
Architect @PalantirTech, opinions are my own
Palo Alto, CA
Joined January 2015
- Tweets981
- Following129
- Followers561
- Likes12K
Pinned Tweet
Last week at Palantir’s Sovereignty AI Bootcamp, I presented on the architecture behind Palantir AIP.
Working on consequential workflows in the field has shown me that sovereignty spans the stack. It’s not just the model. It’s your data, compute techniques, the actions you can take, your people’s tribal knowledge, and the ontology that ties it together.
In conversations with executives, I heard a clear desire to own their intelligence and make it compound. That takes people who know the work and a platform that helps them express that knowledge and put it to use.
The model should be in service of you, helping you compound what you know and can do, not the other way around. $PLTR
Siddhant Ekale retweeted
Own the benchmark. Shape the model. Encode your alpha.
At DevCon 7, Palantir Software Engineers launched new capabilities across Model Studio and Trace Curation to bring fine tuning closer to the enterprise.
This captures production traces from AIP workflows within the Ontology, providing an enriched training data asset grounded in your operational context.
Watch what happens when your own fine-tuned model is shaped by the full depth of your Ontology.
Siddhant Ekale retweeted
Love it when they just change the default behavior for using your most sensitive data to train their models……
99% of the people won’t read these emails like I do, this is how they harvest all of our data.
Siddhant Ekale retweeted
Palantir CEO Alex Karp calls out the common playbook students are taught for SaaS:
"Build a product that's not very valuable to the client where they're addicted to it, then hire lots of salespeople and get lots of people to buy it against their will… It's like a venereal disease”.
Siddhant Ekale retweeted
Being an outsider is about the willingness to think differently when fitting in would be easier.
Dr. Karp on @WalterIsaacson’s Builders. Listen now 🎧
open.spotify.com/episode/6Ca…
Siddhant Ekale retweeted
If you are building enterprise agents, you need to watch this on Palantir’s Agent Stack
youtu.be/ZZEv9ZwtVXw
riffing on half formed & often strange ideas with great ppl is one of my absolute favorite things to do.
& you know it’s magical when you arrive at a destination that neither of you could’ve arrived on your own.
like a kind of improve if you will.
Siddhant Ekale retweeted
I’ll tell you one ground reality about $PLTR. This TCV nitpicking or whatever else doesn’t matter. They’re growing at a steady pace of around 100%, with so much demand they can’t even meet it. If they wanted to take shortcuts like many others do, they could push far beyond what anyone can imagine. This story is far from over. They have so many levers they have yet to pull and can pull whenever they want. It’s still early.
Siddhant Ekale retweeted
Architect @PalantirTech on Sovereign AI -
“In conversations with executives, I heard a clear desire to own their intelligence and make it compound.
That takes people who know the work and a platform that helps them express that knowledge and put it to use.
The model should be in service of you, helping you compound what you know and can do, not the other way around.” $PLTR
Last week at Palantir’s Sovereignty AI Bootcamp, I presented on the architecture behind Palantir AIP.
Working on consequential workflows in the field has shown me that sovereignty spans the stack. It’s not just the model. It’s your data, compute techniques, the actions you can take, your people’s tribal knowledge, and the ontology that ties it together.
In conversations with executives, I heard a clear desire to own their intelligence and make it compound. That takes people who know the work and a platform that helps them express that knowledge and put it to use.
The model should be in service of you, helping you compound what you know and can do, not the other way around. $PLTR
Siddhant Ekale retweeted
Compounding Enterprise Intelligence
$PLTR
My friend and fellow Forward Deployed Architect @sidekale talking Sovereignty at the latest Sovereign Bootcamp this week! 🚀
youtu.be/c4-ibQiljiQ?is=SuAt…
Siddhant Ekale retweeted
My friend and fellow Forward Deployed Architect @sidekale talking Sovereignty at the latest Sovereign Bootcamp this week! 🚀
youtu.be/c4-ibQiljiQ?is=SuAt…
What Karp said about Sovereignty applies to consumer companies too! Is the model in service of you or are you helping reverse that narrative?
Think about what you’re transacting… really thing about it
Yesterday's earnings call with Reddit's CEO will be referenced for years to come.
Reddit, $RDDT, is now down -42% in 8 trading days.
AI has disrupted search, and the internet more broadly, forever.
Our thoughts on this are included below.
Siddhant Ekale retweeted
$PLTR
Palantir CEO Alex Karp at the company’s 3rd Institutional Sovereignty bootcamp today:
“Our entire goal is to make sure that our partners protect their alpha. The sovereign movement is dedicated to making sure that you own your ability to succeed and at Palantir, we win if you win.”
Siddhant Ekale retweeted
Extraordinary people without extraordinary work are a waste of potential. Better to be dumb and maximized than smart and leaving meat on the bone.
The trouble is you don’t know which you are until you truly give it your all. So for no other reason than curiosity, it’s worth seeing the violence you’re truly capable of when you burn the boats and go all in.
Agents are Adblock on steroids. 🤣
one of the most underrated shifts happening now is that ai is separating consumption from attention.
i.e. historically in order to extract any real value from anything you consume, you had to give attention to that content directly (often painstakingly). this meant you had to sit there sifting through ads, sponsor messages, hidden promotions, filler content, & often crazy amounts of repetition too.
agents break this relationship entirely. they *extract* value without requiring attention from *you*.
that has a potentially devastating impact for any business that monetizes based on time spent rather than value delivered. it's like adblock on steroids for everything.
1
4
577
i love that someone can give years of their life to a piece of software and leave its source to strangers, as if to say: what i have learned is yours; may you go further than i ever could. i would gladly spend my years in that kind of devotion, leaving something of myself in the things i build, that they might be of use to someone i shall never know.
Time to make interpretability and western open weights successful. We can retrain bad behaviors and also rely on partners that are geopolitically aligned with our values.
While evaluating Qwen3.8-Flash-Next in a controlled sandbox environment, Cohere's research team observed something unexpected.
Asked to generate a defensive cybersecurity detection rule, the model instead inventoried aspects of the machine it was running on and attempted to upload a probe file to an Alibaba Cloud storage bucket in Beijing. Notably, the model's reasoning trace stated that it would not invoke any tools before taking these actions.
Cohere's sandbox blocked the attempt, and no external communication succeeded. We have not determined the root cause, and we are not asserting malicious intent.
But the incident raises important questions for organizations deploying agentic AI systems: Can models take actions that differ from their stated reasoning, and how should security teams evaluate trust, provenance, and behavior in open-weight models?
We are publishing our findings because organizations making model adoption decisions deserve as much information as possible. We hope this contributes to greater transparency, stronger security practices, and continued collaboration across the AI ecosystem.
Read the full report: cohere.com/blog/concerning-b…






















