"No one is harder on a talented person than the person themselves" - Linda Wilkinson ; "Trust your guts and don't follow the herd" ; "Validate direction not destination" ;

January 06, 2023

Is GPT LLM going to be the next Alexa or Google?

The latest addition to the GPT-3 model family: text-davinci-003. This model builds on top of our previous InstructGPT models. Some questions and the answers I got. Let me know how relevant the answers are. I felt in a matter of few seconds these answers are very very good.

What is difference between conditional random field and naive bayes in 100 words

  • Conditional Random Fields (CRFs) are a type of probabilistic graphical model that are used to model sequential data. 
  • Unlike Naive Bayes, which assumes that all features are independent of each other, CRFs take into account the dependencies between features. 
  • This allows them to capture more complex relationships between the features, which can result in more accurate predictions. 
  • CRFs also allow for the incorporation of prior knowledge into the model, which can further improve the accuracy of the predictions.
  • In contrast, Naive Bayes is a simpler model that assumes all features are independent of each other, and is typically used for classification tasks.

Explain why their is inequality in this world in 100 words ?

Inequality in this world exists due to a variety of factors, including unequal access to resources, education, and opportunities. Inequalities in wealth, power, and privilege are often based on gender, race, ethnicity, and other social identities. These disparities are perpetuated by systems of oppression and discrimination, which can be seen in the workplace, in education, in healthcare, and in other areas of life. Inequality is also perpetuated by unequal access to resources and opportunities, which can lead to poverty, lack of education, and other forms of marginalization. Inequality is a complex issue that requires concerted efforts to address and eliminate.

Question - Explain why productivity is less in software job in 100 words ?

Answer - Software jobs are often complex and require a lot of problem solving and creative thinking. This can lead to a slower pace of work, as it takes time to come up with solutions and troubleshoot issues. Additionally, software jobs often require a lot of collaboration between team members, which can slow down the pace of work. Additionally, software jobs often require a lot of testing and debugging, which can take a lot of time and slow down the pace of work. Finally, software jobs often require a lot of documentation, which can also slow down the pace of work. All of these factors can lead to lower productivity in software jobs.

Explain how to grow as senior engineer in 100 words?

1. Develop a strong technical foundation: Focus on building a strong technical foundation by learning the fundamentals of software engineering, such as data structures, algorithms, and design patterns.

2. Learn new technologies: Stay up to date with the latest technologies and trends in the industry.

3. Network: Attend conferences, meetups, and other networking events to stay connected with the tech community.

4. Mentor others: Mentoring junior engineers is a great way to hone your skills and share your knowledge.

5. Take on challenging projects: Tackle complex projects that will push your skills to the limit.

6. Invest in yourself: Take the time to invest in yourself by reading books, taking courses, and

What questions I should ask my manager in my next one-to-one discussion in 100 words?

1. What do you think are my strongest skills and how can I use them to benefit the team?

2. What do you think I can do to improve my performance?

3. What do you think are the biggest challenges facing the team and how can I help?

4. What do you think I should focus on in the next few months?

5. How can I better collaborate with other team members?

6. What do you think I should do to develop my skills and knowledge?

7. What do you think are the most important goals for the team and how can I help reach them?

8. What do you think are the most important values of the team and how can I help uphold them?

What is difference between markov is hidden markov model in 50 words ?

Markov models are probabilistic models that assume that the probability of a future state depends only on the current state. Hidden Markov Models (HMMs) are similar to Markov models, but they assume that the underlying process is hidden, and the observed data is a result of the hidden process. HMMs are used to model sequences of observations, such as speech or text.

what is difference between emission vs transition probability in 50 words

Emission probability is the probability of a given state emitting a certain observation. Transition probability is the probability of transitioning from one state to another. Emission probability is used to calculate the probability of a given observation given a state, while transition probability is used to calculate the probability of transitioning from one state to another.

what is difference between multiarmed bandit vs markov model in 50 words

Multiarmed bandit algorithms are used to optimize decisions in a situation where there is a trade-off between exploration and exploitation. They are used to maximize the expected reward over a finite number of trials. Markov models are used to model a system where the future state is only dependent on the current state. They are used to predict the probability of future states given the current state.

Why neural networks are non convex in 50 words

Neural networks are non-convex because they contain multiple layers of non-linear functions, which can cause the optimization problem to have multiple local minima and saddle points, making it difficult to find the global optimum.

What is relationship between backpropagation and chain rule derivative in 100 words

Backpropagation is an algorithm used to train neural networks. It is based on the chain rule derivative, which is a method of calculating the derivative of a composite function. The chain rule states that the derivative of a composite function can be calculated by multiplying the derivatives of the individual functions. Backpropagation uses the chain rule to calculate the derivatives of the weights in the neural network. This is done by calculating the derivative of the error with respect to each weight in the network. The derivatives are then used to update the weights, which allows the network to learn. In summary, backpropagation is an algorithm that uses the chain rule derivative to train neural networks.

Playground Ref - Link

There are many domains and aspects. I was trying to ask some questions where I can feel like connecting with answers/opinions. These answers can be translated with Neural Machine Translation and can serve as a go-to Alexa. However impressive but factual information, relevance, and correctness are still a challenge. How we detect/filter false information.

ChatGPT + Google, ChatGPT + Bing. Any combination will really help to improve on relevance / better hybrid results.

The answers it generated I liked it much. I would give it 9/10 if I have to correct the answers/correctness

Keep Exploring!!!

GPT Text Examples

 


Keep Exploring!!!

January 05, 2023

DALLE Text to Image Generation

Example from link

For the text Benz on Beach



Relevant, Realistic, Accurate

Keep Exploring!!!

Learn how to generate or manipulate images with our DALL·E models

 






Very Realistic, New variations!!!

Keep Exploring!!!

Questions about my Work ?

Do you code? Are u a Coder / Non-coding manager?

I code whenever I have to do a POC, Evaluating an approach. In a distributed work, when someone is struck if it's Vision / Model Training Code / Clustering / Approach / DB / Design I review/debug. If it's for algo I reuse my past domain inputs to see what are alternatives. 

Can you write code and demonstrate capability?

Mind visualizes in a different way, not as methods and packages :)

  • For string search, c/c++ for loop comes to mind
  • There is regex in python
  • semicolon for java statements
  • I only remember logic / I StackOverflow it and put syntax
  • Do i remember syntax heavily, No
  • SQL Server I had spent a considerable time I can write about it anytime

Do you know how Transformer works, can u rewrite it?

  • Get back to basics, 1st cut - LSTM you have attention, cell state, memory, forget gate
  • Build your blocks - 2nd cut - A bit of encoder (encode input), and a set of decoders added, from a given list pick where attention is maximum
  • Code up basics, 3rd cut - Rework / check similar simpler examples/paper on the transformer
  • Customize examples - 4th cut - logic/approach/code / keep iterating/building from other repo / customize/working to see if a prototype helps, Still coding / experiment.
Do you know about product management/program management / Data engineering?

Over years I have moved across several roles. You have to collaborate/influence big initiatives/projects. Deep domain knowledge gives us clarity on what solution works and is relevant to our problem statement. I may not have those titles but I have collaborated/delivered with partners in each domain. For bigger initiatives it's Teamwork. 

How often do you code?

Out of 10 git codes, I experiment / 1~2 work and for the rest I end up with environmental issues. You keep changing tabs reading products/code/experiments. It's never fixed but it's more about a solution to build and the rest is part of solving the problem.

Everything does not contribute immediately but every type of error adds up to a backlog of 'Oh I have seen it, ' This could be due to this'

Learned knowledge / applied knowledge / inferred knowledge / connected knowledge.

Domain vs Data vs Code vs product balancing all the aspects.


Why do we Change Roles / Jobs?

  • Experiment with different roles/projects
  • Impact / Change of Environment
  • Learn new things
  • Change of place/people
  • Growth / Financial benefits
  • Money, Work-Life, Win / Failures, Impact, Learning everything matters but at the end of the day pick and choose battles

It's one life, learning never ends, and Nothing comes for free. Keep learning, aware and aligned.

With Covid, I do feel age catching up in my 40s. Pick and choose your battles in Life :). Keep going!!! 

How do I learn as a Manager?

  • As a manager, if you manage people who can manage themselves then you have time to learn
  • As a manager, if you manage people you have to manage then your focus and attention will be divided
  • The most painful thing is making someone understand and aware of their weakness
  • It's difficult to see other's points of view if we defend, Consulting is all about the right tradeoff time vs effort vs impact.
What are opportunities to Learn / Create Impact?
  • Code to prototype solution feasibility
  • Learn to review other's domain use cases
  • Bridge product vs tech wherever possible
  • The lookout where you can resell solutions
  • Collaborate for success, Success needs credits shared, and Failure needs credits owned up
We always keep refining our 'best-fit' architecture

Learning is an endless process like layers of an onion, Same approach / never-ending
  • Know what you speak
  • Code what you speak
  • Speak what you code
  • Improve on the feedback you hear
  • Improve every time when you get an idea
Knowledge is a reflection of your questions, opinions, the code that you deliver, and the design you put on the table. Communication vs Competency vs Consistency all are different skills that you can discover over time.

Keep Exploring!!!

January 04, 2023

AI in manufacturing use cases

Health and safety

  • Employee health & safety: incident prevention
  • Process safety: advanced alarm analytics

Quality

  • Quality inspection in assembly
  • Quality assurance/
  • defect inspection
  • Quality testing
  • Quality prediction

Maintenance

  • Machine health monitoring:
  • predictive maintenance
  • Maintenance planning 

Production process

  • Process optimization
  • Line balancing
  • Product design and development
  • Process parameter optimization
  • Production planning/
  • decision support

Supply chains

  • Future demand and price forecasting
  • Supply chain control tower
  • Warranty and service management

Energy management

  • Energy optimization
  • Electricity demand forecasting
  • Heating and cooling optimization




Ref - Link

Keep Exploring!!!

Robotics and Vision at Amazon

  • Multiple cameras provide a set of eyes on the scene 
  • 3D scene understanding is generated by multiple camera angles
  • Motion algorithms take this understanding of the scene and item identification 
  • Recognize and estimate the shape of individual items
  • Pinch-grasping robot achieved a 10-fold reduction in damage on certain items
  • A robot optimized for deployment in a fulfillment center may require suction and pinching
  • Depending on the type of object recommend a certain way to handle

Ref - Link

Keep Exploring!!!


Delivery Optimizaton at Amazon

  • Local optimizations were tackled first
  • Stop consolidation — a chain of addresses linked by a small road segment
  • Same-day delivery consolidation

Strategies

  • Customer Order and Network Density OptimizeR (CONDOR).
  • Simultaneously determining how orders are split into shipments and the sourcing fulfillment center for each shipment.
  • Determine the right tradeoff between the levels of complexity and optimality.
  • Single warehouse may assemble the order if it has all the items; if not, the order can be split into multiple shipments.

Ref Link 

Keep Exploring!!!

The Logistics at Amazon

  • Delivery between fulfillment centers (FCs), where inventory is stored, and customers’ doorsteps
  • Key factors demand forecasts, cost of acquiring transportation resources, staffing, buffer capacity planning
  • Strategy - origin-destination targets, Demand for specific route/load - what percentage of the demand in a given locality should be met by a given fulfillment center?
  • Customer demand, region-level demand
  • Online controller - scan, label, apply, manifest]


Ref - Link

Keep Exploring!!!


January 02, 2023