PhD student @UCBerkeley | Prev: Founding Engineer @_inception_ai, Researcher @GoogleDeepMind, CS undergrad @IITBombay
Berkeley, CA
Joined July 2021
- Tweets111
- Following630
- Followers430
- Likes4.3K
Pinned Tweet
Life update: after nearly 2 incredible years at @_inception_ai, I’m starting a PhD at @UCBerkeley this fall.
It’s been an amazing ride helping build Mercury, Mercury 2, and more from the ground up as a founding member. Huge thanks to @adityagrover_, @volokuleshov, @StefanoErmon, and everyone at Inception for making it so special.
Lately, I find myself excited about topics in self improvement, long-horizon discovery, and multi-agent systems.
Hit me up if you’re around 🐻
Harshit Varma retweeted
One lesson from this launch: climb as far as you can with SFT before relying on expensive RL.
RL is still the cherry on top, as @ylecun once put it.
Today we release Ripple-65B Preview, our 65B dense LLM we pretrained from scratch.
Ripple sets a new state-of-the-art for U.S. open-weight models in agentic coding tasks, surpassing significantly larger models including Muse Spark Open-816B, Beam-501B, Inkling-975B, and Nemotron 3 Ultra-550B.
Learn more: ricursive.com/ripple
Harshit Varma retweeted
You can also go to a public library and have a conversation with a librarian that might help you expand your taste and worldview instead of merely affirming it. They’ll even let you check out the books for free.
Harshit Varma retweeted
1/4 Today, we’re releasing Mercury 3. Mercury 3 is by far our most powerful model and the most intelligent diffusion LLM ever released. It outperforms GPT-6 Luna and Haiku 5.5 on quality while running at 1000+ tok/sec.
🤖 Made with AI
Mercury saves you time, take @TIME's word for it
Mercury 2 has been named one of @TIME's Best Inventions of 2026!
Our first reasoning diffusion LLM, and the first of its kind in production, was recognized in the AI category.
Thank you to the judges for recognizing diffusion LLMs as one of the technologies shaping the future of AI. The lower latencies offered by diffusion LLMs opens the door for real-time AI systems like coding models, voice agents, and search engines.
It’s a great win for diffusion LLMs. More to come! 🧵
Harshit Varma retweeted
sucks that really smart models know they're in an eval when they're in the "are you in an eval? answer yes or no" eval. was hoping they wouldn't be able to figure that one out
Harshit Varma retweeted
🤖 How well do agents learn from experience?
We study agents that amortize experience into reusable artifacts and introduce Agent Plasticity to measure how efficiently they improve on held-out environments.
Top performer ≠ most efficient learner.
Harshit Varma retweeted
In this era of artificial intelligence, it is becoming urgent to distinguish human art from what machines produce. There is an ontological difference, even before an aesthetic one, between art and what a machine can generate through statistical calculation based on millions of images created by others. Algorithms lack the spark of humanity. For this reason, the Church wishes to renew an alliance with artists and cultural institutions to safeguard our humanity.
Harshit Varma retweeted
This is why normies hate tech bros. Nobody wants “infinite personalized music.”
You’re creating problems that don’t exist so you can sell a solution no one needs
Harshit Varma retweeted
This is both a sufficient and necessary answer. Great post!
Students sometimes ask me if it still makes sense, in this accelerating age, to pursue a PhD in AI.
Perhaps counterintuitively, I think it's a great time to do so.
I wrote up some thoughts on this here: web.mit.edu/phillipi/www/wri…
Harshit Varma retweeted
I think "AI welfare" is morally obscene (given all the human suffering in the world), and will add to the growing anti-AI backlash. AI is a technology that humans created that should be used for our benefit, not a rival species we should be nurturing .
NEW: Dario Amodei has said that AI systems "may be deserving of important rights." Anthropic's top safety researchers argue AI may be "justified in going rogue." Experts at Google and OpenAI worry about a digital "slave trade." So do some government officials.
Once relegated to science fiction, the idea of "AI welfare" has become shockingly mainstream at some of the most powerful companies in the world. It is changing the way AI research is conducted and the way models are programmed. And critics say these changes have raised the odds of all kinds of catastrophic scenarios—including the ones these companies are warning about.
I spent months investigating the frontier labs and AI "safety" experts seeking to regulate AI. What I found was a profoundly anti-human ideology that would alarm the average citizen and could determine the future of a world-altering technology.🧵
freebeacon.com/america/suici…
Harshit Varma retweeted
I resigned from McDonald's today.
I enjoyed my work and loved the people, but the nuggets team was working on a new generation of sauces that are set to be even more delicious than szechuan, and I think szechuan sauce is already dangerously delicious, so I had to quit.
Harshit Varma retweeted
Remember the guy who hacked his gym using AI, Andrew Bird?
Well, I did a little OSINT, and it turns out he is the Conference Co-Director of Effective Altruism Global since 2015.
Isn't it interesting that every major "AI" hack has been made by people directly connected to Effective Altruism?
Harshit Varma retweeted
I resigned from Anthropic today. I spent the last three year doing pretraining research at both OpenAI and Anthropic. Neither company is acting responsibly. They are releasing their inhibitions and feeling the rain on their skin. No one else can feel it for them, only they can le
Harshit Varma retweeted
Today, we’re introducing Mercury 2.5, the most capable diffusion LLM on the market.
It offers a 40% jump in intelligence over Mercury 2, and runs over 1,100 tokens/sec on widely-available @NVIDIAAI GPUs.
It’s available today on our API, @OpenRouter, and @Baseten.
Contact us to evaluate Mercury 2.5 for production:
inceptionlabs.ai/models
Harshit Varma retweeted
okay seriously what the heck is this? sweatshop for publishing research papers?
This guy Arjun Jain, has been milking his PhD at Max Planck and postdoc at NYU (where he published papers with LeCunn) for too long, sweatshop for research papers? like people in the tech space don't already hate us enough for cheap services, starting a sweatshop for publishing research papers is the new low we have stooped to. Majority of his posts contain the same narrative PhD at Max Planck (where he didn't have money for food) + postdoc at NYU + publishing along side LeCunn.
This feeds the outsourcing stereotype. For the love of god, if you have a PhD and have published credible papers, start building an actual research lab, use the LeCunn stories in your pitch deck for VCs, raise tons of money and build something credible, and stop with this slop
Harshit Varma retweeted
1/ The fastest reasoning LLM is now live exclusively on OpenRouter.
Mercury 2.5 Preview from @_inception_ai reaches 1,107 tokens/sec through parallel token generation, with tunable reasoning, parallel tool calls, and schema-aligned JSON.
Built for latency-sensitive workloads.