Haven’t blogged in a while, thought might be good to put some recent thoughts (X tweets) in one place. Last few years has seen alot of trend shift in AI every 3-6 months. Deep Learning, Transformers, LLM Gen AI, RAG, Agents… But all have same underlying ML DS AI basics Python SQL, so it’s just learning new tools frameworks. Rest is easy, quite fun
I am super pumped excited to build products for different industries this AI wave should not be missed just accelerate build. Lot of countries industries recognize its importance for the future
Regularly using Grok Perplexity for AI ML algorithms code R&D, finding grok best, perplexity is good to try different models like chatgpt claude etc. Rate of improvement is awesome. Claude is not that special for coding, DeepSeek is basic not sure why its so hyped. Deep research is good for more details but finding regular chat models as good, similar to reasoning. Used DeepResearch in ChatGPT Perplexity Grok, ChatGPT is best, takes long but very detailed report. Can see lot of service automation happening @OpenAI @perplexity_ai @xai @xai@AnthropicAI @deepseek_ai
Using llm chatbots for understanding ai ml papers is awesome. Ask it to summarize, main idea, algorithm explain, code. Accelerates the learning curve just have to implement code with algorithm logic. No excuse now, damn this is a coding zone issue now. Uploaded llm to generate code for classic AI ML Hinton papers, backpropagation, deep belief nets. Full easy explanation with code, used Grok Perplexity, both similar. Awesome AI coding is so easy now
AI is definitely having the biggest impact on coding application building, shortening the idea to MVP cycle, increasing the experimentation frequency, resulting in more likely product market fit. Great for builders creators startups and markets customers eventually
My take on learn to code or not in AI llm era, most still can’t give full functional end to end code for any business problem in most industries, bugs design flaws, still require lot of stitching together. But do speed up R&D eval deploy prod iterations very useful
Biggest challenge deploying AI model app on cloud is the complex setup too many services, help docs not good, internal llm chatbots are basic, @awscloud @google @azure. I got better instructions from llm chatbots @grok @perplexity_ai
Speed of dev in robotics is amazing accelerated by AI nvidia you open source , will be fun to see what industries get disrupted in next 5-10 years, log curve hockey stick effect in play. Robotics is going to be exciting fun- Optimus, figure, nvidia, so many other startups. The pace of innovation is mind boggling
Used Crew AI, Langchain for AI problems, good easy to use. But I have done same using direct LLM API calls as functions chained in required way for solving the problem. Nice tools, but given current hype on Agents, is it just for non coding technical people, specific problems
All cloud platforms (AWS, GCP, Azure), MLOps tools (Databricks, Sagemaker, Vertex, etc) are very similar in functionality, UI, pricing. Few days to learn any. Real skill is using all tools algorithms to solve business problems in quick iterative improvement cycles
