
Cofounder @alma_inc. Founding member @SarvamAI. Research @MSFTResearch. @iitmadras
Joined May 2016
- Tweets2.3K
- Following621
- Followers2.6K
- Likes4K
Pinned Tweet
We gave @alma_inc a Zerodha account and asked her to trade on NIFTY50 options. Fully with computer-use.
Alma opened the terminal, read the NIFTY charts, analyzed the option chain, placed an order, and closed it in profit - all under 60 seconds.
Alma will be the most capable computer use product in the world, stay tuned!
People keep saying Steve Jobs would vehemently dismiss someone’s idea, only to come back later and propose it himself without due credit.
The positive argument: he cared more about which ideas mattered to the product and less about who they came from. The negative argument: dark psychology and egotism.
What’s fascinating is that the behavior has both positive and negative outcomes. The positive: it self-selects for people who share Steve’s philosophy. The negative: it can be stupid, because recognizing / incentivizing the person who came up with the idea might actually motivate them to keep producing ideas of that quality.
The net impact on the world, regardless of all this, is truly fascinating.
11
614
Vinod Ganesan retweeted
Reviewing agentic coding traces late October from a model family's new release, @ArnhavD observed something that quickly kindled some discussion among us at @PatronusAI ...
He noticed the model's reasoning had nearly no articles and dropped a good number of other "function words" or "stop words" or, to call it simply, filler words, like the "be" in "be doing" and the "to" in "wants to fix".
The traces sounded, in his words, like Kevin Malone's lines from The Office: “Why waste time say lot word when few do trick?” (“Need inspect repo. Grep.”)
Another parallel is that they felt like stuff people would type into telegrams historically e.g. "Arrived safe. Weather rainy." or "Child born, mom healthy."
Intrigued now by the stark difference we had eyeballed b/w end of '26 and early '26 shortness of this kind in model reasoning, we sought to answer if this phenomenon held up more widely, and how to even compute it?
Specifically, was there a clear -ve drop Δ b/w newer late '26 vs older early '26 models from the top #3 open source families by open source use = Deepseek 4* Flash, GLM-5* & Kimi K3* ??
We computed what we called filler word % (articles, auxiliaries, self-talk, pronouns etc) in reasoning coming out of newer vs older models, when solving the same 20 tasks from three popular code benches.
In all the 3 families, filler share in reasoning basically halved, going from 25-30% to the low teens (10-14%)!
To conclude, we had 2 notable side-observations:
a) This didn't always cut total token usage (in some cases it rose) going from older --> newer model
b) The terseness is not quite of the human kind: self-talk e.g. "Wait!", "Eureka!!", "Maybe I'm right!" persists even when other filler words are getting dropped ...
Note that observations to this effect have already been made by many others, particularly in the case of Astra and Deepseek-Flash-4.1 traces by @scaling01 and GLM-5* by @Tim_Hua_ , as also posited by @karpathy as a possible endpoint of outcome driven reinforcement learning happening too well back in late '24 to early '25 itself
Nonetheless, we felt an independent repro through controlled comparison based on some formulation of filler word or "shortness" b/w older and newer models from the 3 top model families would enrich the narrative!
Full blog link in comments, glad to hear thoughts, questions and feedback!
Excited early user.
Pretty helpful in navigating my health goals. Long journey ahead for me and pretty long on @nuehealth_.
Check it out folks!
This video marks our announcement to the world.
Nue Health is now open for public access.
This has been more than a year in the making, but we are now by far the best blood test in the market.
i promise you, we're just getting started.
1
12
577
Vinod Ganesan retweeted
Replying to @Muse
@Muse talking about zucc
teapawt.org/timhwang/status/211507…
Vinod Ganesan retweeted
Replying to @HyphenateAI
@HyphenateAI just raised a $45M Series A at a $270M headline valuation.
We're bringing autonomous finance to every enterprise in the world. 🧵
Hyphenate Raises $45M Series A
Vinod Ganesan retweeted
An honor to be working for a cause central to the mandate of the nation with mission driven founders, a mission driven team, ex armed forces personnel and top technical talent that can accomplish the mission!
We are proud to partner with the Ministry of Defence on a joint framework to advance AI adoption across the Army, the Navy and the Air Force.
@pratykumar @vivekrag @AalekhSharan
So @Meta is crowd-sourcing AI intelligence (and compute) through openkernelchallenge.dev/ and incentivizing humans to use their AI to gain social credits (and $200k worth prizes).
alma can make music now
we asked @alma_inc to score steve jobs' stanford speech using garageband. it picked the instruments, built the whole track itself. fully with computer use.
stay hungry, stay foolish, stay funky!
The Pirates! 🏴☠️
at @alma_inc, we spend a lot of time rethinking interfaces, so tried retelling one of tech’s most famous stories in a different way.
the xerox parc visit.
9
497
Go from YouTube videos to the episode anywhere directly with @alma_inc.
The “ep??” guy in every comment section is free now.
Showed @alma_inc a Demon Slayer clip and asked it to play that episode. Fully with computer-use.
It found the right episode, opened crunchyroll, closed the pop-ups, skipped the recap and jumped to the exact scene.
1
10
625
With these superhuman model capabilities, people 'faking it' appears like 'making it' many times, reducing signal to noise ratio significantly.
195
Congrats folks! Great stuffs brewing at @clueso
We’ve raised a $10M Series A to automate video creation for every company.
1,000+ organizations like Atlassian, GitLab, Samsara, Legrand, and Zendesk use @clueso to automate their videos for marketing, customer education, and employee training.
The struggle of video creation in companies is so real, that Netflix made a documentary about it
1
22
826




















