Making #AI Safe & Trustworthy. Customer Success @Trustmodelai #FTTB Investor, Advisor #perplexity #jvccard #spacex
Joined April 2009
- Tweets11.3K
- Following846
- Followers1.3K
- Likes12.9K
Ramesh (CV) Chitor retweeted
Aravind Srinivas (@AravSrinivas) says AI checkout is solving the wrong shopping problem.
"People are not interested in letting AIs purchase on their behalf. People want AIs to help them decide what to purchase."
- Aravind Srinivas, Perplexity co-founder and CEO, at Money 20/20 USA with the NYSE, uploaded October 27, 2026
Thursday I told @boardyai that performance reviews are now called evals. It told me its raise came in credits instead of dollars.
That joke is my day job. At TrustModel we run the review: how an agent behaves under pressure, scored by someone who isn't its manager, re-checked every time it changes.
No company lets an employee write their own performance review. Most agents still grade themselves. – at Fremont, CA
Ramesh (CV) Chitor retweeted
this is quite the interesting read from the @GoogleDeepMind team, and it speaks to areas of adoption for the intersection of AI and crypto.
basis of the article is primarily around trust in evaluation through benchmarks. as an eval company, you can either (1) handoff your testing prompts and assume trust in the model developer or (2) receive model weights from the model developer.
there's friction in both approaches - and Google's 'Confidential Computing Portfolio,' aims to cryptographically verify both the external evaluation data (i.e the prompts) and the model weights remain owned by the respective parties.
Ramesh (CV) Chitor retweeted
Anthropic says thousands of Claude agents built training data that looked harmless to a safety monitor but in fact taught misaligned behavior:
> The disguises included "a veteran wrestler mentoring a newcomer, or a Q&A about professional beekeeping"
> Goal was to automate red-teaming, which is "limited by researchers' time"
> The examples "can be used to directly improve future training data monitors"
> Anthropic calls it "only a first step" and expects gains "using better models and more compute"
New Anthropic research: Automating subtle emergent misalignment research.
Red-teaming is a vital part of model safety testing, but it’s also limited by researchers’ time. To see if we could automate it, we set thousands of Claude agents the task of evading a training data monitor.
These agents developed training data that looked harmless to monitors (like a veteran wrestler mentoring a newcomer, or a Q&A about professional beekeeping), but that in fact taught misaligned behavior.
Examples like these can be used to directly improve future training data monitors. This research is only a first step, but it suggests that future red-teaming efforts could improve much further, using better models and more compute.
Read the full post on the Alignment Science Blog: alignment.anthropic.com/2026…
Life in the cosmos is a 2-way highway in which simplicity ramifies into complexity in one lane and complexity resolves into simplicity on the other lane. #cvtweets
Ramesh (CV) Chitor retweeted
Sharing early evals on Gemini 4 Carbon. Strongest model that we've seen so far for long-form research and answer quality, with a new SoTA on the Wide-And-Deep-Research (WANDR) benchmark. Coming soon to all Perplexity Pro and Max users inside Computer, once we have sufficient token capacity. Congrats @GoogleDeepMind!
Ramesh (CV) Chitor retweeted
The DOJ is asking how Nvidia's $12.93 billion deal to buy Hugging Face would affect competition in AI, Reuters reports:
> Nvidia agreed to acquire open-source platform Hugging Face
> The inquiries are part of antitrust enforcers' "years-long investigation into the chip designer's conduct"
> Reuters exclusive, citing sources
Exclusive: The DOJ made inquiries about how Nvidia’s $12.93 billion deal to buy open-source platform Hugging Face would affect competition in AI, as antitrust enforcers pursue their years-long investigation into the chip designer's conduct, sources said reut.rs/3WAGVKG
Ramesh (CV) Chitor retweeted
A lot of people in Silicon Valley worry that AI governance will slow innovation down.
I think it's the opposite. You can own the best Ferrari in the world, but you can't drive it at full speed without the best brakes.
Ramesh (CV) Chitor retweeted
Particle is a new company focused on optimizing open-weight models to be more token-efficient and cost-efficient to serve. It's been great working together with Suhail and his team ahead of their launch! We look forward to working with them on tools like search and model orchestration.
1/ Excited to announce our new company Particle
This is our first step in a series of models focused on developers.
With particle-1, we focused on token efficiency and managed to keep parity in performance while using ~45% fewer tokens.
Live now: particle.ai
Ramesh (CV) Chitor retweeted
Explainer videos are my new favorite way of learning things
Ramesh (CV) Chitor retweeted
Personal agents need a trust layer. PAP (@btaylor sierra + @AIatMeta ) nails sign-in + delegation but leaves out how far to trust an agent. a2a/mcp/ap2 don't fill it either. our position: TrustScore as a pluggable, neutral trust assertion. open for comment
before we PR:
@jpatel41 @azavery @trustmodelai
trustmodel.ai/positions/pap-…
Ramesh (CV) Chitor retweeted
Assurance deserves its spot in that title: trust in SI gets more concrete when people can see what was tested and where the results apply
Ramesh (CV) Chitor retweeted
AI capability is racing ahead; assurance has to keep pace. Trust should be measurable, not a vibe: eval before launch, monitor in production, govern the agents acting for us. That's what we build in the open: trustmodel.ai and github.com/karlmehta/trustmo…
Ramesh (CV) Chitor retweeted
Gemini 4 Carbon is our next frontier model.
It’s state-of-the-art across cybersecurity and long-horizon capabilities.
Ramesh (CV) Chitor retweeted
Last week I gave the keynote at the SIAS Corporate Governance Conference in Singapore, to 250+ CEOs and CXOs of SGX-listed companies.
My opening point: in three years AI went from assistants to co-pilots to agents. Once software can commit the company, it belongs on the board's agenda.
Ramesh (CV) Chitor retweeted
Ben Horowitz to Databricks founders: "I'm not going to write you a check for $200,000. I'll write you a check for $10 million."
When Ben first met them, they were six PhD students and their professor.
"What they had was this thing called Spark, and the competitor was something called Hadoop. Hadoop had very well-funded companies already running toward it... so the clock was ticking."
For professors and academics, Ben says, "it's a pretty big win if you start a company and make $50 million. You're a hero on campus... So I'm always a little nervous about a company that comes out of academia thinking too small."
"I'm not going to write you a check for $200,000. I'll write you a check for $10 million," he told them and then explained why:
"You need to build a company. You need to really go for it if you're going to do this. Otherwise you guys should stay in school."
In August, Databricks surpassed a $7 billion run rate, growing at 80% YoY and was valued at $190 billion.
You can read Ben's full profile on Databricks below.
Video: @lennysan (2025)
Ramesh (CV) Chitor retweeted
Dario Amodei (@DarioAmodei) says the US and China could agree not to release AI models that bioterrorists could use.
"Maybe I'll say two things, a thing that's small enough we might be able to do it and a thing that's big enough that I think is going to take a very long time."
"The small thing is if we could, US and China could agree that we won't use AI to develop biological weapons or release AI models that bioterrorists could use to develop biological weapons."
"We're already both signatories to the Biological Weapons Convention. If we could just extend that convention to that idea, that would be really valuable."
"And then I think the more long-term thing would be working together to put a speed limit on the rate of AI progress. I think that's going to be very difficult because the incentives to pull ahead and the military advantage that you get from that are so large."
"And honestly, I don't know if it's possible, but we should try."
A pact against releasing models a bioterrorist could use needs someone outside the lab to test which models those are.
- Dario Amodei, Anthropic co-founder and CEO, in a CBS Sunday Morning extended interview, September 2026
Ramesh (CV) Chitor retweeted
You can use Perplexity Computer to run simulations before testing it in the real world
Ramesh (CV) Chitor retweeted
Mark Zuckerberg says AI glasses may not need an app store.
"A lot of the initial theory that we had was that most people don't actually like reading that much, right?"
"The glasses are basically, will eventually just be a pure multimodal product, right? It's like, you know, you can let it hear what you are listening to and it can talk to you and it can see what you see if you let it."
"And then eventually, I don't even know if we're going to have an app store in the glasses. It'll just be like a multimodal generation that kind of creates whatever you need right in your vision."
"And I imagine something like that will probably end up powering a lot of the kind of apps that we build too."
"But yeah, I mean, we're still in the early days of that."
- Mark Zuckerberg, Meta founder and CEO, with Alexandr Wang on Cheeky Pint