#23 Your Pressing AI Questions
Plus: State of AI Adoption, A Critical Look at Perplexity's $9bn Valuation, OpusClip, Notable AI Developments and a Humble Brag
Welcome back! In this issue, I cover:
Notable developments:
Apple’s AI Debut: A Two-Tier Approach to Intelligence
Genesis and NVIDIA Join Forces to Unlock New Medicines
Is AI Scaling Hitting Its Limits? OpenAI's Orion Sparks Debates
Salesforce Hiring 1,000 Salespeople to Promote Agentforce
AI Health Startup Forward Shuts Down: A Reality Check
Directors’ Corner: State of AI Adoption
From the Trenches: Your Pressing AI Questions
Perplexity's $9bn Valuation: A Critical Look
AI Tool Spotlight: OpusClip
One More Thing: A Humble Brag
Enjoy.
Notable Developments
1. Apple’s AI Debut: A Two-Tier Approach to Intelligence
Apple has finally entered the AI race with a unique strategy that reflects their privacy-first mindset. iOS 18.1 introduces a hybrid system that uses two types of language models: a smaller ~3B parameter model runs directly on your device for quick, everyday tasks (think simple rewrites and summaries), while more complex operations tap into Apple's larger cloud-based model. Think of it as having both a quick-thinking assistant for immediate needs and a deeper analytical engine for complex tasks.
While Apple's on-device AI might not match the creative prowess of the larger models like ChatGPT, it offers something unique: your data stays on your device instead of being processed in the cloud. This commitment to privacy remains a clear differentiator.
2. Genesis and NVIDIA Join Forces to Unlock New Medicines
In a significant move for AI-powered drug discovery, Genesis Therapeutics received a new investment from NVIDIA's venture arm and announced a strategic collaboration. While the investment amount wasn't disclosed, it follows NVentures' participation in Genesis' $200M Series B last August. The partnership will enhance Genesis' AI platform (GEMS) to better understand and design potential new drugs in 3D. This collaboration aims to discover treatments for diseases that have been resistant to traditional drug development methods. It's another example of how AI companies are teaming up with tech giants to tackle complex healthcare challenges.
3. Is AI Scaling Hitting Its Limits? OpenAI's Orion Sparks Debates
OpenAI's next-generation model, Orion, is reportedly showing more modest improvements over GPT-4 than previous model updates, challenging the industry's assumption that bigger models automatically mean better performance. Two key factors might be driving this change: we're running out of high-quality training data (projected to exhaust public text by 2032), and the computational costs of running these models are becoming unsustainable.
This has led companies to rethink their approach. OpenAI is now developing specialized models for different tasks, rather than pursuing a one-size-fits-all solution. This shift suggests that the path to artificial general intelligence (AGI) may require fundamental breakthroughs in how AI learns and reasons, not just larger models with more data.
4. Salesforce Hiring 1,000 Salespeople to Promote Agentforce
In a twist of irony, Salesforce is hiring 1,000 salespeople to promote Agentforce, its autonomous customer service and sales development tool that reduces the need for human sales roles. Already adopted by Saks and OpenTable, this massive investment signals Salesforce's ambition and conviction to lead the AI agent market before competitors like Microsoft and ServiceNow catch up.
5. AI Health Startup Forward Shuts Down: A Reality Check
Forward, a primary care startup that raised $400M including a $100M Series E in 2023, has ceased operations. The company's AI-powered CarePods - self-serve diagnostic kiosks deployed in malls, gyms, and offices - represents a common pitfall in healthcare innovation: focusing on flashy technology without addressing core healthcare challenges. Forward's story highlights that expensive AI hardware solutions need to demonstrate clear improvements in medical outcomes and quality of care to survive in today's market.
State of AI Adoption
As boards hold their final 2024 meetings and plan for 2025, I'm struck by a paradox: while some directors still view AI as peripheral to their business, recent data tells a different story. The pace of AI adoption has outstripped historical technology adoption curves, making it a critical strategic consideration. Therefore, it is timely to highlight some takeaways from the “Growing Up: Navigating Generative AI’s Early Years – AI Adoption Report” by The Wharton School and GBK Collective. The report is full of interesting takeaways, and I will only include a few here:
Weekly use of gen AI jumped to over 70% and mid-sized companies with revenue between $50M and $2B are leading the charges. If you still think AI is not relevant to your business, you are in the minority.
Not all business function leaders are seeing high impact yet. IT (code generation) and business intelligence are leading.
Companies typically start with training and experimentation internally, which I believe is the right approach. We have to start with business goals and work flows before introducing AI to amplify employees. Otherwise it is easy to fall into ‘a hammer looking for nails’ situation.
Some other interesting takeaways:
On one of the most asked questions by boards - who leads the AI strategy - Chief AI Officers are now in 21% of companies surveyed.
Top use cases are not always top performers — idea generation, legal contracts, and fraud detection perform well.
Many still do not have an AI use policy in place. This statistics might give nightmares to boards.
Many expect a slower increase in AI investments in year 2-5 vs the near-term. Muted demand for model training investments, falling costs, and time for ROI analysis are part of the reasons.
Accuracy concerns, privacy risks, and integration challenges are the biggest hurdles to adoption. However, concerns around cost and trust issues have decreased.
The message for boards is clear: If you're still questioning AI's relevance, you're falling behind your peers. As we enter 2025 - projected to be a year of widespread AI deployment - boards must shift from "whether" to "how" in their AI governance discussions.
From the Trenches: Your Pressing AI Questions
You know what they say about conference season in the business world - September and October are like wedding season! I've had the privilege of speaking at and attending numerous discussions around AI, finance, and governance lately. [Here is a list of my recent speakings if interested. ] I thought it might be fun to share some frequently asked questions along with my thoughts here.
"We gave everyone access to Microsoft Copilot (or other AI tools), but adoption is lower than expected. Why?"
This one comes up a lot! In my experience, it all starts with the 'why.' I've noticed that without a clear connection between AI tools and both business goals and individual professional success, people naturally hesitate. After all, who wants to embrace something they worry might eventually replace them?
"How do we unlock AI's full potential in our company's workflows?"
Here's a counterintuitive approach I love sharing: have your AI expert (internal or external) "watch your colleagues suffer" - in a constructive way, of course! What I mean is: observe your teammates going through their typical workflows, honestly assess their energy levels and value creation in the tasks, then collaborate with them to redesign these processes. Use "what-if" questions to imagine AI-amplified workflows. The insights from this exercise are often surprising and invaluable.
"What if we're a smaller company without massive amounts of data for AI?"
Don't let this hold you back! I always emphasize starting with your business model and unique competitive advantages. Choose battles you can win. Interestingly, by leveraging generic off-shelf large AI models with techniques to tap into your private data, you can still extract powerful insights from your 'small data.' In fact, there's a growing movement questioning the "bigger is better" data mentality - check out the thought-provoking piece "Big Data is Dead" for more on this.
"How can we trust AI when it still makes things up?"
This is a crucial question I hear everywhere. The key is understanding that predictive AI and generative AI each have their sweet spots. We're seeing exciting developments in improving accuracy, reliability, and transparency - from Retrieval Augmented Generation (RAG) to Knowledge Graphs (KG) and sophisticated prompting and AI agent techniques. Many applications now achieve accuracy rates from 80% to high-90%, which is often more than sufficient for most use cases.
"AI's energy consumption seems at odds with my company’s sustainability goals. How do we reconcile this?"
The energy consumption of AI is indeed significant, and yes, it's likely to increase in the near term. However, I'm encouraged by several promising trends: AI data centers being strategically placed in areas with abundant renewable energy, the rise of energy-efficient small language models and on-device processing, advances in power-efficient AI chips, and continued infrastructure improvements.
As always, these are complex challenges without simple answers. Peer-to-peer exchanges can help tremendously too. Please share your pressing questions in the comment or though LinkedIn with me.
Perplexity's $9bn Valuation: A Critical Look
Have you heard the latest about Perplexity, the AI search engine? They're reportedly about to raise a whopping $500 million at $9 billion. This would be Perplexity's fourth round of funding in just the past year – a pace that's turning heads even in the fast-moving world of AI startups. In 2024 alone, its valuation has moved up quickly:
January 2024: $520 million valuation
Summer 2024: $3 billion
Now: $9 billion (according to The Information)
Investors are clearly excited about Perplexity's potential to challenge Google in the search market. And the numbers do look impressive at first glance: 15 million daily queries and $50 million in annualized revenue from Pro subscriptions.
Even as an early fan and current Pro user of Perplexity, I'm struggling to make sense of its valuation and the non-stop fund raise activities.
The Capital Need Question
Here's what makes me pause: Perplexity doesn't actually train its own AI models. Instead, it's more like a really search engine with third party access to existing models like ChatGPT or Claude. So why the massive capital needs so frequently? Is all this investment really building long-term core competence, or is it more about ‘raise while you can’ before competitors catch up on product development?
The Moat Challenge
Don't get me wrong – Perplexity is a great product with many passionate users. I wrote to them a few times with my feedback on features like Pages (now Spaces) and domain specific search workflows. They've created an elegant search experience with fast response, logical answers, inline source quotations and smart follow-up questions. But here's the thing: these features, while impressive, aren't exactly impossible to replicate. We're already seeing similar elements pop up in Claude, Google NotebookLM, ChatGPT Search, Google Search AI Overview, Apple Intelligence, and others.
In addition, many enterprise search and AI agents are designed to deliver more accurate and reliable answers to user questions. They are competitors as well.
User Acquisition Concerns
Their recent user acquisition strategy has raised some eyebrows – mine included. I've watched them offer a full year of Pro service ($20/month value) for free to college students. These users tend to be early adopters with characteristically fluid loyalty to tech products especially when they get them for free. This reminds me of tactics I saw during the consumer internet boom, where companies used similar promotions to pump up membership and usage numbers before an upcoming round of fundraising.
If Perplexity is burning cash primarily on user acquisition and 3rd party AI model inference costs, that's a significant shift from their early organic growth.
Unproven Business Models
Looking ahead, there are still some big question marks:
Copyrights and Advertising: They're planning to share ad revenue with content providers as a strategy against media lawsuits. Will revenue sharing be sufficient for major publishers like The New York Times? And will advertisement affect the user experience and therefore adoption?
Real Traction of Enterprise Solutions: Their enterprise product is just getting started. The big questions around privacy, security, and compliance in enterprise workflows are still unanswered.
Time will tell if Perplexity can live up to its sky-high valuation, but these rapid funding rounds and aggressive user acquisition tactics suggest they might be racing against the clock. As someone who wants to see them succeed, I hope they're building something sustainable to grow into the hype-like valuation.
AI Tool Spotlight: OpusClip
Many of you have asked about AI tools that could help with content creation, particularly for video content. While I'm not a video creator myself, I've listened to the founder of OpusClip speak and think it is a great tool for you to try out.
Think of it as an AI-powered video editor that transforms your longer content – keynotes, panel discussions, or webinars – into social-media-ready clips. Several keynote speakers I know have started using it to extend their thought leadership across platforms like LinkedIn, Instagram, and YouTube Shorts. It is super easy to use and you can save hours of time!
[Sidebar: The rise of short-form video content isn't just a trend – it's becoming a key channel for business communication. According to recent data, executives who share insights through short video clips see significantly higher engagement than those who stick to text-only posts.]
What makes OpusClip interesting from a business perspective is its approach to the time-cost equation. The AI reportedly does the heavy lifting by:
Identifying key moments from longer presentations
Adding captions automatically (crucial for accessibility)
Suggesting relevant B-roll footage to enhance visual appeal
Formatting for different social platforms' requirements
Beyond professional use, you can use it to create year-end video highlights for the family. Your Gen-Z audience will be impressed!
One More Thing: A Humble Brag 🙈
Life has a funny way of surprising us, doesn't it? Five years ago, if you'd told me I'd be working at the intersection of AI, finance, and governance, I probably would have laughed. Yet here we are!
I started this newsletter as a way to share what I was learning, hoping it might help others navigate this complex landscape. What I didn't expect was the incredible community that would form around it. Your insights, questions, and experiences have taught me as much as any startup adventure!
[Sidebar: Someone once told me that the best communities are built not around answers, but around shared questions. How right they were!]
So here's the slightly awkward part where I share some news (hence the section title 😊): I've been nominated for the "Women in Tech Global Award" at WomenTech Network. While deeply honored to be considered alongside truly remarkable leaders in this space, what makes this meaningful isn't the nomination itself – it's the validation of our collective journey in making AI in finance and governance more accessible and responsible.
Lastly, if you find this newsletter valuable, please share it with your network. I greatly appreciate it.
Thank you!
Joyce Li