DEV Community: Sarthak Sharma

Either You Die a Developer, or Live Long Enough to See Yourself Become a Product Manager

Sarthak Sharma — Wed, 24 Sep 2025 12:22:28 +0000

📌 Read here: For a slightly better reading experience Sarthology.com

I'm not implying that Product Managers are bad or that they are not good at their job. Although I have seen my engineer friends hating on Product Managers and Testers all the time 😄. Being a Product Manager with a technical background, I have seen both sides of the coin. Luckily enough, when I was a tech lead, my PM also had a technical background. Instead of tension, there was mutual respect and even inspiration.

If you ask me, I have a perspective on it. So if a Developer is a Pokémon, they might evolve into a Manager and later a Technical Lead and then a VP of Engineering—if they like being founders too. But this doesn’t mean all of us will, or even should.

Then here comes the AI and the irony…

If you think about it, vibe coding has not only shifted the tech landscape but also turned the tables, hasn’t it? Suddenly you might find yourself in the shoes of a tester 😅, and writing a good prompt feels exactly like writing a detailed user story.

Working Feature = Well-Defined Spec x Rigorous Testing

You are now wearing three hats: Product Manager + Tester + Developer. 🤯

🌓 The Changing Tech Landscape

I don't want to be Baba Vanga or something, but the tech job exodus is on the horizon now. When my old mentees ask me for advice, I tell them that they're great developers, but it's time for the Pokémon to evolve. I tell them to develop product sense. But the question is—can everyone adapt to it?

If I have to categorize developers, I would put them into three broad categories:

1. Product-Minded 🧙🏻:

Signature traits: You naturally think like a user. You write stories before code. You notice friction, propose better flows, and often pre‑empt bugs with thoughtful tests.

For example: You may have seen the OG OSS developers—yes, I'm talking about them. They relate to a problem at a user level. Most OSS developers (from my experience) are trying to solve a problem for themselves. So planning comes easy; defining user stories is easy. A few of them have good design sense, and they end up creating loved and viral products. Not just solving problems, but solving problems with style. They are like Michelin‑star restaurant owners, baking a good experience along the way.

🤔 What it looks like at work: You push a story from “works” to “feels right.” Testers rarely find surprises.

✨ Why this thrives now: AI supercharges people who give it crystal‑clear specs.

🚀 Trajectory: PM/Tech Lead comes naturally—you’ve been doing the job without the title.

2. Problem Solvers 🥷🏻:

Signature traits: You love hard problems, patterns, and clean solutions. LeetCode is your playground. UI polish isn’t always your first instinct, but you can flex when needed.

🤔 What it looks like at work: You crush tricky tasks, sometimes under‑invest in UX.

💊 Hard pill: If you don’t evolve into clear spec + user lens, you’ll be arm‑wrestling AI that’s getting better at pure problem‑cracking.

🚀 Trajectory: With a bit more user empathy and end‑to‑end testing, you’ll drift into Category 1.

3. Job Holders 👻:

Signature traits: You ship volume but miss acceptance criteria, hand off bugs to QA, and treat stories as tickets rather than user outcomes.

Sadly, these developers are more in number, thriving in big corporate environments. They don’t care to read user stories and often miss the acceptance criteria. These are basically junior developers with fancy job titles, which they got just because of their time in the industry. They have no motivation to evolve and are sadly milking big cash because they have been like this for 5+ years. God knows why companies value numbers over quality. They are valuable because there is an ecosystem that supports them: bad coder → more testers → longer product timeline → more money to milk from the client. So everyone wins except the client of these companies (sorry for my rant there 😅 and also I acknowledge some time bad leadership also lead to this practice).

🤔 What it looks like at work: Projects stall, cycles stretch, and quality debt grows.

🤷🏻 Reality check: This model won’t survive long.

🚀 Trajectory: Start with specs, acceptance criteria, and your own test checklist— or themarket will do the sorting for you.

If you never stop learning, you’ve already tilted the odds in your favor.

Evolving is part of human existence, but today it has become more than that—it’s about survival.

So where to start your journey....

All of this assumes you can read and write code with semantic clarity. Fundamentals stay non-negotiable; product thinking is an extra layer, not a substitute.

📜 Spec‑Driven AI Tools (Why ‘Vibe Coding’ Works)

Vibe coding: using AI with rich context and clear specs so models produce predictable, testable output.

AI performance rises or falls with how clearly you define the problem. And that’s not just for coding—check my article on context engineering (for prompting win with AI).

But this mindset might not come easy for everyone, and that’s why vibe coding tools are themselves shifting your attention there.
A few notable ones include:

Kiro (by Amazon): focuses on specs‑first development, pushing you to think about requirements clearly before writing code. AI can draft specs for you, but you still need to review them to avoid hallucinations.
Agents.md: helps structure prompts and tasks in a systematic way.
Spec Kit (by GitHub): an open‑source toolkit for spec‑driven development. Read more about it here.

Pick one tool and run a 1‑sprint pilot. Start with a 5‑minute spec template, then compare bug counts and review cycles before/after.

✨Tips

Product‑Minded → formalize with Spec Kit;
Problem Solver → narrate user acceptance with Agents.md;
Job Holder → enforce acceptance criteria with Kiro before coding.

🧠 Developing a Product Mindset

Some may say a product mindset is an inbuilt feature for most people, but I disagree. With hard work, everything can be learned and mastered.

The first step is simple: OBSERVE. Every day we interact with countless apps and products as users. Your learning can start there. These apps are filled with hidden gems. Ask: Why do you love a product over its competitors? What feature do you love and why? What design choices are ingenious? That’s where you develop a sense of what might work and what doesn’t.

Observe → Frame → Test → Ship → Measure

1️⃣ Observe: Screenshot two moments you loved or hated in apps you used today and write one sentence on why.

2️⃣ Frame: Turn each into a problem statement and a desired outcome.

3️⃣ Test: Draft acceptance criteria (Given/When/Then) so “done” is unambiguous.

4️⃣ Ship: Build the smallest coherent change that proves the idea.

5️⃣ Measure: Pick one HEART metric (Happiness, Engagement, Adoption, Retention, Task success) and watch it for two weeks.

Lastly, build an OSS of your own—a problem personal to you and a hinge that others want a tool or app for. Launch it on Product Hunt and ask users for feedback. In no time, you will develop the product mindset—and who knows, you might build a business around it too.

Principles to steal: Start with the problem. Think big, start small, learn fast. Ship often.

📌 The Product‑Minded Dev Playbook (6 Steps) 📚

Here is a small playbook to develop with a product mindset 👉

User story: Who is the user? What pain? What job‑to‑be‑done?
Acceptance criteria: Given/When/Then, including edge cases.
Tests first: Red → Green → Refactor. (Even one tiny test!)
Generate: Ask AI for the smallest implementation to satisfy the tests.
Review: Read the diff like a PM—does this match the intent?
Ship & measure: Add analytics or a measurable proxy for value.

📌 Two Pocket Checklists 📋

Checklist to keep in mind, if you want create prefect spec for AI:

👉 Spec checklist

User → Problem → Outcome in one sentence.
3–5 acceptance criteria with edge cases.
Non‑goals (what this won’t do).
Observability (how we’ll know it works).
Rollback plan (what if it fails?).

Checklist if you are working on a small feature and prefer vibe coding:

👉 Prompt checklist

State role and constraints (language, style, deps).
Provide examples (tests, input/output).
Ask for a diff or single file, not essays.
Request assumptions and uncertainties explicitly.
Cap scope (“only implement X to make tests pass”).

📌 Product‑Minded Shifts (Mindset Upgrades) 🧠

From output to outcomes: “I closed 5 tickets” → “I reduced signup drop‑off by 12%.”
From features to jobs: Tie every PR to a user job‑to‑be‑done.
From perfect to iterative: Ship a walking skeleton, then layer quality.
From solo to orchestration: Treat AI/agents as teammates.
From intuition to measurement: Add a metric to every spec.

✨ An Optimistic Future

The safest path is evolution: widening your impact beyond code so your career compounds.

🤞 Here’s why I’m optimistic:
Developers have a tactical advantage in this age of AI. If they understand the product, test it, guide the AI to fix an issue in code—or fix it themselves—they can be a solo force‑multiplier. They will ship better products, faster, and create more jobs in the market. SaaS will be built and iterated like fast fashion—short cycles, rapid feedback, constant remix.

Open source and local‑first are accelerating; smaller teams can ship meaningful fixes faster. Pick one OSS you use, close a small bug this month, and document before/after in a 5‑line changelog.

Evolve your lane: formalize specs (Product‑Minded), narrate acceptance (Problem Solver), or enforce criteria (Job Holder)—then ship.

💪 30‑Day Challenge: Become a Product‑Minded Dev (While Still Coding)

Week 1 – Pick one user pain; write a crisp spec; ship a walking skeleton.
Week 2 – Add tests, observability, and one measurable outcome.
Week 3 – Run a mini user check (1–3 users/stakeholders); revise spec; ship v2.
Week 4 – Write a short retro: what moved the needle? What didn’t? Publish it.

May the force be with you.

Last remark

I’d love to hear your story—did you drift toward product, stay deep in code, or carve a hybrid path? What skill actually moved the needle for you? Drop a comment 👇

P.S. If you want an advice shoot me dm anytime @sarthology, or I do a few 1:1 sessions each month,Check out that here

See you next time. Until then.

Evolve, don’t wait. ⚡️

The DeepSeek Disruption: AI, GEOPOLITICS AND Future of Open Source👑

Sarthak Sharma — Fri, 31 Jan 2025 05:50:39 +0000

Almost a year ago, I wrote an article "Rise of Local LLMs?", which was well received. Now feels like the perfect time to revisit this idea—especially after recent events that have shown open-source AI overtaking proprietary models.

Chapter 1: AI, Global Politics, and the Battle for Control 🏛️🤖

In many ways, AI today is being treated like the atomic project at Los Alamos. If you’ve read "Supremacy" by Parmy Olson, you’ll see a familiar pattern—a group of researchers driven by the belief that they are "saving the world."

The difference? This time, it’s not governments racing to build the next atomic bomb—it’s corporations. And if history has taught us anything about the internet, it’s that money matters more than anything else. 💰

Tech giants are striving for full control. Imagine a world where only the U.S. had nuclear weapons. Would history have played out the same way? Now apply that thought to AI.

The first prototypes of any technology are always expensive and resource-intensive. But eventually, others catch up and make it cheaper and more efficient. We’ve seen this happen with space travel (🚀 SpaceX, ISRO) and now with AI—DeepSeek is proving that local models can challenge industry giants. And this is just the beginning.

DeepSeek’s Checkmate ♟️

If you haven’t been in a coma, you’ve probably heard of DeepSeek by now. But what wiped out $1 trillion from the U.S. stock market wasn’t just that it hit #1 on the Apple Store. 📉

It was something far bigger:

DeepSeek became the first open-source reasoning model. 🏆

We've seen the U.S. ban TikTok and label it a "Chinese spy app," but DeepSeek is different. If you don’t trust it? Run it locally.

Still worried about it being a Chinese spy tool? Read the research paper and train your own model.

The real shocker? Companies like OpenAI, Google, and Microsoft have spent billions on AI development, but DeepSeek’s success proves that state-of-the-art AI can be built for under $10 million without needing NVIDIA’s expensive chips. That changes everything.

What’s Next? ⏳

OpenAI has already accused DeepSeek of using its models to train their AI. That drama is just beginning. Expect to see the U.S. try everything to discredit DeepSeek’s achievement in the coming weeks.

But here’s the bigger picture:

The open-source AI community is thrilled, with notable figures like Yann LeCun from Meta and organizations such as the Linux Foundation praising the shift toward decentralized AI. Many developers and independent researchers have taken to forums and social media, celebrating the new wave of AI democratization and the potential it brings for global collaboration. We’re about to see more efficient, cheaper, and accessible models in the coming months—mark my words. 🔥

Just recently, Alibaba dropped another powerful open-source LLM:

🚀 Qwen 2.5 Max

And Moonshot AI released:

🌙 Kimi Ki 1.5

“There will be another DeepSeek by the end of this month, and many more this year in the U.S. and worldwide,”

— Jim Zemlin, Executive Director of the Linux Foundation

This is the new normal. Open-source AI isn’t going anywhere. And honestly? That’s how it should be. 🏆

The Revolution Isn’t Coming… It’s Here. 🚀

The rise of local LLMs is inevitable. Governments and corporations may try to control AI, but the open-source community has already proved that AI belongs to everyone.

Another side of the debate

WARS and AI ⚔️🤖

In "Supremacy" by Parmy Olson, the DeepMind team once feared that Google might use their AI technology for the U.S. military. At the time, there was a company-wide protest, and the project was eventually shut down.

Fast forward to today: amid the Israel-Gaza conflict, reports suggest that Google has provided AI assistance to the Israeli military. Meanwhile, AI is also being used in the Ukraine-Russia conflict, with both sides leveraging AI-driven strategies.

The reality is clear—there is no stopping AI from being used in warfare. We are already living in that future.

Check this Video for more on that

This is yet another reason why technologies like these should never be controlled by a single nation or military power. Open-source AI ensures that knowledge and advancements are distributed, rather than being hoarded for geopolitical dominance.

Chapter 2: Why Open-Source AI Models Running on Your Hardware Are the Future? 💻🚀

Privacy is the obvious reason to run AI models locally, but let’s be real—billions of users willingly share their data on Instagram, YouTube, and TikTok every day. So, it’s safe to say most people don’t actually care about privacy anymore.

But that’s not the only reason local AI is the future. Let’s explore some key factors.

1. The Cost Factor 💰

According to this article, here are the most common use cases for ChatGPT:

📚 Homework Assistance
💬 Personal Advice & Communication
🌍 Travel & Lifestyle
🔞 Sexual Content (Failed attempts 🤦‍♂️)
🖥 Coding Help

Apart from the last one (coding), do you really need a state-of-the-art reasoning model for these tasks? Absolutely not.

Instead of paying for expensive AI subscriptions, you’d be better off using smaller models locally with tools like Jan AI—completely free!

If your goal is just to get horny with AI, for god’s sake, at least use a fine-tuned model built for that. 😂

Save Money on AI Coding Assistants 🧑‍💻

For coding, DeepSeek R1 is now available to run locally, making it a solid alternative to GitHub Copilot or Cursor. That means you can cancel your subscriptions and still get top-tier AI assistance.

🚀 Check out this Dev.to article to set it up in Visual Studio Code—or, if you're feeling adventurous, build your own AI plugin in under 5 minutes!

2. The Environmental Impact 🌱⚡

Did you know that a single ChatGPT query consumes 10-100x more energy than a locally run LLM response? That’s because cloud-based AI models rely on massive compute clusters running 24/7.

Meanwhile, running a quantized 7B LLM on your laptop consumes just 20-50W—about the same as browsing the web.

To put it in perspective, here’s a side-by-side comparison:

🔋 Energy Consumption

Factor	Local LLMs 🌿	Cloud-Based LLMs ☁️
Compute Power	Runs on consumer hardware (low power)	Requires massive data centers
Efficiency	Optimized for single-user inference	Thousands of power-hungry GPUs
Scalability	Less scalable but efficient	Highly scalable but wastes idle compute
Carbon Footprint	Lower (if optimized)	Very high due to cloud operations

🌍 Which One is More Sustainable?

Factor	Local LLMs (Edge AI) 🌿	Cloud-Based LLMs ☁️
Energy Efficiency	✅ Lower power use	❌ High due to data centers
Hardware Sustainability	✅ Uses existing hardware	❌ High GPU demand & e-waste
Data Privacy & Network	✅ No internet needed	❌ Heavy data transfer
Carbon Footprint	✅ Lower	❌ Higher (always-on servers)

Read more on AI’s environmental impact here

So next time you use a cloud-based LLM just to role-play as a medieval knight or write a poem about your ex, maybe think twice. 😏

3. Speed & Latency 🚀

One of the biggest advantages of local AI models is speed. When you run an LLM on your own hardware, the response time is instantaneous because:

✅ There’s no internet latency – no data needs to travel back and forth to a remote server.

✅ You bypass API limits – no rate-limiting, no throttling.

✅ Your AI responds as fast as your hardware allows – no waiting on cloud server congestion.

Building Small Apps running on consumer hardware instead of relying on Cloud Apis is much faster and cheaper.

Final Thoughts

Between cost savings, privacy, and sustainability, running AI locally on your hardware isn’t just a geeky alternative—it’s the future.

I know it’s exciting, but many will argue that hardware requirements are still a bottleneck—and I won’t deny it. Not everyone can afford high-end GPUs to run these models offline. But hardware is evolving fast, and soon, this will no longer be a limitation.

Take Apple’s new Mac Mini M4—it packs an impressive AI-capable chip at a reasonable price. Meanwhile, NVIDIA is working on making its chips more efficient and accessible, knowing that the future of AI isn’t just in data centers but in everyday devices.

Soon, AI won’t just be something you access through a cloud API—it’ll be embedded in basic laptops, phones, and even IoT devices. On the other side, LLMs are becoming more optimized for today’s hardware, making it possible to run powerful AI without needing a supercomputer.

The DeepSeek story proves that the future isn’t about a single dominant AI or an AGI (Artificial General Intelligence) that does everything. Instead, it’s about specialized models working together—where you can pick and choose the best models for specific tasks and build applications on top of them.

We are entering the era of multi-modal AI systems, where applications will combine multiple specialized AI models to create something far more powerful than a single, monolithic AI.

Beyond that, we're seeing:

On-device AI acceleration 🚀 – Smartphones, gaming consoles, and edge devices are integrating dedicated NPUs (Neural Processing Units) to run AI locally with minimal power consumption.
Open-source hardware initiatives 🏗️ – Projects like RISC-V AI chips are working toward affordable, open AI processing units that break dependency on proprietary silicon.
Federated AI 🌍 – A system where AI models train across multiple decentralized devices, improving efficiency while maintaining privacy.
Energy-efficient AI ⚡ – The rise of low-power AI models designed to run on minimal hardware while achieving state-of-the-art performance.
AI compression & quantization 📉 – Techniques that significantly reduce model size and computational requirements, making high-quality AI accessible on everyday devices.
Community-driven innovation 🤝 – Open-source developers constantly improving and sharing models, ensuring that no single corporation can monopolize AI.

Upcoming

In the next article, we’ll explore:

🔥 Where to get started with local AI models

💡 Ideas for your own Open-Source AI project using offline models

Isn’t this exciting? What’s your take on local AI? Let’s discuss in the comments! 🌟 Do you see yourself experimenting with local AI models? Have any cool ideas in mind? Drop a comment below—I’m all ears!

If you want to discuss a project idea, need guidance, or just want to chat about AI, hit me up on Twitter @sarthology. Let’s build something awesome together! 🚀

Stay tuned. 🔥🚀

How AI Is Transforming Code Learning—for Better and for Worse

Sarthak Sharma — Sun, 26 Jan 2025 06:41:55 +0000

It all began in my third year of college, back in 2013, when I started giving free lectures to my classmates and juniors. Since then, I’ve mentored more than 150 developers. My goal has always been to ignite a passion for coding in young minds—helping them realize that coding is not merely a skill to land a fancy job, but also a hobby that can pay well. My mentoring style is “less lecture, more code and creation.” I always tell my young padawans: “It’s 20% learning and 80% making.”

Enters ChatGPT 🤔

A few months ago, I was mentoring a young coder and noticed a pattern: much of the assignment code was AI-generated. I’m sure many senior developers have encountered this too. But if we think about it, isn’t that what AI tools are meant to do—generate code? Then, I asked this young developer why he included a particular piece of code, and he had no answer. That’s when it struck me: this can be a real problem for those just starting out.

🚀 The Race: Challenges Facing Young Developers

Thanks to Instagram and TikTok, our ability to focus is declining—sometimes referred to as “Brain Rot.” 🧠 This issue affects all of us, but it’s especially troublesome for people who are trying to gain new skills. Millennials had an advantage: we were accustomed to long-form content when we were learning. Gen Z, however, often struggles with lengthy tutorial videos and documentation—they want fast solutions and quick results.

Add a challenging economy and job market—plus the existential fear of AI replacing developers—and you get a sense of urgency. Big tech companies marketing AI as “Use it or be replaced” amplifies that pressure, making the obvious solution for many newbies: “Let AI write the code for me.”

Is there a better way? 🤷

🤖 Learning Code With AI to Your Advantage

First of all, breathe. You can’t rush learning, and you shouldn’t. The time you invest now will pay off later in your career. You also have something we didn’t have back in the day: the power of AI. Think of it not as a slave to do your work, but as a teacher that never tires of your 1,000 silly questions. Use it to your advantage.

🔍 Fix Your Focus

We often hear the obvious advice: “Meditate.” Sure, that helps some people, but you don’t need to become a Jedi 🧘 to fight smartphone-induced brain rot. Instead, set aside “block time”—when you’re learning, put your phone on Do Not Disturb and trust that you aren’t missing out on anything crucial.

Start small. There’s a well-known technique called the Pomodoro Technique: focus on your task for 20 minutes, then take a 5-minute break. Gradually increase your focus intervals over time.

Another strategy is to deliberately engage with things you find “boring.” For example, read just one page of a book every day—something like The Daily Stoic by Ryan Holiday. Or watch longer, more in-depth videos on YouTube about topics that interest you. Don’t start with a 30-minute video right away; ease into it like you would with weightlifting. Focus is a muscle to be trained.

🏗️ Master the Fundamentals

Ask questions, and truly understand the underlying concepts. Don’t just ask for the code solution—ask “Why?” If you come across a tricky concept in your programming course, use AI to guide you until you really grasp it.

When practicing, turn off your AI coding assistant now and then. You’ll do fine. Over-reliance on AI won’t help you in the long run. Once you’re confident and have enough experience with basic tasks, switch it back on and think of it as a junior coding assistant. Let it handle scaffolding, but always double-check the work it produces.

📚 Use Official Documentation

I highly recommend reading documentation. Senior developers often put a lot of effort into making it as clear and engaging as possible. Additionally, consider using tools like Wiseone or Arc Max to ask questions directly about specific documentation pages.

LLMs can sometimes struggle with error handling or hallucinate (fabricating details), which can lead you further from a solution. In those cases, take a moment to do a quick Google search. You might be surprised how helpful GitHub Issues and Stack Overflow can be. And if you’re feeling lazy, use an AI tool to summarize those pages.

🤝 Use AI as Your Critic

Here’s something I’ve recently tried: let AI critique your code—technically roasting it. 🤭 You can even give the AI a fun personality to spice things up. It’s both insightful and entertaining. You can also do pair programming with AI: ask it questions like “Is this approach better, or am I missing some edge cases?”

🌐 Contribute to Open Source

This is age-old advice but still relevant. Don’t just make mindless PRs that fix grammatical errors to boost your GitHub profile—AI has probably taken over that niche anyway. Instead, create your own repository. Something small but genuinely useful, like a static workout app for your gym buddies, deployed on GitHub Pages.

💡 Conclusion

The bottom line is that AI is a great tool—if you use it to your advantage. Telling people not to use AI at all isn’t helpful, but over-reliance on AI can undermine the core developer skill: problem-solving. The strategy should be yours; AI just helps you execute it. After all, developers aren’t just people who write code—some architect it. 🏗️

Let me know how you’re using AI in your learning adventure and what challenges you’ve faced while learning with AI!

How you used AI in 2024?

Sarthak Sharma — Tue, 07 Jan 2025 18:04:49 +0000

2024 seemed to be all about AI—everywhere you looked, it was AI this and AI that. But once we cut through the marketing hype, what was actually useful to you?

I'm curious to hear from the community:

What AI use cases did you integrate into your life in 2024?
Which products or tools did you find yourself relying on the most?
To make it more structured, let’s divide this into two broad categories:

Personal Life: Did AI help you with daily tasks, hobbies, or self-improvement?

Professional Life: Were there any tools or workflows that transformed how you work or made a significant impact on productivity?

Feel free to share both what worked and what didn’t for you, and everything in between. Let’s build a collective insight into how AI has shaped our lives this past year!

My Experience with AI in 2024:

Personal:

I used Arc Search by Arc Browser a lot as a replacement for Google Search when I wanted to find or learn something.
I used Ollama + Mixtral + Obsidian for journaling, marketing write-ups, and more. The Ollama Plugin enabled me to do AI-powered writing even offline.
I also used ChatGPT for various purposes, such as gardening, cooking, travel planning (along with Arc Search), interior design ideas, diet, and exercise planning.

Work:

ChatGPT is at the top of the list. While I used it less for handling complex bugs (I often resorted to forums and GitHub for those), it was very helpful for research and planning new projects.
Copilot has been great for improving programming speed. With new features and better models being released, I’m excited to explore them further. Overall, it’s an excellent autocomplete tool.
Ollama + Mixtral + Obsidian were also useful for work journaling.

I’m sure I’ve missed a lot, which is why I’m curious: How have you incorporated AI into your life in 2024?

Looking forward to hearing your experiences! 😊

Rise of Local LLMs ?

Sarthak Sharma — Wed, 28 Feb 2024 13:30:36 +0000

In the not-so-distant past, dabbling in generative AI technology meant leaning heavily on proprietary models. The routine was straightforward: snag an OpenAI key, and you're off to the races, albeit tethered to a pay-as-you-go scheme. This barrier, however, started to crumble with the introduction of ChatGPT, flinging the doors wide open for any online user to experiment with the technology sans the financial gatekeeping.

Yet, 2023 marked a thrilling pivot as the digital landscape began to bristle with Open Source Models. Reflecting on an interview with Sam Altman, his forecast of a future dominated by a handful of models, with innovation primarily occurring through iterations on these giants, appears to have missed the mark. Despite GPT-4's continued reign at the apex of Large Language Models (LLMs), the Open Source community is rapidly gaining ground.

Among the burgeoning Open Source projects, one, in particular, has captured my attention: Ollama. Contrary to initial impressions, Ollama isn't just another model tossed into the ever-growing pile. Instead, it's a revolutionary application designed to empower users to download and run popular models directly on their local machines, democratizing access to cutting-edge AI technology.

What's the Hype?

Okay, instead of just telling you, let me show you something.

Download Ollama onto your local machine.

Once that's done, try chatting with the model here:

In case you encounter a CORS error, run this command:

OLLAMA_ORIGINS=* ollama serve

With that, you're now able to chat with the Llama2 Model, in an article, running on your local machine.

Take a look at the code—it's just a simple fetch request:

fetch("http://localhost:11434/api/generate", {
      method: "POST",
      headers: {
        "Content-Type": "application/json",
      },
      body: JSON.stringify({
        model: "llama2",
        prompt: document.querySelector("input").value,
        stream: true,
      }),
    });

llama2 isn't the only Model available. They have a long list of models in their library that you can download and run locally. All you have to do is change model name , model: "llama2", in the fetch request.

Fun right. Now you have powerful models running on your computer (offline).

What's fascinating is that you can run it on any machine with 8GB of RAM (7B models). To run bigger models you might need more RAM.

How This Can Change Everything?

Major players like OpenAI and Cohere offer similar APIs, but there's always a cost associated with them, not to mention the privacy concerns.

Sure, hosting and running these models online is an option, but just imagine the possibilities that come with operating these models offline.

For instance, take a look at this plugin, Text Gen, that you can use in Obsidian. You can set Ollama as your LLM Provider in the settings and specify the model to be used like this:

And voilà,

You now have auto-completion in Obsidian, just like that.

Alternatively, you can use Continue as a replacement for GitHub's Copilot. Simply download codellama with Ollama, select Ollama as your LLM provider, and voilà—you've got a local Copilot running on your computer.

I mean, now you can do this...

No internet needed. Zero cost.

Interested in more cool experiments built on Ollama? Here are 10:

// Detect dark theme var iframe = document.getElementById('tweet-1757333815448711560-71'); if (document.body.className.includes('dark-theme')) { iframe.src = "https://platform.twitter.com/embed/Tweet.html?id=1757333815448711560&theme=dark" }

And...

Imagine adding an Ollama provider in your project. If the user has Ollama installed, you can provide powerful AI features to your users at no cost.

Conclusion

This is, of course, just the start. The newest versions of Ollama feature some really cool models, such as Gemma, a family of lightweight, state-of-the-art open models built by Google DeepMind that were launched last week.

Or consider the LLava model, which can endow your app with computer vision capabilities.

Ollama is just the beginning, I would say, but it's not the only project embracing this philosophy. There's also an older project called WebLLM that downloads open-source models into your browser cache, allowing you to use them offline.

So, where does it go from here? No one really knows. But if I had to guess, I'd say these models will soon be part of operating systems. I mean, they almost have to be. It's possible that Apple is already working on one—who knows? Their machines certainly have the powerful chips needed. 🤷‍♂️

Offline LLMs (Large Language Models) will be the future, especially given concerns about privacy and security. Before long, even your mobile devices will be equipped with LLMs. The question is, how will we utilize them? Trust me, chatting is not the only use case; there will be many more. The latest mobile devices like the Samsung Galaxy S24 Ultra have already started taking leverage of AI powered apps, and they're relatively well-integrated into the OS as a whole. My guess is that as time progresses and local LLMs get more powerful, this will further empower our mobile devices in even more novel (and secure) ways.

I would encourage you to try your hand at this technology. You can visit Ollamahub to find different model files and prompt templates to give your local LLMs a twist.

Consider using a library like LangChain to further build upon this technology, adding long-term memory and context awareness. You can even try a low-code setup locally with Flowise.

Isn't this exciting? I'd love to hear what you think! 🌟 Have any cool ideas started bubbling up in your mind? Please, don't hesitate to share your thoughts in the comments below—I'm all ears!

And if you're itching to chat about a potential idea or just need a bit of guidance, why not drop me a line on Twitter @sarthology? I can't wait to connect with you. See you around! 💬✨

How to Achieve true AI Alignment?

Sarthak Sharma — Thu, 19 Oct 2023 22:54:27 +0000

Today, while having a random conversation with a friend on Twitter about AI, a thought crossed my mind. So I decided to share it here among brilliant minds and ask, "What will you build with AI now?" 🤖🤔💡

I can confidently say that AI can solve many problems, but the most crucial ones are often found within ourselves. If we can genuinely recognize our flaws, AI can be a valuable assistant in addressing them. Flaws within us may persist even when our intentions are good. However, where humans may falter is in finding the courage to confront them. This is because it's a challenging task and requires hard work, and AI can help bear the weight of that effort, enabling us to achieve genuine AI alignment.

What are your thoughts on it? Got an Idea to share? 🤔

Where are the Good Bots?

Sarthak Sharma — Tue, 17 Oct 2023 08:20:14 +0000

Hey all 👋

Before I start, I want to mention, take this as a thought experiment, and jump in with an open mind. I'm coming from a place of empathy here; it may be naive to ask this, but my intentions are in the right place.

So, if the social media platforms are filled with evil bots, mostly doing harmless stuff, but in conflicts, they can be weaponized. Let's assume social media platforms can get overwhelmed tracking them all.

So, where are the good(unbiased) bots?

Chapter 3: Three Worlds

Sarthak Sharma — Sun, 15 Oct 2023 18:23:10 +0000

Hey, everyone,

It's been a week, and as promised, here is Chapter 3.

Don't forget to play the background music I composed for all of you. Try listening to it while reading. 📚🎶🎵

Three Worlds

The land of Internet may be divided into two parts, but these two parts have in them three entirely distinct worlds.

It has been 70 years since the first wave of Engis arrived here, wave 'Boom', commonly known as 'Voomers'. They were the ones who laid down the first colonies on the surface. The 'Idea Domes' did help in combating the harsh atmosphere of the surface during daytime, but Voomers still had a rough time in the scorching heat; so they called the surface city on both side of SPCT-RIM 'Hard-Ville'.

The Left Region was named 'Breachers' as Engis on the region believed they could breach their way into the HAPI-NS.

The Right Region was named 'Keepers' as they believed in the ancient wisdom of Humani-T. "Things will fall into place when the right time comes," they would often repeat the saying, being true believers. They had faith in their ancestors and believed that there was a reason they preserved it, and when it's time, the way into HAPI-NS will reveal itself.

The Breachers would soon discover something inside the DeTA caves, something special that could elevate Humani-T to newer heights and lead it to achieve what it never could in the past. They thought the days of the Keepers were long gone.

And what’s that special thing you may ask?

Magic 🪄

You see, the light source emerging is not just to provide lighting in the caves. The Second Wave on the Lands of Internet, Wave X, also know as VXs, discovered it in DeTA caves.

Voomers made enough gold from initial settlers, but the people new to Inter-Net couldn’t afford it. They went with the old ways and stayed in the caves. VXs were the first ones to settle there. The surface was called Hard-Ville, but ironically, it’s actually harder to live in DeTA caves. That’s also the place where they discovered Magic. Every day inside the caves was a challenge, survival took everything, and the residents were no strangers to tragedy. But as they say, “The deeper you fall into tragedy, the more truth you discover”.

And the truth was that the meteor strike had a more profound impact on the caves than it did on the surface. To begin with, it tore the entire land into two pieces. It also gave birth to the infamous ELGO Vents within the caves, which emitted a constant blue light, creating an effervescent ambiance in the rooms. Besides the illuminating glow, there was something peculiar about these vents in the cave. Many had experienced hushed whispers and strange noises emanating from within. It didn't seem significant at the time, but some believed it to be the source of Magic.

With their suspicions confirmed, the VXs first established Magic Camps in DeTA caves on each side of the SPCT-RIM. They then proceeded to teach other VXs how to conjure spells and let the magic take its form.

Fundamentally, there were two types of magic that existed on the planet, and the type of magic you manifested was purely based on your innermost intentions. The scholars who studied and developed them further classified them into various utility-based subcategories. You could learn healing magic to mend and help others. Alternatively, you could delve into mech magic, allowing you to transform metal into formidable machines. Another path was to become a spellcrafter, merging different spells to create entirely new ones with diverse purposes.

After that, everything seemed possible.

Every wave brings their own flavour to Inter-Net, and the next wave, 'Wave Min', was no exception. They created the third world, the Astro Realm as it is now called.

Remember the web-like structures in skies? 'MinLis' crafted a spell that allowed them to send an astral projection of themselves into floating pods with just a snap of their fingers.

These pods appeared small, but they held the power to contain thousands of astral projections, all projected into vast imaginary lands. This innovation resolved the last issue of the Land of Inter-Net: connectivity. People from both lands could coexist in unknown worlds and assume any form they desired. They could let their imaginations run wild with their projections.

Don't like the way your hair looks? Snap your fingers and you've got a new hairstyle. Fancy a cool nose ring? Zap! Now you have one. Be whatever you want to be and meet whoever you want.

In the Astro Realm, the sky was the limit. All your desires could be fulfilled.

And now comes the Last Wave, Wave Z, also known as 'VNZs.' Our hero belongs to this wave and he choose to be in the Breachers city in the last chapter. GeAK, I hope you still remember him, right?
Let's trail back to him now, shall we?

4 years later....

GeAK stood in front of the room that held the decision for their future. He was scared to enter, but he reminded himself that this wasn't about him. It was about Lily, after all; it was her big day.

When a child turned 18 in their world, it was time for the "Spark Ceremony" in the magic camps. After years of learning, apprentices ignited their first flame of magic in either hand. The color of the flame determined which side their core was aligned with. If the flame turned Neon, it meant the left side, and if it turned Radiant Yellow, it was the right side.

As GeAk moved through the crowd, he realized that he had just made it in time. Lily was up next for the ceremony.

"Lils! We love you!" GeAK shouted and waved to his little sister.

"Now, Miss Lily," the Camp's Chief called Lily onto the stage.

She came to the stage, gave a nod to GeAK with a hint of a smile, and took her position.

"Let the truth be told," the Camp's Chief declared.

For a moment, GeAk could hear his heartbeat clearly. He knew in his heart that it had to be Neon. He had always hoped - no, not hoped - he had always known it. After all, that's what Lily had chosen for GeAK when she was a kid... It had to be Neon, right?

SNAP!

That loud sound from Lily's fingers shook him out of his reverie.

"The Flame has been chosen. Miss Lily, welcome to the Keepers,' said the camp’s chief.

"Ah," GeAK thought with a bittersweet smile. "The bag was a good idea, I guess."

As he looked down, the bag he had prepared for his little sister came into focus. It now served a purpose, as they had to leave for the desired land directly from the ceremony.

GeAK could feel it—the pain of the unexpected. His eyes were almost wet, but then he saw Lily.

'She is happy.' A smile spread across his face.

'So this is it, eh Lils?' said GeAK.

Lily ran towards her brother and hugged him.

'I will be fine, broski,' Lily reassured him. She didn't sound too convinced herself. There was a strange mix of emotions in the air; in that moment, they both understood each other without words. Lily was excited for the land of Keepers but sad to leave GeAK behind. GeAK was also sad, knowing she would be gone in minutes, but knew what he had to do in order to look happy for his sister's sake.

They both decided on silence. In a way, it conveyed what words could not. They looked at the incoming Mrain 🚂 and knew it was time.

'Don’t worry, Lils. I’m gonna come to visit you. One day, I will be the best Spell Crafter there is,' GeAK confidently proclaimed with a big smile. It was a promise.

'I know, broski,' said Lily, a hint of sadness in her voice, then recovered quickly and stuck her tongue out at him mockingly. 'I'll be waiting then."

She boarded the Mrain, and GeAK was soon left alone at the platform as it departed.

to be continued....

[CREDIT ROLL]

Editor 1 - Sneha Saxena It's her birthday today 🥳

Editor 2: Utkarsh

Post Credit

Back in his room, emotions ran high. The room had never felt so empty. Was it the room or was it him?

He knew that choosing the right side would be a daunting journey. Becoming a spell crafter seemed like a distant dream, especially since the very first step required a substantial amount of gold – a Spell Crystal to merge different spells.

"It may take forever to collect that much gold," GeAK said, his voice heavy with the weight of the reality sinking in. He acted confident in front of his sister, but when faced with reality, he felt miserable.

'How the hell am I going to pull this off?'

Suddenly, the voices emanating from the caves ceased, and the room filled with strong gusts of wind, the source of it somewhere unseen. GeAK could barely see as the winds whipped through the room at a rapid pace. Unaware of what was happening and utterly confused, could only yell "Stop!" and close his eyes.

As he cautiously opened his eyes, he couldn't believe what he saw—a Spell Crystal now sat in the middle of the room.

Behind GeAK, concealed in the shadows, a hidden passage revealed itself. A mysterious figure emerged, whistling with neon and yellow fire dancing in their hands.

"The lies are coming down!" ~ Mento

Chapter 2: SPCT-RIM

Sarthak Sharma — Sun, 08 Oct 2023 18:41:30 +0000

SPCT-RIM

The most dire challenge posed by the Inter-net land was the colossal canyon that emerged after the meteor collision. This deep fissure broke the vast expanse into two distinct regions, forever known as the "SPCT-RIM," symbolizing the great divide.

The Engis embarked on construction efforts on both sides of the SPCT-RIM, strategically dividing their men as needed.

However, the once-natural landscape of the land was now leveled. Yet, the underground cave systems endured, providing a foundation for their new endeavor—building subterranean cave settlements on each side.

Thanks to the marvel of light-emitting magic, underground cities were meticulously crafted on both sides of the "SPCT-RIM."

Intriguingly, each side employed distinct magical techniques, resulting in varying hues of illumination. This distinction allowed people to choose the color of their preference, adding a touch of personalization to their underground abodes.

While the enchanting magic unfolded below ground, a remarkable revelation occurred above. Engis laboring on the surface uncovered a transformation in the land's silica caused by the meteor impact. This newfound material, when melted and molded, produced a unique glass capable of shielding people from the harshest of elements.

This discovery marked the turning point for the Inter-Net, transforming it into the City of Glass Domes, a marvel of magnificence. Each region's distinctive magical light imbued it with its true beauty, rendering it a breathtaking spectacle in every sense.

With the radiant yellow and white light on the right side and the vibrant neon glow on the left, the City of Glass Domes in the Inter-Net became a mesmerizing fusion of two distinct worlds, a testament to the ingenuity of the Engis and the beauty of their dual magic.

"So, I hope you enjoyed the story. Left or right?" inquires the officer at the gate.

"I'm asking you, sir. What's your name?" officer persists.

"Who, me? I'm Geak, ma'am," says the young lad.

"Welcome to Inter-Net. Which side do you reside on, left or right?" the officer at the gate inquires.

The young lad, like any of us, was clueless. He looked at his little sister and realized they were now one step closer to HAPI-NS. Left or right didn't matter to them.

"Hey Lils, what's your favorite color?" Geak sat down on his knees and asked his little sister.

She chuckles and points, her tiny finger guiding Geak's gaze towards a particular hue, a simple yet profound choice in a world where colors held special meanings.....

To be Continued.......

Internet And Shadow Groups

Sarthak Sharma — Sat, 07 Oct 2023 15:10:06 +0000

Hey all, it's been a long time since we all had a conversation. First, the pandemic, and then the noise of AI-generated content has given me no reason to post, other than the mandatory duty posts that I have to do because of my business endeavours. 😅

Well, I feel like there's still a reason to connect, not on a literal level – I mean, come on, how many people can you actually talk to – but on a deeper level, through ideas, philosophy, and stories. 🤓

Stories. 📖

I have one for you all. 😄

I know in this hustling world, why would you read a stupid story? Because, in a way, it's our story, connecting all of us together. So bear with me. And it's my first attempt at this; that's why I came here to share this. Back to home. 🏡

"INTERNET AND SHADOW GROUPS" 🌐👥

In a galaxy distant and unknown, there existed a planet named 'Humani-T'. It was a world populated by nomads, amidst a stretched wasteland.

The scorching temperatures left the surface unbearable, forcing the inhabitants to seek refuge within the natural cave systems of the planet, 'De-Ta', during the fiery days.

As the sun descended below the horizon, the people prepared to make their ascent...

And to where, you might wonder...

Well, on Humani-T, a closely guarded secret existed, which is known as "HAPI-NS." It's a world within a world, flawlessly crafted by the ancient beings, safeguarded by a mysterious enchantment. Should you seek answers from the elders, they will relive tales of days gone by when individuals could dive into the depths of HAPI-NS. However, in recent years, none have managed to break its mystic barrier.

Yet, the legends of HAPI-NS remained majestic, painting vivid pictures of lush forests and cascading waterfalls. Although the sheer reality of Humani-T's downfall might have led some to question the existence of such a paradise, a deep longing for it remained etched in every soul.

And why shouldn't it? 🌿💧🌟

For in its truest essence, it was indeed a paradise...

A few years ago, due to a rather significant incident, everything changed.

You see, reaching HAPI-NS was never a challenge; everyone knew how to get there, and its location was always well-known.

But the spell...

The incident? An asteroid collision with planet Humani-T. Fate, it seemed, had chosen its course, guiding the celestial rock to hit right next to the HAPI-NS Globe.

And that single event reshaped the entire landscape..

Suddenly, strange lines began to turn up, decorated with bright neon lights, rising from their depths.

This captivating event drew the attention of Humani-T's most ingenious and sly builders, a tribe known as the Engis from the NaR region.

As they dared closer, they realized that these lines were extending from HAPI-NS, stretching all the way into the atmosphere above.

Indeed, much like a sprawling spider's web.

And so, the Engis, most craftly, bestowed upon it the name "Inter-net," signifying the Open Land, the gateway to HAPI-NS.

So it lured settlers from far and wide, drawn by the promise of the Inter-net's mystical allure.

To be Continued.....

[CREDIT ROLL]

Chief Editor - Sneha Saxena

More to be added soon.....

💯 7 Free cool Form Components made in TailwindCSS

Sarthak Sharma — Mon, 04 Apr 2022 13:48:28 +0000

In this series, we have explored the web for all the cool tailwindCSS components that one can find for free and brought it to you. Today, we are exploring the form elements available on the web for free. Login Form, Signup forms and much more. So let's get started.

1. Login form by Erich Haemmerle

💵 Price: Free
🔗 Link: https://codepen.io/erich93063/pen/MWvLxNN
A cool cherry color Login page.

2. Registration form

💵 Price: Free
🔗 Link: https://tailwindcomponents.com/component/form-register-component
A split screen Login page, used quite often these days

3. Checkout Pay

💵 Price: Free
🔗 Link: https://codepen.io/cruip/pen/wvqVpbP
A Simple checkout form made in Tailwind CSS with both card and PayPal options.

4. Banner Image Signup Form

💵 Price: Free
🔗 Link: https://tailwindcomponents.com/component/login-7
Another cool Login Page with Big Side banner image, a lot of people use it these days.

5. Payment Form by Joe

💵 Price: Free
🔗 Link: https://codepen.io/dope/pen/gORRMOo
A minimalist payment form for Card payments

6. Account Form

💵 Price: Free
🔗 Link: https://tailwindcomponents.com/component/account-form
Edit account form for you users with multiple fields.

7. Checkout Form by Scott Window

💵 Price: Free
🔗 Link: https://tailwindcomponents.com/component/account-form
Checkout form made in tailwind for Ecommerce sites.

Conclusion

Thanks a lot, We will see you next week with some more components for you. Free components are useful for you start learning and explore a framework but sometimes you need Premium Components to speed up the Development process.

That's why we made DollarTailwind a UI kit with over 200 components to use in all kind of projects. We are giving an additional discount of 20$ to all our readers. Use code "DEVTAILWIND" during checkout, avail your discount and get 400$ worth of components for 79$.

Just released Notion Pro Clipper Web Extension

Sarthak Sharma — Sun, 21 Nov 2021 11:18:50 +0000

Just Released our product Notion Pro Clipper on Producthunt. After working with 700+ beta user from Notion Community our it's finally here.

So, what can you do with Notion Pro Clipper?

🌐 Page Clipper

3 ways to clipper the information on web using Notion Pro Clipper:

Just URL: The plan simple way to save urls to Notion
Content of Web Page: Clip the entire content on webpage to Notion
Section of Web Page

📸 Take Screenshot

Notion Pro Clipper lets you capture screenshots of any page in one click and adds them to whichever Notion page you want. It lets you take screenshots in 3 ways.

Visible Section: Section visible currently
Entire Page: From top to bottom
Free clip: Select any area you want on a page

🌄 Image Clipper

Just hover on any image on web you want to save and save it directly to Notion.

Bonus

Save any Youtube video directly to Notion with one click.

I hope you all are ready to use this. Get it Notion Pro Clipper

We are on Producthunt today. Go show some love here.

Thanks Again 🖖🏼