Driving business growth with AI | Chief AI Officer & Advisor @ MechanizedAI | ex-AI Principal @Apple | Georgia Tech, Stanford, MIT
Joined May 2013
- Tweets2K
- Following556
- Followers122
- Likes4.3K
Pinned Tweet
We will soon (are) see a MASSIVE acceleration in AI startups created. If you are leveraging an AI product, ensure the CTO or CDO has actually built and managed AI in production for large companies. Anything else will cause silent failures that will be eaten by the customer.
Aditya Murali retweeted
Yep and that actually enables you to solve biz problems rather than building BS “cool” stuff that never delivers biz value
Aditya Murali retweeted
you usually don't need fine-tuning.
but when you're pushing the boundaries of performance and efficiency, you do
🧠 RFT - train reasoning models on SotA tasks
⚖️ DPO - a/b optimization, tone matching
⚡️ SFT - distillation, formatting, efficiency
when fine-tune w/ @openai? 👇
check out my full @aiDotEngineer workshop on everything fine-tuning, where i dive into each approach, some anecdotes from fine-tuning with customers at @openai, and hands-on examples.
many thanks to @swyx and team! (and @TSautory for awesome RFT resources)
Aditya Murali retweeted
It is exciting to combine skills across modalities. Here we explore model merging as a training-free tool to compose perception of VLMs and reasoning of LLMs. Check Shiqi's thread for more details!
Share our another #ICML25 paper: “Bring Reason to Vision: Understanding Perception and Reasoning through Model Merging” !
(1/5) We use model merging to enhance VLMs' reasoning by integrating math-focused LLMs—bringing textual reasoning into multi-modal models. Surprisingly, this avoids catastrophic forgetting and yields strong performance gains as a free lunch ! 🍱 We further leverage merging as an interpretability tool and uncover a key insight: perception and chain-of-thought reasoning are naturally decomposed in the parameter space! 🌍🔍
Paper: arxiv.org/pdf/2505.05464
Code: github.com/shiqichen17/VLM_M…
Aditya Murali retweeted
don't miss that OAI also published a prompting guide WITH RECEIPTS for GPT 4.1 specifically for those building agents... with a new recommendation for:
- telling the model to be persistent (+20%)
- dont self-inject/parse toolcalls (+2%)
- prompted planning (+4%)
- JSON BAD - use XML or arxiv 2406.13121 (GDM format)
- put instructions + user query at TOP -and- BOTTOM - bottom-only is VERY BAD
- no evidence that ALL CAPS or Bribes or Tips or threats to grandma work
Aditya Murali retweeted
💁♂️ How to Play Long Term Games:
Systems > Goals
Discipline > Motivation
Trust > Distrust
Principles > Tactics
Writing > Reading
Vulnerability > Confidence
North Stars > Low Hanging Fruit
Trends > News
Habits > Sprints
Questions > Answers
Problems > Solutions
People > Projects
Aditya Murali retweeted
Understanding MCP From Scratch
MCP is an open protocol connecting AI models w/ data + tools. Here, we explain how it works, build an MCP server for LangGraph docs from scratch, and connect it to Claude / Cursor / Windsurf.
📽️:
youtu.be/CDjjaTALI68
Aditya Murali retweeted
The DeepSeek innovation is a signal that the world will buy more NVIDIA not less. Its a fundamental misunderstanding of how hardware is consumed to suggest that a 10x improvement leads to less hardware demand.
1. There is no amount of hardware that ever satisfies software. Software simply runs up to the line and demands more. Trust me.
2. More AI means more inference around the world, inference is the real consumer of NVIDIA compute.
3. Optimized AI just signaled to little tech that they can compete with big tech. Little tech will expand AI beyond LLMs, every compute use case is open.
Aditya Murali retweeted
I personally think that entering fields with ai is a decent bet, but the more valuable thing imo is being able to critically think and have creative ideas.
Creativity will take a bit of time for ai to figure out, and until then, humans will be able to drive a lot of value there.
Rote memorization and execution oriented fields (doctors that don’t research and invent, cs people that don’t invent, lawyers that don’t add unique value like argument driving and persuasion) will be hard to find jobs for.
Aditya Murali retweeted
I used Cursor to setup a backend in just a few minutes
I have no idea how to code (you can clearly see that in the video 😅)
But just by talking to Cursor I got a working database in a couple minutes
Step-by-step tutorial (incl. bug fixes) for beginners:
Aditya Murali retweeted
Michael I. Jordan on ChatGPT and AI: "We're fearful in 10 years it can get out of control and all that. That's just not true. ChatGPT is predicting the next word in a sentence, just remember that's what it does".
youtu.be/mszj94hEKH4
1/3
Aditya Murali retweeted
Check out @karpathy's "State of GPT" - it's 40 mins well-spent! It packs in some serious knowledge:
🧵
Aditya Murali retweeted
🤖💼 Ever wondered how to leverage #AI for your business? In this era of digital transformation, AI isn't a luxury but a necessity to stay competitive. Tag a business that needs to see this! #AIinBusiness #DigitalTransformation
🤖💼 Ever wondered how to leverage #AI for your business? In this era of digital transformation, AI isn't a luxury but a necessity to stay competitive. Tag a business that needs to see this! #AIinBusiness #DigitalTransformation
🌐 Remember, the goal isn't to replace humans but to augment their capabilities. AI can handle routine tasks, freeing up your team to focus on more strategic, creative work. Share if you agree! #AIandHumans
1
1
80
🎯 Productizing AI can be a game-changer for your business. Start today and reap the benefits of increased revenue and reduced costs. Like, retweet, and reply with your thoughts! Stay tuned for more insights on leveraging AI in business. #AIforBusiness
1
75