPU Execution 0000 Serkan demonstrates Spell's command-line interface for running remote GPU jobs seamlessly. The segment consists entirely of a product video demonstration without host interaction.
Scientific Rigor, Empirical Engineering, and Collaboration Tools 0000 Serkan details how Spell brings scientific rigor and reproducible experiment tracking to empirical deep learning. The host does not participate in this monologue segment.
Meta-Learning, Automated Machine Learning, and Beta Announcement 0000 Serkan discusses meta-learning, automated hyperparameter optimization, and opens the beta signup. This segment is a continuous monologue with no host involvement.
Fireside Q&A on Infrastructure, Datasets, and Roadmap 4313 Matt opens the fireside Q&A, asking targeted questions about cloud hardware and challenging initial impressions of Spell's target market. Serkan clarifies product positioning while maintaining an agreeable tone.

Statements from this episode (7)

Insight
Piantino: Big tech dominates AI research due to superior compute resources
“Increasingly at a lot of the conferences and looking at the papers that were being developed, it was more and more just big tech companies like ourselves that were showing up in the literature, and it was more and more a field becoming dominated by big players…”
Serkan Piantino May 18, 2018 ▶ 2:05
Opinion
Piantino: The term 'AI' has become overloaded and pop-culture-ish
“I'm not a big fan of the word AI. It's kind of, has become overloaded. It's very pop culture-ish. It means different things in different contexts.”
Serkan Piantino May 18, 2018 ▶ 3:59
Insight
Piantino: Expertise is the primary bottleneck restricting machine learning progress
“And my question is, what is the bottleneck? Which of these ingredients is in shortest supply? And I think it's the expertise.”
Serkan Piantino May 18, 2018 ▶ 5:26
Assertion Not checkable as stated
Piantino: Startups building deep learning products struggle with infrastructure
“If you have startups in your portfolio, I know we have at least one VC in the audience, so and they're trying to do something in deep learning. They're trying to build a new product with some of this technology. I think you'll find that they do struggle with i…”
Serkan Piantino May 18, 2018 ▶ 10:36
Prediction Not checkable as stated
Piantino: Great future products will be built using machine learning workflows
“I think a lot of the great products in the future are going to be built this way, and consequently, I think this style of software engineering is becoming mainstream and becoming this sort of differentiated tool set for software engineers and machine learning …”
Serkan Piantino May 18, 2018 ▶ 14:53
Assertion Supported
Piantino: AI image identification on ImageNet has surpassed human accuracy
“We estimate humans can do this with about five percent error, so we're already better at identifying things and objects than humans.”
Serkan Piantino May 18, 2018 ▶ 17:40
Disclosure
Piantino: Spell built custom deep learning hardware alongside public cloud support
“We have our own hardware, which we built custom purpose for doing deep learning. But we also can use resources on any of the public clouds.”
Serkan Piantino May 18, 2018 ▶ 19:29
Made with StarZero

Turn any episode into a week of clips.

This entire site, over 400 conversations transcribed, diarized, checked and made playable, runs on the StarZero media pipeline. Drop in your own episode and the podcast clipper finds the moments worth sharing, cuts them, captions them, and reframes them for every feed.

Try the podcast clipper Learn more