Blogs/ Articles / News
- Relational Model: Designed to separate underlying hardware interfacing (storage details) from definition and manipulation of data.
- Problem with Early Databases: Any update to table structure required changes to the underlying storage strategy, which had to be manually handled.
- Solution: The relational model abstracts storage management. Users specify what they need in a high-level language, while the database management system:
- Optimizes queries using relational algebra
- Handles schema migrations
- Manages storage efficiently
- Considers available processing power
- Emergence of NoSQL (Especially Document Stores):
- Object-oriented programming led to a preference for storing objects directly instead of computing joins and reconstructing them from relational tables.
- Avoids issues like redundant data updates and manual object reconstruction, similar to challenges faced in a simple file database.
- Relational Algebra symbols & Fundamentals
- The metric originated in cellphones where-in companies needed a way to measure product health across different market segments.
- Averages are easily skewed by outliers - With just an increase of a single high-paying customer, the ARPU could increase. But this didn’t tell you about customer increase OR anything about pricing model changes.
- It also doesn’t tell you anything about profitability or price-discrimination ability of the model.
- It doesn’t indicate anything about the efficiency of your sales funnel. How much revenue are you leaving on the table?
- It’s a very transient number, so going on the basis of this for any kind of decision is just a bad move.
- Binural beats are two tones which operate at different frequencies. So each ear hears a different frequency. Interestingly, your brain perceives a 3rd tone which is operates at the frequency difference of the two tones you hear.
- The Brain Wave Entrainment Hypothesis suggests that the brain’s electrical activity entrains or synchronizes with the binural beats. different entrained frequencies are thought to correspond to different mental states -
- Delta (0.5–4 Hz): Deep sleep, relaxation
- Theta (4–8 Hz): Meditation, creativity
- Alpha (8–14 Hz): Relaxed focus, mindfulness
- Beta (14–30 Hz): Alertness, problem-solving
- Gamma (>30 Hz): High-level cognition, information processing
- There’s some apparent secondary effect which depends on the actual individual frequencies of the two tones used, though binural studies mostly study the frequency difference rather than the impact of individual tones.
- Generally left-brain processes right-ear auditory signals. The left-brain is mostly associated with logic & reasoning. In most studies, the right-ear is stimulated at the higher frequency.
- Augmented LLM –> LLM + retrieval + tool-use + selective memory (long & short-term)
- Workflows vs Agents: Workflows having fixed steps in-order. Agents are completely dynamic, we don’t know how many steps their execution will take.
- Types of workflows :
- Simple chaining: A –> B –> Check/Gate –> C or Exit. Here A,B,C are all tasks handled by LLMs
- Routing: A router (LLM based) selects one of several simple-chained prompts
- Parallelization: Single input to multiple simple-chained workflows –> Combine outputs using an aggregator.
- Orchestrator: Hybrid between routing & parallelization. Select one or more in parallel and combine with a synthesizer.
- Evaluator + Optimizer: One handles the task, one handles evaluation. These steps repeat until the evaluator is satisfied.
- Agents: They interact with the human until the task is completely clear. They then operate with tools and get feedback from the environment until the task is sufficiently completed. When they are stuck or need verification, they come back to the human for assistance. We don’t know how the agent might solve a task, but we can limit the maximum number of tools it uses or how many times it takes feedback before completion. This pattern is still emerging.
