Edition 36: The Next Generation of AI
Welcome to the AviaryAI Newsletter!
Thanks for joining us as we explore the intersection of GenAI and finance with practical learnings and the latest relevant insights. Let’s get started.
This week you’ll learn:
- 🧠 Meet O3: OpenAI’s biggest step towards Artificial General Intelligence
- 🤝 Salesforce drops Agentforce 2.0 - A digital workforce
- 📊 Google's new model makes AI decisions crystal clear
- 🎓 Your daily AI lesson: Understanding Epochs
- 💳 What to know about AI holiday scams
Subscribe to the AviaryAI Newsletter here.
GENERATIVE AI THIS WEEK
The coolest things we're watching and why you should care
OpenAI's o3: When AI Starts Actually Thinking
OpenAI’s biggest release of the year has been unveiled: O3 and O3-mini - a new generation of AI models that OpenAI claims to, in certain conditions, approach AGI. These models demonstrate unprecedented problem-solving abilities, achieving near-perfect scores on advanced mathematics exams and surpassing human experts in programming challenges. The o3 model scored 87.5% on the ARC-AGI, a test that evaluates whether an AI can “learn” new skills, tripling the performance of their previous model, o1.
So what?
We're moving from AI that simply follows rules to AI that can think through problems and explain its reasoning. This shift fundamentally changes how organizations can use AI - from handling routine tasks to solving complex problems that previously required teams of experts. The transparency in O3's decision-making process also makes it easier to trust and implement in regulated environments, similar to having a highly capable advisor who can clearly explain their thought process and safety considerations.
Salesforce launches Agentforce 2.0 - Digital Labor
Salesforce has launched Agentforce 2.0, introducing AI agents that can think and reason through complex problems instead of just following pre-set rules. At its core is the Atlas Reasoning Engine, which enables AI to understand context and make decisions more like humans do. Early results are impressive - the system handles 83% of customer support queries independently and has cut human escalations in half in just two weeks of deployment
So what?
While most organizations are still thinking about AI in terms of automation and cost-cutting, we're entering an era where AI becomes a genuine workforce multiplier. This isn't just about replacing repetitive tasks - it's about expanding what your organization can achieve without traditional hiring constraints. Leaders who start viewing AI as a workforce to be managed rather than software to be deployed will have a significant advantage in scaling their operations beyond the limitations of local talent pools and training capacity.
Google Releases New Reasoning Model: Gemini 2.0 Flash Thinking - Rivals OpenAI’s o1
Google has unveiled Gemini 2.0 Flash Thinking, an AI model that processes both images and text while explaining its reasoning process step-by-step. Unlike previous "black box" AI systems, this model allows users to see exactly how it reaches conclusions through a dropdown menu interface. Early tests show impressive speed and accuracy, with the model solving complex problems in seconds while handling up to 50 pages of input text.
So what?
This shift toward transparent AI decision-making fundamentally changes how organizations can approach AI adoption. When you can see how an AI reaches its conclusions, it becomes less of a leap of faith and more of a practical tool – like having a brilliant analyst who can clearly explain their thinking. This transparency doesn't just build trust; it enables better oversight and more informed decisions about where and how to apply AI effectively.
GENERATIVE AI WORD OF THE DAY
Epoch
An epoch represents one complete cycle through the entire training dataset in a machine learning model. During each epoch, the model processes all training examples, computes errors, and updates its parameters to improve accuracy. Multiple epochs are typically required for optimal model performance, as each pass allows the model to refine its predictions and internal representations.
GENERATIVE AI IN FINANCE
The latest news at the intersection of GenAI and Finance
AI-Powered Holiday Scams
Major credit card companies are reporting unprecedented spikes in fraud attempts this holiday season, with Visa blocking 200% more suspicious charges and Mastercard intercepting nine times more fraud attempts compared to last year. The surge is driven by fraudsters using AI to automate sophisticated scams - from mass-producing convincing phishing messages to generating fake IDs and even deepfake verification videos. Even financial giants like JPMorgan Chase and Visa, despite investing hundreds of millions in fraud prevention, are struggling to keep pace.
The scale of the threat is staggering: U.S. consumers lost $8.7 billion to fraud through Q3 2024, up 14.5% from last year. Traditional scams are being supercharged by AI tools that help criminals create convincing fake shopping sites, test stolen credit cards en masse, and bypass identity verification systems - all while targeting vastly more potential victims than ever before.
So what?
AI has flipped the economics of fraud prevention. While institutions invest millions in better detection, criminals can now launch sophisticated attacks for pennies. The math no longer works in favor of traditional security approaches. Organizations need to rethink their entire security infrastructure - from authentication methods to transaction flows - with a focus on making attacks economically unfeasible rather than just detectable. This requires implementing friction strategically: making legitimate transactions smooth while forcing attackers to expend more resources than they can reasonably expect to recover.
About AviaryAI
AviaryAI is the next evolution of financial interaction. Enhance your team with proactive outbound voice agents to welcome new members, encourage credit card activations, and drive non-interest revenue.
