StarZero All shows Try the podcast clipper →

Sep 28, 2017 · 26m · mad

AI and a Different Type of Competitive Advantage // Omar Tawakol, Voicera (FirstMark's Data Driven)

Omar Tawakol · 21m spoken Matt Turck · 40s spoken Recorded Audio Sample · 2s 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

In this FirstMark Data Driven presentation, Voicera CEO Omar Tawakol presents how AI exoskeletons enhance workplace productivity, demonstrating Voicera's meeting assistant Eva while detailing technical pipelines and compounding competitive advantages unique to enterprise AI platforms.

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 2.9% of the talking time here. How this is scored →

Matt as informed peer 0.8 Guest teaching 0.8 Guest disagreement 0.6 Matt pushing back 0.4
05100:0010:0020:000:26–3:58 · Matt as informed peer 0/10 Agenda and AI Exoskeletons vs. Robots Omar gives a monologue presentation contrasting job-displacing AI robots with exoskeleton AI tools like Grammarly and Voicera. The host does not speak during this presentation segment.3:58–8:34 · Matt as informed peer 0/10 Enterprise Voice AI and the Problem with Meetings Omar demonstrates Eva, Voicera's live enterprise meeting assistant, detailing real-time transcription and action item tracking. The host is not participating beyond brief live audio feedback from the assistant.8:34–10:38 · Matt as informed peer 0/10 Building Competitive Advantage: Data Networks vs. AI Compounding Effect Omar explains how AI competitive advantage differs from traditional data network effects by building a compounding algorithm cycle. This is entirely a monologue keynote presentation with no host involvement.10:38–14:21 · Matt as informed peer 0/10 Building the Data Pipeline and Model Automation Omar discusses building automated data pipelines, addressing false positive/negative labeling, and explaining survivor bias using Abraham Wald's airplane analysis. No host interaction takes place.14:21–26:17 · Matt as informed peer 4/10 Refining Scope to Customer Meetings and Presentation Wrap-Up Matt Turck opens the Q&A session asking insightful questions about vertical customization and team composition in early-stage AI startups. Omar explains Voicera's ensemble architecture and human-in-the-loop validation process.0:26–3:58 · Guest teaching 0/10 Agenda and AI Exoskeletons vs. Robots Omar gives a monologue presentation contrasting job-displacing AI robots with exoskeleton AI tools like Grammarly and Voicera. The host does not speak during this presentation segment.3:58–8:34 · Guest teaching 0/10 Enterprise Voice AI and the Problem with Meetings Omar demonstrates Eva, Voicera's live enterprise meeting assistant, detailing real-time transcription and action item tracking. The host is not participating beyond brief live audio feedback from the assistant.8:34–10:38 · Guest teaching 0/10 Building Competitive Advantage: Data Networks vs. AI Compounding Effect Omar explains how AI competitive advantage differs from traditional data network effects by building a compounding algorithm cycle. This is entirely a monologue keynote presentation with no host involvement.10:38–14:21 · Guest teaching 0/10 Building the Data Pipeline and Model Automation Omar discusses building automated data pipelines, addressing false positive/negative labeling, and explaining survivor bias using Abraham Wald's airplane analysis. No host interaction takes place.14:21–26:17 · Guest teaching 4/10 Refining Scope to Customer Meetings and Presentation Wrap-Up Matt Turck opens the Q&A session asking insightful questions about vertical customization and team composition in early-stage AI startups. Omar explains Voicera's ensemble architecture and human-in-the-loop validation process.0:26–3:58 · Guest disagreement 2/10 Agenda and AI Exoskeletons vs. Robots Omar gives a monologue presentation contrasting job-displacing AI robots with exoskeleton AI tools like Grammarly and Voicera. The host does not speak during this presentation segment.3:58–8:34 · Guest disagreement 0/10 Enterprise Voice AI and the Problem with Meetings Omar demonstrates Eva, Voicera's live enterprise meeting assistant, detailing real-time transcription and action item tracking. The host is not participating beyond brief live audio feedback from the assistant.8:34–10:38 · Guest disagreement 0/10 Building Competitive Advantage: Data Networks vs. AI Compounding Effect Omar explains how AI competitive advantage differs from traditional data network effects by building a compounding algorithm cycle. This is entirely a monologue keynote presentation with no host involvement.10:38–14:21 · Guest disagreement 0/10 Building the Data Pipeline and Model Automation Omar discusses building automated data pipelines, addressing false positive/negative labeling, and explaining survivor bias using Abraham Wald's airplane analysis. No host interaction takes place.14:21–26:17 · Guest disagreement 1/10 Refining Scope to Customer Meetings and Presentation Wrap-Up Matt Turck opens the Q&A session asking insightful questions about vertical customization and team composition in early-stage AI startups. Omar explains Voicera's ensemble architecture and human-in-the-loop validation process.0:26–3:58 · Matt pushing back 0/10 Agenda and AI Exoskeletons vs. Robots Omar gives a monologue presentation contrasting job-displacing AI robots with exoskeleton AI tools like Grammarly and Voicera. The host does not speak during this presentation segment.3:58–8:34 · Matt pushing back 0/10 Enterprise Voice AI and the Problem with Meetings Omar demonstrates Eva, Voicera's live enterprise meeting assistant, detailing real-time transcription and action item tracking. The host is not participating beyond brief live audio feedback from the assistant.8:34–10:38 · Matt pushing back 0/10 Building Competitive Advantage: Data Networks vs. AI Compounding Effect Omar explains how AI competitive advantage differs from traditional data network effects by building a compounding algorithm cycle. This is entirely a monologue keynote presentation with no host involvement.10:38–14:21 · Matt pushing back 0/10 Building the Data Pipeline and Model Automation Omar discusses building automated data pipelines, addressing false positive/negative labeling, and explaining survivor bias using Abraham Wald's airplane analysis. No host interaction takes place.14:21–26:17 · Matt pushing back 2/10 Refining Scope to Customer Meetings and Presentation Wrap-Up Matt Turck opens the Q&A session asking insightful questions about vertical customization and team composition in early-stage AI startups. Omar explains Voicera's ensemble architecture and human-in-the-loop validation process.

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 19.8% · guest 80.2%15:00 · Matt 19.8% · guest 80.2%18:00 · Matt 0.7% · guest 99.3%18:00 · Matt 0.7% · guest 99.3%21:00 · Matt 2.8% · guest 97.2%21:00 · Matt 2.8% · guest 97.2%24:00 · Matt 4% · guest 96%24:00 · Matt 4% · guest 96%
Sharpest disagreement ▶ 1:15 Critique of Job Displacing Robots

Omar forcefully argues against AI applications that displace human jobs, positioning his vision of exoskeleton AI as superior to destructive automation.

 Hardest push from Matt ▶ 15:06 Host Question on Vertical Slicing

Matt Turck pushes Omar on whether Voicera needs specific domain models for industries like finance or healthcare versus general enterprise speech models.

 Biggest teaching moment ▶ 22:38 Explaining Human-in-the-Loop Audio Correction

Omar details why a 1% human-in-the-loop fall-back mechanism is necessary for enterprise AI accuracy where standard consumer assistant error rates are unacceptable.

 Matt holds his own ▶ 15:06 Host Demonstrates Industry Domain Insight

Matt Turck shows domain knowledge by specifically questioning how enterprise speech models handle vertical-specific terminology and language customization.
the scores for every segment, with the reasoning behind each
ChapterTopicMatt as informed peerGuest teachingGuest disagreementMatt pushing backWhy
Agenda and AI Exoskeletons vs. Robots 0020 Omar gives a monologue presentation contrasting job-displacing AI robots with exoskeleton AI tools like Grammarly and Voicera. The host does not speak during this presentation segment.
Enterprise Voice AI and the Problem with Meetings 0000 Omar demonstrates Eva, Voicera's live enterprise meeting assistant, detailing real-time transcription and action item tracking. The host is not participating beyond brief live audio feedback from the assistant.
Building Competitive Advantage: Data Networks vs. AI Compounding Effect 0000 Omar explains how AI competitive advantage differs from traditional data network effects by building a compounding algorithm cycle. This is entirely a monologue keynote presentation with no host involvement.
Building the Data Pipeline and Model Automation 0000 Omar discusses building automated data pipelines, addressing false positive/negative labeling, and explaining survivor bias using Abraham Wald's airplane analysis. No host interaction takes place.
Refining Scope to Customer Meetings and Presentation Wrap-Up 4412 Matt Turck opens the Q&A session asking insightful questions about vertical customization and team composition in early-stage AI startups. Omar explains Voicera's ensemble architecture and human-in-the-loop validation process.

Statements from this episode (10)

Insight
Tawakol: AI transcription needs human-in-the-loop systems for acceptable user experiences
“What they don't realize, it's pretty hard for a human to understand the last few percentage points of accuracy, how that impacts the experience, and typically in many areas, like in transcription and other areas, it's a pretty bad impact if you're not, you kno…”
Omar Tawakol Sep 28, 2017 ▶ 3:37
Assertion Not checkable as stated
Tawakol: Over 200 companies are building meeting infrastructure software
“There's like, 220 companies doing that.”
Omar Tawakol Sep 28, 2017 ▶ 4:26
Assertion Not checkable as stated
Tawakol: Multitasking during meetings drops your cognitive performance by 20 IQ points
“You are literally 20 IQ points dumber when you are multitasking, and everybody's multitasking, and that's the problem.”
Omar Tawakol Sep 28, 2017 ▶ 4:37
Assertion Not publicly verifiable
Tawakol: Oracle Data Cloud grew into a $500 million business
“By the time, you know, we went into the Oracle data cloud and continued to grow the business and make acquisitions, it was a five hundred million dollar business.”
Omar Tawakol Sep 28, 2017 ▶ 9:13
Insight
Tawakol: AI data flywheels take years to build and create strong moats
“So you start to get much more nuanced and better algorithm results, which lets you extend the surface area of what you're doing way beyond your competitors, which then gives a better experience, and you get a virtuous cycle, and it takes years to develop these…”
Omar Tawakol Sep 28, 2017 ▶ 10:10
Disclosure
Tawakol: Voicera bootstrapped its voice AI using phone transcripts and Mechanical Turk
“Buying available data assets in terms of You know, voice that's out there, voice over the phone, transcribed voice over the phone, which we did, and then we went out and we bought labels through things like Mechanical Turk.”
Omar Tawakol Sep 28, 2017 ▶ 10:50
Assertion Not checkable as stated
Tawakol: Most AI companies forget to label false negatives in their data
“False negatives are harder And most companies will do solutions like this and forget to figure out how to label the false negatives.”
Omar Tawakol Sep 28, 2017 ▶ 11:47
Insight
Tawakol: AI development teams require statistics experts, not just software engineers
“So these kinds of judgment errors are just very commonplace, and that's why you obviously need people on the team who aren't just You know, ah, really good at writing the code, and you also knew people very familiar with statistics.”
Omar Tawakol Sep 28, 2017 ▶ 14:03
Assertion Not checkable as stated
Tawakol: Voicera uses human error correction for 1% of processed audio
“We end up doing the error correct about one percentage of the audio time.”
Omar Tawakol Sep 28, 2017 ▶ 22:39
Insight
Tawakol: Enterprise AI demands significantly higher accuracy tolerances than consumer AI
“If you tell Alexa to play a song and she plays the wrong one, no problem. You tell Eva to update Salesforce and automatically update Salesforce record, and it's the wrong record, there's a problem.”
Omar Tawakol Sep 28, 2017 ▶ 23:11
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