I build AI and AI-powered products. My focus areas are healthcare, making SMBs operationally effective and algorithmic trading.
Bangalore
Joined March 2009
- Tweets9.5K
- Following3K
- Followers1.9K
- Likes8.6K
Success is 99% perspiration. 1 % inspiration.
It is even more true in this day and age of AI agents.
I would add focus also to it. Even sun rays burn only when focused on a particular point.
Everytime a long build is happening in Opus 5.5, I keep asking it "status and eta?".
It gives a darn good picture of the status - with clear challenges and what might take time etc.
But it still is horrible in eta. I've had it give me timelines for 6+ weeks when it finished the entire task overnight.
It has definitely been trained on some terrible real world s/w plans :)
Here's my weekend read/watch list (mostly a watch list)
• What are recursive language models?
• What goes in building a kick-ass software factory?
• The new agent architecture
• How to build an always-on agent?
• A new video series on reasoning models from scratch
Like the city that never sleeps, we now have AI agents that never sleep. They are more commonly referred to as "always-on" agents.
The early wave was openclaw, hermes etc. But that all appealed to the hackers. But with Meta's Muse, Grok's @bot, OpenAI's dot and now Claude's Duet - this has become mainstream and is accessible to everyone.
But what goes on in building this? A great talk by @leerob at Stanford
xitter.cf/leerob/status/21086502…
How do always-on, proactive agents like @Bot work?
Watch my lecture at Stanford CS146S:
2:35 How we got here
6:13 What changed in the models
10:06 Inside an always-on agent
18:07 The harness
27:26 Context engineering
34:26 Where this is going
Following up on his fantastic book, @rasbt has created an amazing video series on Reasoning Models from Scratch. Highly recommend writing and executing the code along with watching the series.
xitter.cf/rasbt/status/211432509…
The complete reasoning-from-scratch video series is now up (and linked) here: magazine.sebastianraschka.co…
(Taking a short break but will be back with LLM model and research updates in November!)
I like how Claude can co-ordinate runs across projects and load balance the runs. The orchestrator (Opus 5.5) has a md file that gives the priority of the projects (based on deadline, criticality etc).
It runs all night - picking up tickets and fixing it across projects - and scaling up or down the sub agents so that machine load/temperature/ram usage are manageable.
There is a classifier on haiku that decides which model the sub agent should use. Based on available usage and Tibo's tweets - it will decide whether to go all in on codex sub agents or stick to claude family.
Currently, my machine is my limiting factor!
100% agree to this. In this age of agentic engineering, kid you not - the person working on the code matters a lot more than ever!
Thanks to @inventcures for pointing me to this !
One for the F1 buffs like me from the stables of @collision .
Timing couldn't have been better - Antonelli is on the verge of being crowned the youngest ever F1 champion.
Huge comeback for Mercedes this year. This is going to be a good listen!
youtube.com/watch?v=mFLcqTGp…
Suddenly there is a deluge of voice-to-voice models. While they all excel in being fast and expressive voice, they all struggle in long context.
For most of this year, I've battled using all kinds of voice models and platforms to solve intake questionnaire. These are crazy long number of questions - and while this is critical and mandatory in healthcare, I've seen this in other fields too.
I am yet to come across a reliable setup for this.
Games are a fascinating test bed to both learn and try out new strategies with AI Agents. I loved this talk where @maxbittker talks about agent SDK design, evals and multi-agent simulated economies. I found this insightful and entertaining.
youtube.com/watch?v=cNm4zyUt…
Love this. This is just the start of the agentic wave in manufacturing. Like how the AI software factory has rapidly progressed and taken shape, we're going to see something similar in manufacturing.
CAD + simulation + stress testing would unlock massive value.
Introducing SKY-001, the first Engineering World Model.
On CAD and BIM tasks, it beats GPT-6 Astra and every frontier model we tested:
-> 51-125x faster
- > 150-210x cheaper
-> Higher IOU than all frontier models
SKY-001 lets you dream a physical object before you build it.
18 months of work. Built by the team out of Maluuba's deep learning lab (acq. by @MSRResearch).
The first ‘AI General Engineer’ will be the backbone for Autonomous Manufacturing.
What brilliant marketing this is! And that's quite some storytelling and editing.
Hi, it's Darshita. I am a designer.
And I'm pitching myself in public 'cause I'm done with a year full of anxiety.
Everyone says design is dead and AI killed it. But the real parasite was hiding in plain sight.
So here's my film, DESIGN IS DEAD. She's bold, fearless and very PINK 💖
Hiring? My DMs are open.
I started my career almost 25 years ago - working on COBOL and the mainframe stack - on the world's largest cobol program.
There was absolutely no documentation. Everything- literally everything - was tribal knowledge. The SOPs were all tribal knowledge too!
Once, I went 48 hours non-stop working to fix a critical piece of code - because there was no documentation.
Excited about frontier labs tackling legacy software. So much value to be unlocked from this!
Our latest model advances bring stronger capabilities for building and migrating software in low-resource programming languages such as COBOL, Fortran, and Visual Basic.
These languages power software that banks, governments, and businesses depend on. Progress here means understanding decades of business logic and helping developers modernize those systems while preserving what works.
We’re building these capabilities around the challenges developers and enterprise customers bring us. If your team is working on these systems, tell us which languages and workflows we should tackle next.
openai.com/index/low-resourc…
Bargava retweeted
We’ve just made public 125,091 DICOM radiology images from across my cancer journey (link below). The data is both longitudinal and multi-modal, dated and linked to my treatment timeline, so you can compare imaging across initial diagnosis, standard of care, first remission, “founder mode cancer care”, and second remission. It’s viewable in OHIF, the standard open-source viewer used in radiology research.
Given the long history of applying deep learning to radiology, we’re hoping to see a variety of methods and models applied to my data. If you find something interesting, please reach out to [email protected], or if you know a researcher who should see this, tag them below.
Thanks to Terra LaFerriere for organizing these images, and Kamil Slowikowski for building osteosarc.com into the resource it is.
