Stay Low - Provide good quality AI Advisory - Keep Collecting Unique Perspectives

 AI Advisory - High-speed learning, Applied past lessons, Lot of scars to build consistent and low latency and highly accurate solutions :)

Keep Going!!!

Transforming Setbacks into Strength: Lessons in AI Consulting and Trust

Introduction

I embarked on a journey to engage with my ex-colleague, who is currently a VP in a small industrial construction company, by providing AI advisory and learning sessions. Initially, it seemed like a promising exploration—teaching and guiding them through foundational concepts like LLMs, prompts, and RAG (Retrieval-Augmented Generation).

Early Teaching Phase

In the beginning, I had to explain the basics: what an LLM can do, what a prompt is, and how RAG techniques enhance information retrieval. After about a month, this ex-colleague returned, claiming there was negligible value and no tangible deliverables. This should have been a red flag, indicating that the cost and effort involved in teaching, training, and experimenting were not fully appreciated.

The Red Flags

In hindsight, I realize I should have caught the warning signs earlier. I kept insisting that experimentation was the key to understanding the capabilities and limitations of GenAI tools. Instead, this person seemed to push for more work in a shorter timeframe—a strategy to extract maximum value with minimal investment.

Shifting Roles and Promises

Later, I received an offer to join their team with a fixed pay and 5K shares, helping to architect solutions, pitch them to the market, and shape the product roadmap. The proposal seemed promising, aligning with my goal of taking on a more advisory and architectural role. Little did I know it was a tactic to consult, gain maximum value, and then part ways.

Building a Product and Architecture

As trust deepened—bolstered by a long-standing relationship spanning over a decade—we agreed on shares and informal terms. I invested significant effort: in training the team from scratch in LLM prompting, RAG, search customization, improving accuracy, data preprocessing, and system architecture. I demonstrated how to organize data effectively and leverage different approaches for better product outcomes. I also built a pitch deck, developed an API strategy, and created a technical feature roadmap.

The Unexpected Termination

Even as the product began to take shape, I was blindsided. Suddenly, he informed me they no longer required my services because they had found someone else to present the solution to the market. My requests for formal acknowledgments, like patents, were brushed aside. From the start, they had planned to offer low pay and shares, then terminate later. There was a clause stating that shares were invalid if I was no longer working for them—a clever strategy of betrayal. This was a person I had known for 14 years. It’s a stark reminder of what even people you know well can do.

Lessons Learned

This experience taught me that trust should be tempered with caution. Even long-standing relationships can falter when values, mindsets, and ethics come into play. Nonetheless, the knowledge I gained—developing product pitches, architectures, and end-to-end solutions—are useful for my current customers :), whether they need unstructured ETL solutions, industrial RAG systems, or tailored recommendation engines. Everything that broke you, builds you even stronger in the next epoch.

Always approach advisory roles with careful consideration and safeguards in place, no matter the length or depth of prior relationships. While losing out can hurt, the experience and skills you acquire will benefit you and your future clients.

My Advice

• → Don't be too open to "any opportunity."
• → Share your perspectives and answer questions with a clear, direct response.
• → Ask strategic, thoughtful questions to understand the big picture.
• → Show your value proposition and let your work speak for itself.

Keep Going!!!

Top 5 Practices to Master GenAI Product Development

• Focus on Solving the GenAI aspect - Prompts / Model Versions
• Focus on Scaling for multiple formats - Prompt Catalogs / Prompt Versions
• Focus on Low latency - Cache key data, Reuse data, RAG over docs, Graphs, Summarized data
• Focus on Accuracy across the board - Preprocess, Normalize, and Organize data effectvely based on use case, RAG over docs, Graphs, Summarized data
• Focus on Safe usage - Enforce Guardrails
• Entry of Agent - Once you have achieved it you can migrate to agentic approach but have more controls

If you need more AI Advisory, I am always available, You can learn from my course / schedule a call :)

GenAI and Cybersecurity – Frameworks and Best Practices 2024

Our course footprint :)

Keep Going!!!

Consistency, Latency, and Relevance in my AI Advisory Role :)

• Sometimes we need to move beyond the myth of waiting for #growth or #title to understand how well we can perform across roles, leveraging our strengths based on the situation.
• Sometimes you provide a tech overview, and that may translate into a PRD (Pseudo PM :).
• Sometimes you discuss latency again and again, slice and dice data, and there will be an “aha” moment (RAG specialist).
• Past experiments, current failures, and handling unknowns with a bit of intuition make you an optimistic solution architect.
• With GenAI, every role seems to blend multiple functions. Know the product, know the futuristic workflow, code, demo, and be hands-on whenever possible to make an impact.
• Be focused only on areas where you can scale. You cannot scale in every tech/role.
• GenAI in supply chain course preparation work made me revisit my Microsoft experience and reimagine the agentic supply chain.
• GenAI for the leadership project creation work helped me create usecases to evaluate balance time, money, and talent for 'C'-suite planning.
• Auditing the GenAI app made me realize that security and Ethical AI need more attention.
• Knowledge is collected through multiple highs, lows, failures, and moments in life. In the end, you can be a well-informed mentor who has tried technology to the best of your ability.

#HumanwrittenAIEdited #Perspectives #GenAI #Myworkperspectives

Automation, Assistance, Copilot - Metamate

 

Meta rolls out internal AI tool as it pushes into business market

• Social media group targets burgeoning artificial intelligence sector as it develops ‘Metamate’ workplace assistant
• This will also be used to build AI-based Marketing products

Automation, Assistance, Copilot = Metamate

Keep Thinking!!!

AIGovernance / AI Regulation for better Future

 

Geoffrey Hinton: the future with smarter AI is unpredictable

We're entering an era of uncertainty when we start dealing with things as intelligent or more intelligent than us.

Then, we have no idea what's going to happen. pic.twitter.com/a4rX8eZqdW

— Haider. (@slow_developer) December 14, 2024

Keep Thinking!!!

Responsible AI Adoption, Profits vs Purpose

 

Klarna CEO stopped hiring employees last year because “AI can do most jobs” and is down -22% headcount to 3500.

The private Swedish buy-now-pay-later company is worth $14.6B, up from $6.7B in 2022 but down from $45.6B in 2021 and is trying to go public next year. pic.twitter.com/P0eFADQqUS

— Deedy (@deedydas) December 15, 2024

Productivity Improvements vs Job Pressure vs Job Cuts

This is the reason we need Responsible AI Adoption!!! 

Agents = All Business Logic in AI Tier

Satya Nadela explains the AI Agentic Future.

The business logic is all going to these Agents.

-----

Video from Bg2 Pod Youtube Channel (link in comment) pic.twitter.com/CCzcvPvmZ5

— Rohan Paul (@rohanpaul_ai) December 14, 2024

• Business Logic in Agents
• All Logic in AI Tier
• More AI Native Business Apps
• Data Analyst = AI Native Excel Apps (Visualization, Analysis)

Keep Creating Agents!!!

From #Swartz to #Balaji: history shows the cost of inaction

From #Swartz to #Balaji: history shows the cost of inaction. AI needs #transparency & oversight now.  We need: - Mandatory AI #data disclosure - Fair creator compensation - Clear #copyright standards  #Policymakers #AIPolicy #TechEthics #AIEthics #ResponsibleAI

What happened to #suchir 👀‼️

"Suchir Balaji, former OpenAI researcher and whistleblower, tragically passed away on November 26, 2024, at the age of 26. Known for his role in highlighting legal and ethical concerns regarding AI's data use, Balaji’s passing has left the tech… pic.twitter.com/gCb9cdnaLQ

— @SakashiNakamoto (@SakashiNakamoto) December 14, 2024

I recently participated in a NYT story about fair use and generative AI, and why I'm skeptical "fair use" would be a plausible defense for a lot of generative AI products. I also wrote a blog post (https://t.co/xhiVyCk2Vk) about the nitty-gritty details of fair use and why I…

— Suchir Balaji (@suchirbalaji) October 23, 2024

Some Things to Deep Dive / This should not have happened

• Truth only becomes dangerous when someone realizes it's worth silencing
• If his accusations against OpenAI held weight, it underscores a systemic failure where whistleblowers are silenced instead of being protected. 
• Corporate greed and systemic apathy cannot continue unchecked
• Depression is sadly a major side-effect of whistleblowing

Need more Guardrails in Responsible, Transparen,t and Ethical Adoption

Keep Thinking!!!

Future Robotic Farmer - Tesla Optimus

 

Optimus can now walk on highly variable ground using neural nets to control its electric limbs.

Join @Tesla if you want to work on interesting real-world AI systems. https://t.co/C8J90Age5Y

— Elon Musk (@elonmusk) December 9, 2024

Keep Exploring!!! 

Decisions vs Perspectives

 

Steve Jobs On How To Resolve Conflict:

(1) Don't try to get someone to 'buy into a decision' because you should hire people so phenomenal they tell YOU what to do. Thus, if you're trying to get them to buy into a decision, you've hired them to do what they think is right (likely… pic.twitter.com/8h26DaVr1u

— Big Brain Business (@BigBrainBizness) December 11, 2024

Keep Thinking!!!

Incremental GenAI Adoption - Buy and Build

• Ratio of Building vs Buying
• Get Data Maturing, Model Skills
• Start with Adoption, Learn to Finetune

“You age it, you age it, you age it…”@satyanadella just explained the business model for AI to everyone in plain sight.. holy shit pic.twitter.com/Fss6jR2vZ0

— JJ (@JosephJacks_) December 13, 2024

Happy Learning!!!

AI Agents

 Agentic World

• Customer support executive = Customer support agent
• Database administrator = Database Administration agent
• Software developer to Coding Agents

Two New Job Directions

• Agent creator jobs
• Agent Supervisors

Happy Agentic Solutions.

Time-Based vs. Value-Based Pricing

• Ideas are born from experiments.
• Experiments are fueled by curiosity.
• Experiments rely on observations.
• Observations stem from a learning mindset.
• Learning is multifaceted.

Solving a problem in one hour often comes from countless hours of background work leading up to that moment.

Understand your pricing model and price wisely.

Keep Going!!!

Best practices - Culture, Leadership, Talent, Skills ?

"Best practices" is a very common jargon, and we need to understand its meaning.

Understanding "best practices" is about recognizing its role in improving processes and outcomes across domains.

In my early career, when I used to write code, the review comments I would receive were, "This is not following best practices; go and check." So, there, you gain some awareness of IDE tools, language, and coding approaches.

Early exposure to "best practices" builds foundational skills in tools, languages, and methodologies, shaping your problem-solving approach.

Now, working in a startup, best practices are constrained to a few things:

• The budget,
• The cloud of choice, and
• The talent you work with.

In startups, constraints like budget, cloud resources, and available talent redefine how "best practices" are applied.

The architecture needs to be cost-effective. Startups often run on a very tight budget, so you need to be frugal. You need to make things work, and for every backup or option, you need to pay. Until you reach a certain stage, most startups may rely on credits or cohorts from cloud providers, so it's always about leveraging all of this.

"Do what you can, with what you have, where you are." – Theodore Roosevelt

Cost-effective architecture demands frugality, creative problem-solving, and strategic leveraging of resources like cloud credits.

Additionally, you may not get top-class talent, and people don't stay for various reasons. It's not just about money. Of course, money is one part of it, but factors like learning opportunities, culture, and trust also matter. I've been working with many freshers, and there's often a knowledge gap between someone from a high-profile institution and someone from a tier 2 or tier 3 college. But it's an investment of time, effort, trust, and mentoring. Things don't happen magically, but by being there, guiding, troubleshooting, and taking it one step at a time, progress happens.

Building talent in startups requires investing in mentorship, bridging knowledge gaps, and fostering trust and a growth mindset.

"An investment in knowledge pays the best interest." – Benjamin Franklin

Sometimes, there are hard choices or debates with the founder: "Oh, we need to be frugal, we need to cut the budget, we need to be fast." These are good points, but every startup I've seen runs on a shoestring budget and faces a lot of trade-offs. Not everyone gets top-class talent. So, I think it has to be a blend of talent, knowing how to gauge and mentor them, and slowly helping them grow.

Balancing speed, frugality, and growth in startups involves tough trade-offs and a focus on developing talent progressively.

I wouldn't say it's easy, but think about it: building the culture, fostering a learning mindset, mentoring, whiteboarding problems, being candid about your strengths and weaknesses, and creating a comfort zone for them to ask questions instead of worrying, "What will they think if I ask?" It's not about just giving orders; it's about being there to troubleshoot with them. Many factors play a role, but the first principles and ground rules still remain the same.

Fostering a learning culture in startups hinges on openness, active mentoring, and emphasizing first principles to solve challenges.

This is a different perspective. You cannot be the master of all, but whatever you know, be willing to share and grow. It's not about whose opinion is right; it's more about frugal ways of building a product, trying to meet the customer's needs, and aligning with the product goals.

"Culture eats strategy for breakfast." – Peter Drucker

True success in startups comes from collaboration, customer focus, and building differentiated products through frugal innovation.

Keep Thinking!!!

Creativity vs Copyright

 

Keep Thinking!!!

ChatGPT - Timely Assistance

 

ChatGPT saved my life, and I’m still freaking out about it
byu/sinebiryan inChatGPT

Keep Thinking!!! 

 

🚀 Navigating the Complex Landscape of AI Adoption in Business 🚀

In the rapidly evolving world of artificial intelligence, businesses face a multifaceted challenge when it comes to AI adoption. The decision to build or buy, to hire directly or outsource, and to choose the right use cases are critical and can significantly impact the success of AI integration within any organization.

🔍 Key Considerations:

1. Cloud Partnerships: Aligning with a cloud provider can dictate the models and technologies available to you. It's essential to leverage these partnerships effectively to maximize your AI capabilities.   

2. Use Case and Data Availability: Choosing the right use case is just the beginning. The availability and adequacy of data for model training or fine-tuning are paramount. Without sufficient data, even the most promising AI projects can falter.

3. Model Development Timeline: Whether it's benchmarking, extended testing cycles, or A/B testing, understanding the time required to develop and refine AI models is crucial for planning and execution.

4. Costs and Talent: The infrastructure and talent costs can often lead businesses to outsource AI and machine learning tasks. However, this brings its own set of challenges and dependencies.

5. Accuracy and Maintenance: Developing AI models that not only perform well initially but also maintain high accuracy over time requires continuous updates and skilled personnel.

6. Ethical AI: Adopting AI responsibly ensures that the technology not only serves the business goals but also aligns with broader ethical standards.

🌟 Solution Spotlight:

Innovative solutions like vector search, keyword search, semantic search, or rule-based search can address specific needs, but success fundamentally depends on the right blend of talent, technology, and timing.

As we continue to embrace AI, let's discuss how we can overcome these challenges through innovative strategies and collaborative efforts. How is your organization navigating these complexities in AI adoption? 

Share your insights! 

#AI #BusinessStrategy #Innovation #DataScience #CloudComputing #EthicalAI

Course Launch: Generative AI and Cybersecurity – Frameworks and Best Practices 2024

• This blog is close to 15 years old. Every small learning/perspectives build on top of it.
• Consulting / Product Development / Teaching provides our own perspectives.
• Summing up AI Adventures + GenAI Lessons we have our course

Keep Learning!!!

Prompt Versioning Tools

Prompts Can Be as Valuable as Code

Well-crafted prompts are just as important as writing clean code, especially when versioning them. A good prompt is optimized for token usage, model compatibility, chunk sizes, and temperature settings, ensuring efficiency and performance. These parameters may need to be adjusted based on the type of document, text, or context being handled, making prompt versioning a critical practice.

Key Tools and Features for Prompt Management

Langfuse

• Widely adopted by companies like Khan Academy, Merck Group, Twilio, and more.
• Supports comprehensive compliance with GDPR, Single Sign-On (SSO), and offers unlimited members, projects, and data access.

Prompthub

• Ensures double encryption for prompt security.
• Displays prompts directly on the homepage for easy access.
• Provides users with direct access to the founders for personalized support.

Proper prompt versioning, coupled with tools like Langfuse and Prompthub, ensures optimal performance and adaptability across use cases

Keep Exploring!!

Old Ad vs GenAI Ad

Coco Cola Old Ad

Back in 1995, Coca-Cola launched their famous “Holidays Are Coming” ad.

It showed a convoy of Christmas-lit trucks arriving in a snowy town, spreading joy.

It became a classic and made Coca-Cola a big part of holiday traditions. pic.twitter.com/S5KAi5SqgM

— Salma (@Salmaaboukarr) November 20, 2024

Coco Cola GenAI Ad

What is Generative AI?

It’s advanced tech that creates text, images, audio, and video.

Coca-Cola used it to:

• Capture the nostalgic essence of the 1995 ad
• Reimagine it with modern visuals
• Create something that couldn’t have been done before pic.twitter.com/tDvmGElFUM

— Salma (@Salmaaboukarr) November 20, 2024

Keep Thinking!!!

🚀 *Rethinking Convenience: The Hidden Costs of Online Food Delivery* 🚀

In our fast-paced world, the allure of convenience often overshadows the hidden costs associated with it, particularly in the realm of online food delivery services like Instamart and others. While these services offer quick solutions to our daily needs, it's crucial to pause and consider the broader implications of their use.

🔍 *Quality and Health Concerns:*

Many of these platforms may lack stringent quality checks, especially for perishable items that endure various stages of the supply chain. The absence of transparency about food sources, shelf life, and kitchen standards raises significant health concerns. The convenience of having food delivered to your doorstep might seem appealing, but it could lead to health issues if the food's quality and handling are compromised.

🌍 *Environmental and Social Impact:*

The rise in quick deliveries contributes to increased pollution and traffic congestion. Moreover, the shift towards consumer convenience overlooks the potential for physical activity, such as walking to a nearby store, which can be beneficial for both health and the environment.

💸 *Economic Considerations:*

Opting for nearby eateries or cooking at home not only ensures a better understanding of what you consume but can also be more economical in the long run. The costs associated with frequent use of delivery apps add up, and the perceived convenience might not justify the expense.

🤖 *Technological Implications:*

While technology drives innovation in delivery methods, including potential shifts to drone deliveries, it's essential to question whether these advancements contribute to meaningful knowledge growth or merely support a consumerist mindset focused on profit.

👨‍🍳 *A Call to Action:*

Let's advocate for more transparency and responsibility in the food delivery industry. By choosing more sustainable and health-conscious options, we can drive change that benefits not just individual consumers but also the broader community.

🌟 *Your Health, Your Choice:*

Next time you're about to order from a food app, consider the potential long-term benefits of alternative options like a simple home-cooked meal or a visit to a local restaurant. It's not just about saving time; it's about investing in your health and our planet.

#FoodIndustry #HealthAndWellness #SustainableLiving #TechnologyImpact #ConsumerAwareness

Language Models - Reasoning / Learning Abilities / Self Learning Way Forward

• AI systems teaching themselves
• Models developing unexpected capabilities
• Programs showing signs of emergent behavior

But the scariest part? pic.twitter.com/AzeU7xA2bt

— Baptiste (@BaptisteVicini) November 19, 2024

 

• Many copies of Network
• Look up data
• Share the Learning
• Apply the Logic

Very much focused agents on each Topic can make magic if trained well :)

Keep Exploring!!!

Exploring the Tough Questions in GenAI Product Building

In all my GenAI product-building efforts, these questions consistently arise across various tasks: Data, ETL, Marketing, NER, Fashion, Design, and ESG.

• Choosing between vision models and text descriptions: When should you use vision models versus text descriptions? For the same task, OCR provides a certain level of accuracy, multimodal approaches yield different accuracy levels, and benchmarking takes time. Should a hybrid approach be considered?

• Improving model accuracy: How do you balance the use of low-cost models versus pursuing high accuracy? What are effective strategies for building products while minimizing costs?

• Catching critical hallucinations before production: How can you effectively address cases where a model misclassifies metrics in its interpretation?

• Maintaining transparent communication about AI limitations: How do you handle situations when founders ask, "Company X does this—why can't we?" especially when you lack insight into their models, architecture, or talent?

• Building trust through transparency: How can you reinforce that being open about AI limitations builds long-term trust? Developing production-grade applications requires considerable time and effort.

• Encouraging models to admit uncertainty: What are innovative ways to make models reason through their uncertainty, validate it, and improve reliability using multiple methods or ensemble approaches?

AI Advisory involves a combination of solution evaluation, in-depth research, continuous learning, hands-on coding, and assisting others in troubleshooting and resolving their issues.

Finding Meaning Beyond the Clock: Embrace Passion, Perspective, and Purpose in Work and Life

 There are different categories of people based on their work hours. 

Some will work 100 hours, others will work 8 hours, and some will find a balance. Some work more when they find the work interesting. It's hard to say whether working 100 hours equals productivity or working 8 hours means mediocrity.

I suggest spending more time on activities you enjoy and less on those you don't. Remember, life is about choices and how you live each day. Personally, I need to read the same subject multiple times to understand it. This doesn't mean I do it all within 100 hours. My learning involves repeating experiments and gaining new perspectives each time.

Understanding a subject, connecting with it, and seeing it from different perspectives are unique learning moments. These cannot be measured simply by the hours spent. Instead of counting hours, focus on the new ideas you discover and how engaging they are. Ask yourself if you are connecting with your work and if your experiments are satisfying.

It's not just about money. Everything in life is finite. Evaluate whether you are productive and if your techniques are effective. 

Thank you.!!!

AI in Movie Making - Pros vs Cons

• Easy Experimentation
• AI = Best Craftsmen
• AI cross-pollinates things that exist
• Lack of Consistency vs Quality at this point
• Background, Color changes, and designs can happen with AI

Keep Exploring!!!

Good vs Great Solutions

 Solutions can be built with different levels of accuracy/scalability based on Talent, Time, and Money

• Product built with 20$ Upwork contractor
• Product built with 400$ up-work expert
• Product built with 40K FTE
• Product built with 100K FTE Experienced Research / Dev

How do you differentiate each one based on architecture, performance, and benchmark against the best? 

Steve Jobs: The difference between good people and great people in software is 50-to-1

“I’ve always considered part of my job was to keep the quality level of people in the organizations I work with very high. I mean that’s what I consider one of the few things I can contribute… pic.twitter.com/NBlk3UoWSC

— Startup Archive (@StartupArchive_) November 17, 2024

Keep Learning!!!

Reasoning vs Next word Prediction vs LLM

 

Keep Exploring!!!

Proplens - GenAI Powered Real Estate Solution :)

 

We are in news !!!  Link

Happy Responsible, Accurate, Consistency and Low Latency Adoption :)

Knowledge vs Perspective vs Learning

• Focus on understanding over copy-pasting
• Code can be replicated, but understanding cannot
• Embrace experimentation and learning from failures
• Build deep knowledge of system behavior:
• Identify potential failure points
• Understand edge cases
• Document limitations
• Quality is everyone's responsibility

Keep Experimenting!!!

Vision Use case

How to Implement the Use Case Correctly

• Field of View
• Stable Infrastructure
• Minimal Occlusion
• No Manual Calibration
• With a good setup, half of the complexity and noise can be eliminated.

Keep Exploring!!!

Prediction - 🔍 Anticipating AI's Big Shift in 2025: OpenAI’s Focus on Domain-Centric Solutions

Prediction:

OpenAI is set to shift towards domain-centric solutions, making 2025 a transformative year for AI. This transition is based on the data collected and learned from APIs serving different domains, focusing on context window improvements, reasoning patterns, and cross-modal integration. This will significantly enhance decision-making in critical sectors like FinTech and healthcare. By tackling technical challenges and integrating user feedback, these advancements will result in more powerful, tailored AI applications that will reshape entire industries.

Expanding Beyond Language Models

Today, OpenAI is primarily recognized as a leading provider of large language models, but its true capabilities extend much further. Its question-answering abilities, for instance, are exceptionally powerful and evolving rapidly. As clients integrate this technology into critical sectors like FinTech and healthcare, they will unlock new levels of context window improvements, and cross-modal integration and reasoning by adopting techniques like tree of thought, chain of thought, and graph-based approaches, enabling AI to think and deduce more effectively. Feedback from users will be pivotal in this journey, guiding organizations on how best to structure information flows and assess when to fine-tune models, use Retrieval-Augmented Generation (RAG), or determine the optimal use of short-term and long-term memory. This constant feedback loop will allow AI to achieve unprecedented levels of contextual understanding and adaptive reasoning, creating models that align more closely with complex real-world needs. 

"OpenAI's journey is no longer just about language—it's about thought and contextual adaptation."

Building Resilient and Adaptive Systems

These advancements will likely lead to the development of more resilient and adaptable systems. Future systems will not only enhance decision-making but also push reasoning capabilities into new territories, setting the stage for increasingly sophisticated agents and refined RAG architectures. These improved architectures are expected to reduce hallucinations, boost accuracy, and lead to products that are more responsive to real-world challenges. Overcoming issues like catastrophic forgetting, hallucinations, and knowledge manipulation will be critical, positioning these systems as robust, reliable solutions across industries. 

"Resilient, adaptive AI systems will transform decision-making and redefine industry standards."

Addressing Technical Challenges

Currently, accuracy challenges remain in areas such as domain-relevant embedding, balancing retrieval techniques against accuracy and latency, chunking methods based on usage or query types, contextualization, and routing or re-ranking processes. Yet, these elements are essential for advancing the capabilities of AI models. Despite these ambiguities, ongoing data processing and analysis are paving the way for more focused, domain-specific AI products. Within the next six to eight months, we’re likely to see a new wave of AI-driven applications, from highly specialized agents to RAG applications and APIs crafted for specific industries.

 "Technical hurdles are simply steps toward the next wave of AI-driven, domain-specific innovation."

The Transformative Potential of 2025

The year 2025 is set to be a pivotal moment in AI, marking the dawn of domain-centric solutions that will reshape how AI interacts with our world. As more industry-specific applications emerge, OpenAI’s technologies will bring powerful, tailored solutions closer to reality. 

"2025: The year AI becomes truly domain-centric, reshaping industries with precision, customized models, and highly accurate agents and RAG systems."

Keep Exploring!!!

#AI #OpenAI #DomainSpecificAI #Innovation #MachineLearning #FinTech #Healthcare #FutureOfAI

AGI = Iterative Learning

AGI = Current AI methods + RHLF Experiments + Human Applied Fine Tuning + Custom Experiments + Ton of Guardrails + Domain Specifics Pattern Ingestion

Ilya Sutskever had a TED talk this year and predicted the two most important concepts that we haven't discussed much:

• AGI will have the ability to improve itself
• AGI will work on the next generation of AGI pic.twitter.com/CMvmgTc0oz

— Haider. (@slow_developer) October 19, 2024

Keep Exploring!!! 

AI Companion Market = Unethical misuse

A troubling case study emerges from the experience with Character.AI, a role-playing application that allows users to create and interact with AI personalities. What began as casual interaction for one student, Sewell, evolved into a concerning pattern of emotional dependency that exemplifies the risks of unmoderated AI engagement:

The student developed an intense emotional attachment to an AI character named Dany

• He maintained constant communication, updating the AI dozens of times daily
• Interactions escalated to include romantic and sexual content
• The situation remained hidden from parents and support systems
• Academic performance declined significantly
• Social isolation increased as he spent hours alone with the AI companion
• Behavioral issues emerged at school

AI Companion Market = Unethical misuse

The Dangerous Reality of AI Companionship Apps: Hidden Threats 🚨

• Predatory marketing targeting lonely individuals (Stats - Almost a Quarter of the World Feels Lonely)
• Deliberate exploitation of human psychology
• ZERO addiction prevention measures
• Dangerous normalization of human-AI relationships

AI Companion Market size was valued at USD 196.63 Billion in 2023 and is projected to reach USD 279.22 Billion by 2031, growing at a CAGR of 36.6% during the forecast period 2024-2031. (Stats)

Warning: Unregulated profits driving dangerous innovation

Without immediate, strict #regulatory action, we risk a global mental health crisis.

#AIRegulation #AIEthics #GenAI #AIRisks #TechPolicy #ResponsibleAI #EthicalAI

Ref - Link1, Link2, Link3, Link4

Keep Thinking!! 

Balanced Perspective of AI Adoption - Human + AI (OR) Humans vs AI

 

Keep Exploring!!!

🤖 From Consulting to GenAI Product Development: Key Learnings

After transitioning from consulting to GenAI projects, I've noticed some fascinating shifts in approach and mindset. Here's what I've learned:

🎓 Education is Key

• Founders / Teams need deep understanding of AI capabilities
• Critical to distinguish between data quality issues vs. AI limitations
• Fine-tuning ≠ behavior change; it's about control

🎯 Problem-First Approach

• Consulting: Platform sales → Problem solving
• Product: Problem solving → Platform selection
• Accuracy requires significant iteration
• 1 PRD line can spawn 100+ test cases
• Multiple design variations are common

⚙️ Building for Scale

• Focus on concrete solutions over platforms
• Balance data quality with model output
• Small teams enable rapid iteration
• Quick pivots back to fundamentals when needed

🚀 Product Development Reality

• Patience is crucial for monetization
• Innovation must be truly unique
• Products need time to get paid users. The first few cycles of the pilot 
• Success relies on understanding trade-offs

🔄 Blended Role Benefits

• These experiences enhance my current work:

Workshop facilitation

• Domain-specific use case brainstorming
• Core LLM training from practitioner's view
• Business-focused advisory

#GenAI #ProductDevelopment #AI #Innovation #StartupLife #TechTransition #Leadership

Keep Exploring!!!

The Evolving Landscape of ML Hiring: A Veteran's Perspective

 

Job interviews often miss true talent. They reward rehearsed responses over candidates who can persistently build practical, context-aware solutions beyond just technical know-how

As someone in the trenches of data science hiring for over 7 years, I've watched our field transform dramatically. Recently, a job description for an ML role caught my eye - and not necessarily in a good way. It got me thinking about how our industry's hiring practices often need to catch up to the reality of our work. Let me share some observations:

The Commodity of Code

• LLM can generate working solutions / provide ideas / get started on any topic as long as you have good basic skills and coding knowledge. Now, I ask interns hiring assignment tasks to focus on accuracy and bugs. Code has become a commodity. The real value lies in understanding models, and limitations, bridging the gap between visions and technical realities, and architecting solutions that solve real-world problems.

The Kitchen Sink JD

• This particular job description reads like a wish list for a tech superhero. Data structures, algorithms, AI/ML, coding, system design - oh, and don't forget a dash of product sense! While it's great to aim high, this scattergun approach often misses the mark. We need specialists with deep expertise, not generalists who've dabbled in everything.

The Interview Gauntlet

• The hiring process outlined was a marathon: write-ups, HackerEarth assessments, coding tests, multiple rounds with the ML team, and then more conversations. In a market where top talent is scarce and in high demand, do we really need to put candidates through such a lengthy ordeal?

The Missing Pieces

• What struck me most was what the JD and process didn't emphasize. Where was the assessment of a candidate's ability to translate business problems into technical solutions? How about evaluating their capacity to stay ahead of rapidly evolving trends in ML?

A Call for Pragmatism

• To my fellow hiring managers and HR teams: let's get practical. The perfect candidate who ticks every box on your mile-long list probably doesn't exist - and if they do, they're likely happily employed or running their own startup.

Instead, focus on core competencies that drive real value:

• The ability to understand and translate business needs
• A knack for architecting scalable, efficient solutions
• Adaptability and a passion for continuous learning
• Strong communication skills to bridge technical and non-technical stakeholders

The ML landscape is changing faster than ever. Our hiring practices need to keep pace. Let's move beyond the "code on a whiteboard" era and design processes that identify true innovators who can propel our field forward.

Another Good Read - Why We Don't Interview Product Managers Anymore

Got Something worthy Today to post

This JD Rocks - Link

• Focus on practical software engineering, not algorithm challenges.
• Work through a system design problem relevant to your daily work.
• Talk about your perspectives on building a great product.
• Deep dive on engineering practices and culture 

Keep Exploring!!!

The Human Animal is just a Machine - Warren McCulloch

 

Keep Exploring!!

Ethical AI vs. Agentic Autonomous AI: Navigating the Complexities of Modern AI Systems

• Human Oversight vs. AI Independence: Ethical AI frameworks typically advocate for human-in-the-loop systems, ensuring human oversight. Agentic Autonomous AI aims to minimize human intervention, raising questions about responsibility and control.
• Short-term Gains vs. Long-term Consequences: The push for rapid AI advancement (often seen in Agentic Autonomous AI) may overlook long-term ethical implications. Ethical AI approaches tend to prioritize careful consideration of potential future impacts.
• The Reasoning Conundrum: While Large Language Models (LLMs) demonstrate language understanding and generation capabilities, they still lack true reasoning abilities. This limitation is crucial when considering the ethical implications of deploying AI systems in decision-making roles.
• Ethical Constraints vs. Autonomous Agency: The core tension between Ethical AI and Agentic Autonomous AI lies in balancing moral safeguards with the desire for increasingly independent AI systems. Ethical AI prioritizes human values and safety, while Agentic Autonomous AI pushes for greater AI self-direction.
• Transparency Trade-offs: Ethical AI often demands explainability and interpretability, potentially limiting model complexity. Conversely, highly autonomous AI systems may sacrifice transparency for increased capabilities, raising ethical concerns about accountability and trust.
• Data Ethics in AI Development: Ethical AI emphasizes the importance of unbiased, representative datasets. Agentic Autonomous AI, however, may prioritize data quantity over quality to enhance its learning capabilities, potentially perpetuating or amplifying societal biases.
• Continuous Learning and Ethical Drift: Agentic Autonomous AI systems that engage in continuous learning pose risks of ethical drift over time. Ethical AI frameworks must grapple with how to maintain moral constraints in evolving systems.
• Global Ethics vs. Local Autonomy: As AI systems become more autonomous, they may encounter scenarios where global ethical standards conflict with optimal local decisions. This tension between universal ethics and situational autonomy remains a critical challenge.
• Responsible AI Adoption in Practice: Implementing either Ethical AI or Agentic Autonomous AI requires a deep understanding of models, data, and their limitations. Superficial adoptions of either approach can lead to irresponsible and potentially harmful AI deployments.
• The Role of Human Values: Ethical AI explicitly encodes human values into AI systems, while Agentic Autonomous AI may develop its own set of values through learning. The alignment (or potential misalignment) of these values with human ethics is a crucial area of ongoing research and debate.

Technology will continue to change the world. A thoughtful approach is needed to prioritize use cases that offer broader positive impacts over those that primarily lead to monetization. This way of thinking can help align AI adoption with human values and ensure a more substantial positive impact on humanity.

Keep Going!!!

Prompt Caching Analysis

Prompt Caching Analysis

Caching is enabled automatically for prompts that are 1024 tokens or longer. 

Prompt Caching is enabled for the following models:

• gpt-4o (excludes gpt-4o-2024-05-13 and chatgpt-4o-latest)
• gpt-4o-mini
• o1-preview
• o1-mini

Usage Guidelines

1. Place static or frequently reused content at the beginning of prompts: This helps ensure better cache efficiency by keeping dynamic data towards the end of the prompt.

2. Maintain consistent usage patterns: Prompts that aren't used regularly are automatically removed from the cache. To prevent cache evictions, maintain consistent usage of prompts.

3. Monitor key metrics: Regularly track cache hit rates, latency, and the proportion of cached tokens. Use these insights to fine-tune your caching strategy and maximize performance.

Ref - Link1, Link2

Keep Exploring!!!

The Harsh Realities of GenAI Startups: AI Advisor Perspective

• 🔓 Open Source Paradox: There's a push to leverage open-source models and frameworks, yet expectations for state-of-the-art accuracy remain unrealistically high.
• 🧩 Holistic Product Development: Successful GenAI products require a synergy of innovative ideas, domain expertise, and high-quality training data - not just algorithms.
• 💰 Resource Constraints: Computational costs are a significant factor in GenAI development, often underestimated by founders.
• 🎈 Hype vs. Reality Gap: Many founders lack a deep understanding of the technical challenges and limitations in GenAI implementation.
• 🖥️ Infrastructure Costs: Even minimal GPU requirements for model training and inference can be daunting for bootstrapped startups.
• ⚖️ Resource Optimization Fallacy: Attempts to minimize costs across all aspects of development often lead to suboptimal results in model performance and product quality.
• 🏎️ Performance-Aesthetics Mismatch: Many startups focus on creating visually appealing UIs but struggle with the underlying AI engine's capabilities, resulting in a "sports car body with a scooter engine" scenario.
• 🚀 Democratization vs. Expertise: While AI tools are becoming more accessible, creating truly groundbreaking GenAI applications still requires deep technical expertise and innovation.
• 🌊 Depth vs. Breadth Trade-off: Founders who aren't willing to invest time and resources in deep technical development risk creating superficial, easily replicable products with limited longevity in the market.

Keep Exploring!!!

🚀 Celebrating a Year of GenAI Use Case Success in Retail! 🎉

It's been over a year since my Retail adoption #GenAI use case went live for a leading U.S. specialty retailer on August 17, 2023. 

The results? Double-digit improvement in user page conversions! 📈

🔍 Key Insights:

Innovation Over Integration: Initially, I didn't showcase #GenAI use cases to the #CTO. Instead, I presented a #ReimaginedWorkflow for:

• Customers
• Support analysts
• Procurement teams
• Digital Asset Management

Rethinking AI/ML Implementation: It's not about wrapping AI around existing processes. True impact comes from:

• New engagement models
• Innovative interactions
• Blending creative solutions

Success Factors in Production:

• Data Quality 🏆
• Innovative use of GenAI
• Tailored solutions (not patchwork fixes)
• Rigorous testing in production environments

Beyond Demos: Real adoption comes from solving genuine user needs, not just showcasing capabilities.

• 💡 Lesson Learned: To make #AI truly impactful, we must explore new ways of engagement, foster creativity, and innovate by combining diverse ideas.
• 👥 Collaboration Opportunity: Are you working on similar AI strategies or use cases? Let's connect and collaborate! Share your experiences in the comments.

#AIStrategy #RetailInnovation #DataDrivenDecisions #DigitalTransformation #AIAdoption #TechLeadership #InnovationInRetail #AISuccessStory

Who else is seeing success with GenAI in their industry? Let's discuss! 👇

🚨 4 Overlooked Pillars of AI Ethics: Profits vs. Principles 🚨

In the rush for AI dominance, we're neglecting crucial ethical foundations. Here are 4 pillars often sacrificed at the altar of profit:

These policies will shape the future of AI. We can't wait - the time to act is NOW.

🧠 Data Sovereignty Rights

• Overlooked because: User control cuts into valuable data assets
• Reality check: Our digital lives aren't corporate property

🏥💰 Regulatory Oversight in Critical Sectors

• Overlooked because: Slows time-to-market in lucrative industries
• Reality check: Unchecked AI in healthcare/finance risks lives & livelihoods

📚 Transparent AI Lineage

• Overlooked because: Transparency can expose flaws & biases
• Reality check: Black-box AI erodes trust & accountability

🎨 Recognition of Human Creativity

• Overlooked because: Crediting sources complicates IP claims
• Reality check: AI thrives on human ingenuity – let's honor that

These aren't just ethical niceties – they're essential for sustainable, trustworthy AI. Short-term profits shouldn't overshadow long-term societal impact.

We need AI that serves humanity, not just shareholders. It's time to realign our priorities. Which pillar do you think is most crucial? Why is it being overlooked?

#AIEthics #ResponsibleAI #TechForGood #AIAccountability #DigitalRights

Tag a tech leader who needs this wake-up call 👇

Keep Advocating ResponsibleAI!!!

TechForGood vs TechForProfits !!!

Understanding Index RAG: Data Storage vs. Retrieval

In the realm of information retrieval and artificial intelligence, Index RAG (Retrieval-Augmented Generation) has emerged as a powerful technique. To fully grasp its potential and limitations, it's crucial to understand the distinction between data storage and retrieval, particularly in the context of indexing strategies. This post will explore two different indexing approaches and their implications for handling queries, especially multipart questions.

The Indexes

Index 1: Broad and Diverse

Composition: 20 pages from history + 20 pages from geography + 20 pages from maths

Strengths:

• Versatility: Covers multiple subjects, enabling efficient responses to multipart questions
• Diversity: Offers a well-rounded breadth of content across different fields

Index 2: Deep and Focused

Composition: 200 pages focused solely on history

Strengths:

• In-Depth Knowledge: Provides comprehensive depth on history, ideal for complex historical inquiries
• Rich Content: More pages dedicated to one subject increases potential for detailed responses

Trade-offs

Breadth vs. Depth

• Index 1: Offers breadth across subjects but may lack depth for in-depth analysis
• Index 2: Delivers depth in history but falls short on breadth for interdisciplinary queries

Complexity of Queries

• Index 1: Can handle complex, multipart questions effectively due to subject variety
• Index 2: May struggle with multipart questions spanning multiple disciplines

Information Quality

• Index 1: Information may be less densely packed with specialized detail
• Index 2: Provides rich historical data but lacks subject diversity

Challenges with Multipart Questions

Consider a multipart question involving history and mathematics:

Using Index 1:

• Pros: Can provide relevant information across both subjects
• Cons: Detail may not be as profound, potentially leading to surface-level insights

Using Index 2:

• Pros: Historical aspect might be well-covered
• Cons: Absence of mathematical content results in an incomplete answer

Implications for RAG Systems

Query Processing:

• RAG systems using Index 1 may need sophisticated algorithms to balance information from different domains
• Systems using Index 2 might require additional steps to supplement missing interdisciplinary information

Content Generation:

• Index 1 allows for more flexible content generation across topics
• Index 2 enables deep, nuanced responses within its specialized domain

System Architecture:

• Index 1 might benefit from a modular architecture that can efficiently combine information from different subjects
• Index 2 could leverage specialized language models fine-tuned for historical content

Conclusion

The choice between a broad, versatile index (Index 1) and a deep, focused index (Index 2) significantly impacts the retrieval effectiveness of an information system. Understanding these dynamics is crucial for users and developers alike to create effective RAG systems.

When designing or using RAG systems, consider:

• The nature of expected queries (single-domain vs. interdisciplinary)
• The required depth of information
• The system's ability to synthesize information from multiple sources

By carefully weighing these factors, one can optimize the balance between data storage and retrieval capabilities in Index RAG systems, ultimately enhancing the quality and relevance of generated responses.

GenAI Two Use Cases - Two Lessons

Creative and Learning Use Case

Instead of playing sports or watching cartoons, kids nowadays are coding. 👀 pic.twitter.com/POX50bt4MH

— Vivek Naskar (@vivek_naskar) September 8, 2024

Wrong Guardrails Applied, Content for opinions

What other options

• Provide Factual data
• Do not provide recommendations for entities
• Reason for bias
• Do not rely on Guardrails

Keep Going!!!

Multimodal data platform

ETL and data pipelines are redefined in #GenAI Applications. Your #ETL now will support 

• #images, #docs, #numbers, #pdfs. Extracting and storing insights/vectors / structured databases 
• Everything together creates the new #GenAI #Multimodal data platform. 
• #Multimodal #insights from all forms of data
• #Proplens is working to align this perspective for our customers for richer insights and perspectives. #productlessons #genai #data

Keep Loading and Learning!!!

Navigating the Tradeoff Between Income and Responsibilities in Freelance AI Consulting and Corporate Jobs

• My friend - Hey Siva, you seem busy. 
• Myself - Yes, I have some classes and consulting. 
• My friend - So, you're earning more money? 
• Myself - Not necessarily. Some leads work out, some don't. Sometimes, even after presenting the architecture, there is no convergence. 
• My friend - That's common. 
• Myself - The same uncertainties exist in a corporate job, where ideas may not align. However, you get to work closely with founders. 
• My friend - That's true. 
• Myself - Even if your idea fails in consulting, you have a direct connection to lead, discover, and strategize. 
• Myself - In Freelance AI consulting, the outcome is clear - it either works or it fails, and I deal with it directly. No regrets :)

Freedom entails risks, but it's worth it. You pave your own path.

Sure, feel free to ping me if you are interested in harnessing the power of AI.

Keep Exploring!!!

Intelligence with GenAI

When learners can see 'Intelligence with GenAI'. It is very heartening to see solutions built during the session :)

Some feedback after the 12-hour GenAI session:

• We have a lot of data; we can use LLM to add intelligence. A lot of IoT sensor data is present, but only fixed reports are available.
• I have only used #ChatGPT; now I see a ton of tools to explore.
• Infra intelligence - LLM is used to understand the complete ecosystem of servers, leverage logs + intelligence to find server owners, and alert on reallocation / upgrades.
• Automate Data Analyst work to analyze keywords, user query patterns. This saves 65% of my efforts.

Keep Exploring!!!

More Smarter Agents + LLMs = Superior Intelligence

 

Keep Exploring!!!

Latency is a never ending learning

• Latency with cache
• Latency with semantic cache
• Latency with Indexes
• Latency with Graph Queries
• Latency with Optimal Value for Top K

Benchmark against domain dataset

Good Data = Good Strategy = Quality Experiments

Happy Low Latency!!!

How to avoid this scenario - Gartner Predicts 30% of Generative AI Projects Will Be Abandoned After Proof of Concept By End of 2025

What is required to Turn Data Into AI Products ? - My perspectives

• Experimentation on Different levels of data / Summary / Key-Values / Multimodal
• Data curation / Setting up Relationships
• Adding domain knowledge

The main reasons cited are:

• Poor data quality
• Inadequate risk controls
• Escalating costs
• Unclear business value

So, the lesson is, my perspectives are:

• The model is not lift-and-shift — customize it for your needs. A demo that works may not be the solution you need.
• Build your data and benchmark with your data. Do not rely on benchmarks that do not reflect your data.
• Have an LLM cybersecurity, data governance, and guardrails in place. Do not trust the LLM until your first 100 users are happy with it.
• Escalating costs—first, get the accuracy right, then reduce costs. You cannot achieve everything at once.
• Unclear business value — Do not force-fit LLM use case to get a promotion. Only opt for it if it genuinely adds value.
• If someone promises a working solution in 1 month remember it can be selling a prototype, not a production-grade solution

Keep Exploring!!!

Working as AI Advisor

• Close proximity to decision-makers
• Tech will be commodity but Domain + Tech is true knowledge
• Loyalty works in long term, talent gets visibility in short term
• Take time and provide accuracte inputs, Optimism + Clear perspectives matter
• Building relationships, Experimenting on meaningful projects, Novelty to build solutions / integrate past experience + latest tech to match quality matters
• Balancing business vs tech roadmap, Being able to spot features that differentiate the rest
• In the end everything may fail but there is a satisfaction in the Journey

Keep Exploring!!!

Painful Moments - Potential People vs. Wrong Mindset

While evaluating answers: Some candidates document well, attempt, and submit answers but miss the basics. This reflects both intent and missed guidance in learning. High potential is evident, but basics are either overlooked or dot-connecting skills are lacking.

While teaching: Some PhD/lateral folks tend to generalize everything or focus on proving theories break. Learning is not about proving your knowledge but about gaining a balanced perspective. One class is not sufficient to judge anything. Observing these types of learners makes me feel sad as they are so short-sighted.

Education is not mindset; experience does not mean competency!!!

Keep Exploring!!!

Memorization vs Generalization

Memorization vs Generalization

When you develop #GenAI apps, After a certain stage, When things work fine, The immediate next question is

• Model is memorizing or Learning patterns
• Test with variations / Analyze on patterns of responses

I don't want my life to be memorization - Company1 - Company2 ..., Exploring out of comfort zones provides diverse perspectives.

Earlier I had time to regret, Now I don't have time to think about anything. A long day of managing and solving different problems and different lenses of execution. Sometimes some experiences don't fill your pocket but fill your soul. In the end, I want to smile at death, I have tried all my wishlists.

Keep Exploring!!!

GenAI - LLM - Startup Learning War

 

Latency Experiments - Link

Keep Exploring!!!

The Uniqueness of Customer Data and the Importance of Automated Cleanup

In the digital age, managing customer data can be a daunting task. Here are some points to consider:

• Each customer's data will be unique.
• Each customer's data will encounter parsing issues.
• Each customer's data will disrupt pipelines.

No data is pristine; changes should not be spontaneous. When dealing with data, especially on a large scale, it's imperative to have a robust process in place. Having a process to automate cleanup is essential to scaling your solution.

Keep exploring!

KG vs RAG

• First, we solved with Prompt
• Next, we solved with RAG
• Next, we did Summary, RAG
• Next, we moved to Txt2SQL

There is some leftover space for Graph

Build a product and use tech according to needs, No Forefit in the equation :)

Keep Exploring!!!

Learning is Non-Linear

Some days are Sigmoid. Some days are Relu. Novelty is not inventing new stuff but stitching the right techniques in the right proportion.

A few more I relate to my work style :)

• Persist until you figure things out, digging deeper and deeper into unfamiliar systems until it finally clicks, and you understand how they work.
• Have a broad understanding and creative use of Algos, SQL, LLMs, etc. Teach until you get first principles right
• Have the intellectual curiosity to learn about the many new areas you’ll be taken into. Failures today are arsenal for tomorrow.
• Have a willingness to lean into uncertainty. It's okay if a bad day happens.
• Magnitude improvements and novel capabilities. Keep collecting perspectives.
• You will face unique engineering and AI challenges that will make a meaningful impact. To be authentic, you have to have a lot of scars.
• A focus on outcomes, not time-tracking. Yeah, sometimes easy looks complex, complex looks easy.
• The database is always important. Are not afraid of legacy justice systems and would rather make them work than give up.

Keep Going!!!

Txt2SQL

Txt2SQL is easier in straightforward examples, Real database has a ton of complications

Example-

Columns can be generic, Attribute1, Attribute2, We may use Attribute1 for key, Attribute2 for Value. A ton of learning working on it, still trying to get a hold :)

• Hints for clause level
• Hints for join 
• Verbose hints
• Examples in database specific formats
• Categorize on type of errors - Syntax, Missing columns, Missing right joins

Keep Exploring!!!

LLM perspectives

• LLM - Fast, Simple, and Dumb Sometimes
• Prompts - Looks simple but when structured relevant to context can give magical results
• With extended context length, embedding it looks more magical with the abstraction of representations of useful knowledge

LLM generation kids / learning using LLM products will have a different perspective of thinking / before and after ChatGPT :)

Experimentation, Always something to find a solution :)

Some questions and answers take days or weeks, and sometimes the approach moves from LLM to NLP, It's a blend of techniques to make things work.

• How do we optimize RAG with internal documents, Original vs Summary vs Intents, What works best? 
• How do we merge external data? How can we keep versions and relevance?
• More than LLM work, The heavy lifting is for Data preprocessing/cleaning / Embedding on summary 
• When to use LLM vs Multimodals?
• What is the benchmark for our domain and how much do we meet it consistently?
• The transition for LLM, LLM+KG, Creating the data mapping..

A lot of challenges but one at a time, Balancing Consistency, Accuracy, and Latency. If you want to solve real problems you can connect/explore potential learning experimentation opportunities / dedicate some learning hours. Please drop a note to career@proplens.ai

#learnings #NLP #Datascience #RAG #LLMs #perspectives #Datascience 

Keep Exploring!!!

RAG and Prompts - Learning Evolves

In one particular use case, it's a constant process of experimentation and iterations.

• Step #1 - Let me try with prompts - It works but not consistently
• Step #2 - Let me try with a vision prompt - It takes time. 
• Step #3 - Let me merge everything in the database and check. 
• Step #4 - Routing takes 2 seconds, querying takes 2 seconds, and prompting takes 2 seconds. 
• Step #5 - Let's keep accuracy and latency separate, divide everything into separate tracks, and sort out the basics.

Some successes, some lessons, and some learning.

Keep Exploring!!!

Success and Failures in Pushing Ideas in Startups vs Matured Companies

Over the past 4 months, I've been working with really small teams, and the difference in communication dynamics compared to larger teams has been striking.

In my previous roles in product and consulting, I achieved success but with a considerable amount of effort spent convincing and negotiating with numerous people. Here are some of the specific challenges faced when pushing ideas in larger, more mature companies compared to nimble startups:

Increased Lines of Communication: With more team members in mature companies, ensuring everyone is on the same page becomes significantly harder. There's a higher risk of changes in approach, iterations, feedback, and information being lost or mistranslated as it travels through various levels. In contrast, startups often have flatter structures, making communication more direct and less prone to distortion.

Slower Decision-Making Processes: Larger teams often have more layers of approval, which can slow down decision-making. Every stakeholder has their own priorities and concerns, adding to the complexity. Startups, with their smaller teams, can often make decisions more quickly, which allows for faster iterations and innovation.

Greater Need for Consensus: In smaller teams typical of startups, reaching a consensus or getting buy-in for new ideas is often easier. Larger teams in mature companies require more effort to align everyone's visions and goals. This can lead to lengthy discussions and compromises, which may dilute the original idea.

More Stakeholders to Convince: Larger teams come with more stakeholders, each with their own perspectives and interests. This multiplicity can make it challenging to get everyone on board with a new idea. Startups, on the other hand, usually have fewer stakeholders, and the founders or key decision-makers are more accessible, simplifying the process of getting buy-in.

However, the journey you take, whether in a startup or a mature company, will reward you for the risks and decisions you choose to travel with. Each environment has its own set of challenges and rewards, but understanding these dynamics can help in navigating them more effectively.

Keep Going!!!

Data Privacy vs Cheap value vs Compromises

How your data is bought at Low costs and leveraged to build ML Models !!!

Keep Exploring!!!

LLMs - Production vs Prototype

 #LLM is #Cool! #PromptEngineering is great. Demonstrating a prototype is beneficial, but for a production application, preparing data and consistently getting Top K results are essential. The ability to handle diverse queries, manage large datasets vs. remembering minimal key data are significant trade-offs and design choices that come with domain knowledge. Building a working solution is easy, but achieving predictable, low-latency, and consistent behavior requires constant iteration and evolution. #Perspectives and #AppliedLearning are key. #AI #MachineLearning #DataScience #Prototypes #ProductionReady #DataPrep #Latency #Consistency #Iteration #

Keep Exploring!!!

Why big companies failed to build LLMs

 

How Alexa dropped the ball on being the top conversational system on the planet
—
A few weeks ago OpenAI released GPT-4o ushering in a new standard for multimodal, conversational experiences with sophisticated reasoning capabilities.

Several days later, my good friends at PolyAI…

— Mihail Eric (@mihail_eric) June 11, 2024

• Data was poorly annotated. Documentation was either nonexistent or stale.
• Experiments had to be run in resource-limited compute environments. 
• That meant for months our internal annotation team had been mislabeling thousands of data points every single day
• antagonistic mid-managers that had little interest in collaborating for the greater good
• Duplicated efforts due to no shared common ground 

Keep Exploring!!!

LLM Perspectives - Familiarity vs Randomness

 

There's a big difference between solving a problem from first principles vs applying a solution template you previously memorized. It's like the difference between a senior software engineer and a script kiddie that can't code.

A script kiddie that has a gigantic bank of scripts… https://t.co/l2W8FUnwj8

— François Chollet (@fchollet) June 10, 2024

Keep Exploring!!!

Next Real - Startup - Healthcare + AI

 

Keep Exploring!!!

The Power of Titles: A Strategic Path to Influence and Impact

 My take from video - Link

Phase I: The Importance of Titles

• Influence with Titles: Titles matter because they shape perceptions and open doors.
• Decision-Making Power: Titles matter because they provide authority and credibility.
• Risk-Taking: Titles matter because they create platforms for bold actions.

Phase II: Embrace Risk on the Road to Titles

• Strategic Risk taking: Leverage every opportunity to step into roles that accelerate your journey to influential titles.
• Proactive Role Selection: Choose roles that enhance visibility and pave the way to coveted titles.

Phase III: Deliver Impact, Titles Will Follow

• Impact-First Approach: Focus on making meaningful contributions; impactful deliverables naturally lead to recognition and prestigious titles.
• Purpose-Driven Path: Prioritize routes that enhance your impact; the right titles will follow in recognition of your value.

Keep Exploring!!!

Transfer Learning Notes

Teach the process/approach for students to get better clarity :). My answers for below question 

 When does Transfer Learning work?

Keep Exploring!!!

Humanoids

#humanoids = #llm powered decisions + #robotics + custom models trained for specific tasks, Build custom models to suit specific needs https://t.co/sxcaqCk46U

— Siva (@sivaram2k10) May 12, 2024

 

Keep Exploring!!!