AI Innovators - By SaladCloud

Ep 6 - AI adoption & procurement - Henry Stanley - Fabrik

In this episode, Henry Stanley, Chief Product Officer and Co-Founder of Fabrik, discusses the challenges of AI adoption and procurement in the B2B trust ecosystem.
This podcast is brought to you by Salad, the world's largest distributed cloud.
Learn more: https://salad.com/
Try SaladCloud today: https://portal.salad.com
Takeaways
- AI adoption and procurement in the B2B trust ecosystem present challenges related to security, compliance, and trust.
- Compliance is a risk function that helps organizations manage risk and build trust with customers. AI introduces new risks and frameworks for governance and compliance.
- Tools and solutions in the compliance and security ecosystem, such as Vanta and Creo AI, are emerging to address AI-specific governance and controls.
- Fabrik aims to help companies proactively demonstrate their security and compliance posture and build trust with customers.
Chapters
00:00: Challenges of AI Adoption and Procurement
03:24: The Role of Compliance in Risk Management
06:14: New Risks and Frameworks for AI Governance
08:05: Emerging Tools and Solutions in Compliance and Security
10:59: Proactively Demonstrating Security and Compliance

Ep 5 - AI for easier tax filing - Daniel Marcous - April

In this episode, we interview Daniel, a CTO who solved two of the most frustrating problems for everyday people - taxes & traffic.
Daniel Marcous is the CTO of April (AI tax-prep software) and ex-CTO of Waze (popular crowdsourced traffic app). This podcast is brought to you by Salad, the world's largest distributed cloud.
Learn more: https://salad.com/
Try SaladCloud today: https://portal.salad.com
Sound bites:
"So we don't only meet you once a year in April when you need to file your taxes where it's far too late to actually do something impactful, but we accompany you throughout the year in order to save you the big bucks and actually optimize your taxes"
"Taxes need to be 100 % accurate. You can't make mistake one every 100 users. Just not doable. So there's tons of guardrails that we've put basically everywhere so it will be accurate "
"This role called data scientist is going to slowly shrink into becoming something probably not going to remain in existence for longer the way we know it, as the power is being shifted to engineers really, really fast in everything that has to do in AI"
Takeaways
April is an AI-powered tax preparation company that offers tax solutions for financial institutions and individuals.
The use of AI in tax optimization can help individuals save time and money by automating tax planning, tax optimization, and tax filing. Bridging the gap between tax knowledge and technical expertise is crucial in building AI-powered tax software.
The role of data scientists is evolving, and engineers are becoming more involved in AI development.
Cloud costs can be managed by understanding the trade-off between business outcomes and costs and optimizing financial operations.
Choosing an alternate cloud provider should be based on the specific needs of the business and the ability to solve product requirements.
The entry barriers for entering tech are lower, and formal education is becoming less important in the industry.
Daniel's favorite tech gadget is a tag locator, and his favorite way to unwind is by watching anime TV shows.

Ep 4: AI tools, cost & examples of video accessibility done right - Michelle Inaba

SUMMARY
Michelle Inaba, a product manager, discusses the importance of making video content globally accessible. 
This includes providing options such as automatic captions, subtitles, and dubbing in multiple languages. Companies are using strategies like automatic speech recognition and dubbing to make their video content accessible.
The future of video accessibility could include AI real-time dubbing and improved automatic captions. While AI is efficient and cost-effective, human involvement is still essential for tasks that require nuance and accuracy. It is best to use a hybrid approach, combining AI solutions with human verification.
The cost of video accessibility with AI varies depending on factors like content length and complexity. However, the upfront expenses are worth it for the increased audience reach and engagement. Video accessibility positively impacts business metrics like revenue and brand recognition by expanding global reach and attracting diverse audiences. 
There are several AI tools and services available for video accessibility, including Papercup AI, Dubbing AI, Rev.ai, and Google Cloud Translation.
Michelle also provides many real-life examples including Mr. Beast on how accessibility can increase revenue for creators and businesses alike. 
Takeaways
Video accessibility means making content globally accessible through options like automatic captions, subtitles, and dubbing in multiple languages.
Companies use strategies like automatic speech recognition and dubbing to make their video content accessible.
The future of video accessibility could include AI real-time dubbing and improved automatic captions.
While AI is efficient and cost-effective, human involvement is essential for tasks that require nuance and accuracy.
A hybrid approach, combining AI solutions with human verification, is recommended for video accessibility.
Video accessibility positively impacts business metrics like revenue and brand recognition by expanding global reach and attracting diverse audiences.
AI tools and services like Papercuts AI, Dublin AI, Rev.ai, and Google Cloud Translation are available for video accessibility.
Sound Bites
"A good example here is, in my opinion is Mr. Beast. He has his channel in several different languages. And he does that by using dubbing. And by doing that, he is making his content available to so many different people from so many different countries, so many different languages"
 
"In the case of a streamer or any other individual that's like a video content creator, YouTuber or whatever, expanding your markets internationally allows you to have more interest, to be more interested in product placements in your videos. And you become more attractive for brands partnerships"
"In the future, we could expect to see some AI real-time dubbing and an improvement on the automatic captions."
Chapters
00:00 Introduction to Video Accessibility01:35 Strategies for Making Video Content Accessible02:02 The Future of Video Accessibility03:30 The Role of Humans in Video Accessibility07:20 The Cost of Video Accessibility with AI15:49 The Business Impact of Video Accessibility18:49 AI Tools for Video Accessibility
 
 
 
 

Ep 3: AI applications in the enterprise - Chip Ernst from Roli.ai

Chip Ernst, CEO of Roli.ai, discusses the challenges of integrating AI into enterprise applications and the need for easy and robust cloud engineering solutions.
He emphasizes the importance of considering the complexity of enterprise environments and the need for validation and control when using AI models.
Chip provides examples of use cases, such as AI-supported responses for automotive service organizations and doctor case note expansion, where human validation is crucial. He also highlights the need for flexibility and adaptability in AI solutions as technologies and models evolve.
Takeaways
Integrating AI into enterprise applications requires robust cloud engineering solutions.
Validation and control are essential when using AI models, and human involvement is necessary to ensure accuracy and accountability.
Use cases for AI in enterprise include AI-supported responses for automotive service organizations and doctor case note expansion.
Flexibility and adaptability are crucial in AI solutions as technologies and models evolve.
Sound Bites
"Can you do the same old thing smarter with AI?"
"AI technologies are easily accessible, so accessible that every single one of us could sit down at our laptop and query with a prompt."
"The AI is new, but the idea of adding intelligence or some sort of clever process to a business application is not."
Learn more about Salad: https://salad.com
Try SaladCloud today: https://portal.salad.com
 

Ep 2: ASR models, accuracy, cost & the role of humans - Aleks Smechov from Wordcab

In this conversation, Derick Thompson from Salad Technologies interviews Alex from WordCab about transcription, ASR, and accessibility. They discuss the importance of accurate transcripts for global accessibility, the different definitions of verbatim transcription, and the impact of audio cues. They also talk about the best ASR models, tools for post-processing, and the need for human editors in transcription. The conversation concludes with a discussion on the future of ASR and transcription.
 
Takeaways
Accurate transcripts are crucial for global accessibility, allowing people with disabilities to understand audio and video content.
Different definitions of verbatim transcription exist, ranging from including all disfluencies to a more cleaned-up version.
Audio cues, such as laughter or coughing, are important for accessibility and may need to be added during transcription.
The best ASR models for transcription depend on the specific use case and language requirements.
Post-processing is essential for improving transcript accuracy, especially for industry-specific terms and difficult words.
Human editors play a vital role in fine-tuning transcripts and adding value through post-processing and audio cues.
The future of ASR and transcription lies in increasing accuracy, reducing word error rates, and focusing on post-processing capabilities.
Transcription will become a commodity, and the real value will come from what can be done with the transcript after transcription.
Using cost-effective GPU instances and cloud-agnostic tools is important for hosting ASR models.
The goal is to provide reliable and affordable transcription services to meet the needs of different use cases.
Sound Bites
"Accessibility in terms of video and audio, captions and transcription in general, is making sure that people who have some sort of disability, maybe they're hard of hearing or deaf, are still able to understand the captions or subtitles or transcript as well as someone who could hear."
"Transcript editing will always be there as a kind of a last mile thing for edge cases and there will always be edge cases."
"Transcription will become a commodity or table stakes like, you'll have to have excellent transcription, 95% accuracy, et cetera, in the future. And the real value will come in with what you could do after."
Chapters
00:00: Introduction and Overview of WordCab
01:14: Defining Verbatim Transcription and Audio Cues
07:03: Choosing the Best ASR Models for Transcription
09:26: The Importance of Post-Processing in Transcription
12:51: Accuracy, Word Error Rate, and Transcription
14:17: Tools and Approaches for ASR and Transcription
19:43: The Future of ASR and Transcription
21:08: Optimizing ASR Performance and Cost
22:07: Providing Reliable and Affordable Transcription Services
 

Ep 1: How AI dubbing is making videos globally accessible - Doniyor Ulmasov from PaperCut

Summary
In this conversation, Doniyor Ulmasov, Head of Engineering at PaperCut, discusses the process of making videos globally accessible through AI dubbing and localization.
He explains the differences between captions, subtitles, and dubs, and how dubbing involves adapting the source content to the target audience.
Doniyor also shares insights into the multi-step process of dubbing, including transcription, translation, and text-to-speech models.
He highlights the importance of human validation in maintaining quality and discusses the challenges of expanding beyond English.
The conversation concludes with a discussion on the cost-effectiveness of dubbing and the potential for PaperCut to become a global dubbing solution.
 
Takeaways
Video accessibility involves making videos globally accessible in multiple languages.
Dubbing is the process of adapting the source content to the target audience.
The dubbing process includes transcription, translation, and text-to-speech models.
Human validation is crucial for maintaining quality in dubbing.Expanding beyond English poses challenges in accuracy and pipeline management.
Dubbing can be a cost-effective solution compared to traditional dubbing houses.
PaperCut aspires to become a global dubbing solution for video accessibility.
Sound Bites
 
"A video is globally accessible when it can reach as many people as possible and as many languages as possible."
"If you do a literal translation, you're going to lose the joke, right? That's why it's called adaptation, not translation."
"Once we achieve the translation layer, then we move to the text-to-speech model."
 
Chapters
00:00 Introduction and Background
01:19 Caption, Subtitle, and Dubbing Differences
03:05 Text-to-Speech and Voice Assignment
05:03 Serverless GPU Options for Cost Optimization
08:18 Recommended Open Source Models
10:37 Challenges in Expanding Beyond English
11:06 Human Validation in Maintaining Quality
12:04 The Cost-Effectiveness of Dubbing
12:57 PaperCut's Aspiration as a Global Dubbing Solution