2 posts tagged with "Decision Making"

Key Challenges with LLMs​

• Difficulty with tasks requiring complex planning (e.g., travel itineraries, meeting schedules).
• Performance declines with increasing task complexity (e.g., more cities, people, or constraints).

Three Proposed Solutions​

1. Scaling Law
   • Increase data, compute, and model size.
   • Limitation: High costs and diminishing returns for reasoning/planning tasks.
2. Hybrid Systems
   • Combine deep learning models with symbolic solvers. Symbolic reasoning refers to the process of solving problems and making decisions using explicit symbols, rules, and logic. It is a method where reasoning is based on clearly defined relationships and representations, often following formal logic or mathematical principles.
   • Approaches:
     • End-to-End Integration: Unified deep model and symbolic system.
     • Data Augmentation: Neural models provide structured data for solvers.
     • Tool Use: LLMs act as interfaces for external solvers.
   • Notable Examples:
     • MILP Solvers: For travel planning with constraints.
     • Searchformer: Transformers trained to emulate A* search.
     • DualFormer: Switches dynamically between fast (heuristic) and slow (deliberative) reasoning.
     • SurCo: Combines combinatorial optimization with latent space representations.
3. Emerging Symbolic Structures
   • Exploration of symbolic reasoning emerging in neural networks.
   • Findings:
     • Neural networks exhibit Fourier-like patterns in arithmetic tasks.
     • Gradient descent produces solutions aligned with algebraic constructs.
     • Emergent ring homomorphisms and symbolic efficiency in complex tasks.

Research Implications​

• Neural networks naturally learn symbolic abstractions, offering potential for improved reasoning.
• Hybrid systems might represent the optimal balance between adaptability (neural) and precision (symbolic).
• Advanced algebraic techniques could eventually replace gradient descent.

Overall Takeaway​

The future of decision-making AI lies in leveraging both neural adaptability and symbolic rigor. Hybrid approaches appear most promising for solving tasks requiring both perception and structured reasoning.

Only Live "Day 1" = Without Growth, There is Death​

The office building where Amazon CEO Jeff Bezos works is called "Day 1." Over the years, no matter which other building he moves to, he always brings this same name with him. Therefore, he has a lot of authority on this term.

Someone might ask, what is "Day 2"? Day 2 is stagnation, followed by irrelevance, then suffocating, painful decline, and finally, death.

This is why Bezos believes that every day should be Day 1; without growth, there is death. So how do we prevent "Day 2"? There are four foundations.

A True Obsession with Customers​

There are countless business strategies, but why focus on "obsession with customers"? The benefits are numerous, with the biggest being: Customers are always dissatisfied, even when they say they are satisfied. Customers often don’t know what they truly want: they actually want something better. If you want to serve customers well, you must create products and services in their name. For example, the Prime service was not something customers asked Amazon for, but the results proved it was indeed what they wanted.

Maintaining "Day 1" requires patience; you need a lot of experimentation and to accept failure. Planting seeds and growing saplings takes time, but once you see what makes users happy, double down on it.

Resisting Proxies​

As companies grow larger, we often tend to rely on proxies or intermediaries. This form of dependency can take many shapes and is very much "Day 2." Here are two examples:

1. Relying on processes as proxies for results. Good processes serve you, allowing you to better serve customers. You must never serve the process. Why? When you serve the process, you only focus on doing the process correctly, regardless of the outcome. When failures occur, only inexperienced leaders say, "We followed the process," while seasoned leaders say, "We found an opportunity to improve the process." Constantly ask yourself, does the process own us, or do we own the process?

2. Relying on market research and customer surveys as proxies for customers. When you invent and design products, relying on research can be dangerous; "satisfaction increased from 47% to 55%" is a vague statement that can be misleading.

   1. Good investors and designers deeply understand customers; they invest significant energy in developing intuition and study numerous fascinating anecdotes rather than average data from surveys. They exist to design.
   2. Bezos does not oppose public testing and surveys; they help you identify blind spots, but as a provider of products and services, you must prioritize your vision and unique value over customer feedback. Exceptional customer experiences begin with intuition, curiosity, playfulness, courage, and taste—qualities that user surveys cannot provide.

Embracing External Trends​

The trends of the world favor those who align with them and doom those who resist. These trends are not hard to identify, but strangely, large companies often struggle to embrace them. One such trend today is machine learning and artificial intelligence.

Over the past few decades, many tasks could be solved with precise rules and algorithms; next, with machine learning, we can tackle tasks that cannot be described by exact rules.

Much of what happens in machine learning occurs at the foundational level, out of sight, but you can at least call them very simply via APIs.

Fast Decision-Making​

"Day 2" companies make high-quality decisions, but their decision-making speed is very slow. To maintain the energy and vitality of "Day 1," you must make "high-quality and high-speed" decisions. This is important not only because "speed" matters in the business world but also because having an atmosphere of "fast decision-making" is more enjoyable.

How can you achieve fast decision-making? Bezos does not have a complete answer, but here are some thoughts:

1. Decisions are inherently unequal; never treat them all the same. Reversible decisions should use lightweight decision-making processes.

2. Most decisions can be made when you have 70% of the information. Waiting until you have 90% may be too late. Also, in either case, you must quickly identify and address bad decisions. When you are highly responsive, making mistakes is cheap, while being slow is costly.

3. Use a management style of "==I disagree, but I commit to executing well==." This saves a lot of time spent on disputes.

   1. When no one knows the outcome, ask, "I know we have a disagreement, but are you willing to take a gamble with me? I disagree, but I commit to executing well?" The answer you get is likely to be, "Sure."
   2. The party that disagrees does not commit out of indifference but from a genuine and sincere disagreement, allowing the other party to reconsider your "disagreement" while still acting quickly due to your commitment.

4. Identify misalignments early and escalate them immediately. Sometimes, goals between teams conflict, and disputes at the same level cannot be resolved, wasting a lot of time and energy. In such cases, escalating will make decision-making faster and easier.