This Week in AI: Open Source, AI For Investment Fund Boards, Deepfakes, One AI App To Start Using, and More

Feb 6, 2024

Feb 06, 2024

👋 Hi! This is Joyce Li. Welcome to the inaugural issue of "AI Simplified for Leaders," your weekly digest aimed at unraveling the complexities of artificial intelligence for finance and business leaders, board members, and investors.

Each week, I will curate a selection of must-try AI tools, the latest news, strategic insights, and practical use cases, all tailored to seamlessly integrate into your busy schedule. My objective is to enhance your fluency in AI, enabling you to navigate this fast-evolving field without feeling overwhelmed, thereby empowering you to make informed decisions and pose strategic questions with confidence and clarity.

Open Source Goes Mainstream in AI

On Meta’s recent earnings call, CEO Mark Zuckerberg highlighted open-source software infrastructure as integral to Meta's AI strategy:

“Open source software is typically safer and more secure, as well as more compute efficient to operate due to all the ongoing feedback, scrutiny, and development from the community. ”

Open-source has been widely used in software development for decades and it refers to systems where the source code, models, and tools are freely available for anyone to use, study, modify, and distribute. However, open-source approach’s rise to prominence in AI is still worth noting for a few reasons:

Firstly, recent open-source models, such as Mixtral, have shown performance that rivals or surpasses that of GPT-3.5, while also being more cost-effective. The field of open Large Language Models (LLMs) is rapidly advancing, becoming a foundational layer for AI applications.

Secondly, enterprises can fine-tune their AI models or develop their applications using private copies of open-source LLMs, which come with traceability, transparency, privacy, and security features. This not only reduces costs by eliminating fees associated with proprietary models but also empowers businesses to co-create software, securing intellectual property rights for models that incorporate their unique expertise and ensuring compliance with future regulations.

Thirdly, the long-term viability of open AI business models remains uncertain. Open-source Software-as-a-Service (SaaS) companies often rely on a mix of open-core software and premium enterprise features for their business models. In the AI domain, while the open infrastructure, such as foundational LLMs, offers cost savings on customization, it still requires significant investment in computational resources for training. Meta's strategy of keeping its Llama models open helps the company stay competitive by improving its proprietary data and product layers, which are highly profitable. However, this approach may not be feasible for all companies. Continuous investment is likely necessary, as seen with Mistral, the French company behind the aforementioned LLMs, which secured $415 million in venture capital funding last December. The challenge for many AI model companies is not just to reach but to maintain a position on the model leaderboard, as funding can quickly diminish.

One thing to note:

HuggingFace serves as a marketplace for open-source AI models, datasets, and discussions. It's an excellent resource for exploring state-of-the-art models and engaging with educational content. (Incidentally, HuggingFace secured $235 million last August at more than 100 times its annual revenue. While I admire HuggingFace's contributions to AI advancements, the rationale behind such valuations remains a puzzle to me.)

How Investment Fund Boards Should Approach AI

Recently, I've engaged in discussions with esteemed directors on investment fund boards, where a recurring theme emerged: how should fund boards navigate the integration of AI within their fiduciary responsibilities?

I propose that fund board directors should adopt the following strategies:

Understand AI's Impact on Investment Strategies: Directors should gain a high-level understanding of how AI can enhance investment decision-making, risk assessment, and portfolio management.
Oversee AI-Related Risk Management: Ensure that AI integration aligns with the fund's overall risk management strategy. This includes understanding the risks of AI-driven investment strategies and data security concerns.
Ensure Regulatory Compliance and Ethical Standards: Stay informed about regulations concerning AI in the investment sector and in financial services in general. Ensure that AI applications comply with industry standards and ethical investment guidelines.
Guard Against Conflicts of Interest in AI Use: Ensure that the AI tools or data analytics used do not create or exacerbate conflicts of interest, especially in relation to trading practices or portfolio management.
Facilitate AI Transparency and Investor Communication: Advocate for transparency in how AI is used in fund management strategies and communicate its role and benefits to investors.
Encourage Continuous Learning and Advisory: Promote ongoing education about AI among board members and seek advice from external AI experts as needed.

I wrote more on this topic in this LinkedIn post. In future posts, I will share a list of questions board directors can ask executives around AI.

Deepfakes Are Closer Than You Think

A recent incident highlighted the dangers of deepfake technology, where a company suffered a $25 million loss due to a fraudulent video call. An employee, initially skeptical, was deceived by a fraudster impersonating the CFO with remarkable accuracy.

It is incredibly easy to make a fake video. Wharton Professor Ethan Mollick demonstrated how, with just 30 seconds of his video footage, an AI application like HeyGen can create a believable video of him speaking in Italian. This example reflects the dual nature of AI tools: while they can significantly enhance productivity for educational and business content, they also present opportunities for misuse in fraud and scams.

To protect against such sophisticated threats, leaders must prioritize the implementation of advanced security measures and the training of staff to recognize and counteract these attacks. Measures such as secret codes, multi-party validation, or even reverting to in-person interactions for critical decisions may be necessary. Deepfakes could happen to anyone.

One Thing to Learn

This 1-hour Youtube video by Andrey Karpathy offers a clear explanation of how Large Language Models (LLMs) work. It is a little technical, but at the right level, so quite approachable for business people. I promise you will be able to cut through the noise and ask key questions next time people throw all these AI LLM jargons to you.

AI in Finance: Real Life Use Cases

In corporate finance, AI is transforming operations, as illustrated by OpenAI's use case in the accounting department.

In this podcast, Sowmya Ranganathan, the Controller at OpenAI, shared several accessible strategies for leveraging AI to address everyday challenges and enhance the capabilities of finance teams. A pivotal piece of Sowmya's advice is the importance of incorporating flexibility into internal AI projects. Given the rapid pace of technological advancement, a solution that is effective today may become obsolete in the near future. The primary objective is to liberate team members from routine tasks, enabling them to devote more time to strategic thinking about business requirements and to reengineer workflows to better meet those demands.

Must-Try AI Tools

This issue features a special focus on "Perplexity," a tool that has become indispensable for my research tasks across various platforms. I've integrated Perplexity into my daily routine through its desktop application, Chrome plugin, and mobile app, leveraging its capabilities to enhance my research. The sentiment I have about this tool is aptly captured by a headline from The New York Times below:

Thank You

I hope this issue gives you some actionable ideas and some AI topics to consider. Your feedback and questions are invaluable while I am testing and adjusting the content and format. Please share your insights, use cases, or challenges with me at joyce@averanda-ai.com. Let’s get AI fluent together.