Mohit Kulkarni 🧠 retweeted
Replying to @CLESZN22 @dwarkesh_sp
There’s something deeply human about building something that helps us see a little further into a world we didn’t choose to enter but want to make sense of.
Mohit Kulkarni 🧠 retweeted
i don’t think you guys understand. you will get the chance to go to space in your lifetime. new era
Mohit Kulkarni 🧠 retweeted
I bet against this and I bet against Dwarkesh.
He conveniently sweeps under the rug that climbing a metric involves defining it. Then, supposing you define it (in a way that avoids Goodharting, reward hacking, etc), you need data/envs, which do the lifting. On defining good metrics - this is most of the game of making AGI (converting fuzzy concepts to formalized objects you can climb without gaming, which requires costly fixes later). He elides the hard problem by saying 'AI will just automate data'.
How do you know whether a change belongs in your final recipe? You ablate at scale. When you make N changes, on even a subset of combinations, this compute cost is naively exponential. So, you need a faster signal to tell you when a task that isn't immediately verifiable improves end behavior. Finding that signal is the bottleneck, which doesn't seem to be helped by 'automating the great algorithmic thinker starting from the empty string on paper'.
Mohit Kulkarni 🧠 retweeted
With Instinct bringing AI agents to iMessage groups (and planning to raise at $30B), this seemed like as good a time as any to share our NeurIPS 2026 paper on the security risks of AI agents in group chats.
Turns out, putting multiple users into the same agent's context can lead to some pretty interesting failures.
A thread 🧵
Instinct in group chats is now available to everyone on iMessage.
Add Instinct to a group chat, and it works for the whole group, including for friends who don’t use Instinct yet. The group gets its own Instinct that only knows what’s in the chat, not anyone’s accounts or data. When it needs something from you, it goes through your personal Instinct, which confirms with you first.
A few things early access groups have used it for:
- Planning a dinner party with friends
- Organizing a road trip
- Decorating an apartment
Add Instinct to any group chat to get started.
@GajanNagaraj @anuda_w @saagarwashere
Mohit Kulkarni 🧠 retweeted
Working with AI models, I've arrived at what I consider the most significant pure-theory breakthrough of my research (in theoretical neuroscience) so far:
arxiv.org/abs/2610.12426
Chaotic systems show what is colloquially called the butterfly effect, namely, a tiny nudge to the initial state grows exponentially in time. How fast the nudge grows depends on its direction. Each direction has its own growth rate, called a Lyapunov exponent, and the full set of rates is the Lyapunov spectrum.
The Lyapunov spectrum is a rich (and diffeomorphism-invariant) characterization of chaos, telling us how many directions are chaotic, how many dimensions the attractor occupies, and how fast the dynamics generate information.
In 1988, Sompolinsky, Crisanti, and Sommers (Physical Review Letters) introduced a nonlinear recurrent-network model of N randomly coupled neurons that becomes chaotic when the coupling strength exceeds a critical value. This network has since become a cornerstone of theoretical neuroscience, both as a model of the spontaneous activity of cortex and as the starting point for training recurrent networks on tasks. I have worked on this model for much of my research.
In the same paper, the authors obtained the largest Lyapunov exponent from the ground-state energy of a Schrödinger equation and named the full Lyapunov spectrum as an open problem. In 2023, beautiful work by Engelken et al. (Physical Review Research) computed the full Lyapunov spectrum numerically. However, an analytical calculation of the full Lyapunov spectrum has remained out of reach.
Four decades after the problem was posed, this paper derives the full Lyapunov spectrum at large N from a simple self-consistent single-site problem.
This calculation has three steps.
First, we lower every exponent by a constant s, which turns the question "how many exponents lie below s?" into the question "how many tangent directions decay?" We show that, if we drive the shifted tangent dynamics with a source and take the minimum-norm solution, the one-step response is the projector onto the decaying directions. Thus, the trace of this projector counts the decaying directions. This step is exact at finite N.
Second, to work with the minimum-norm solution, we cast it as the vanishing-regularization limit of the solution of a least-squares problem. This regularized solution in turn solves a dynamical system with one field running forward in time and another running backward.
Third, we apply the cavity method, that is, we add one neuron to the network, along with its forward and backward fields. The effect of the rest of the network on this neuron then reduces to the classical dynamical mean-field description of the activity of this neuron, plus two kernels that couple the forward and backward fields of this neuron to each other across time. Thus, the full Lyapunov spectrum follows from a single-site problem, which boils down to the self-consistent calculation of these kernels.
The resulting theory agrees closely with simulations for the Lyapunov spectrum, the attractor dimension, and the entropy rate, and recovers known results, including the Schrödinger equation of Sompolinsky, Crisanti, and Sommers for the largest Lyapunov exponent.
As a byproduct, the theory establishes analytically that the chaos is extensive, i.e., the attractor dimension and the entropy rate grow in proportion to N.
I had assumed that this problem was analytically intractable. After working independently for 100 minutes with Max thinking, GPT-6 Astra produced the initial derivation, yielding a self-consistent problem whose solution matched simulations of large networks.
I found this derivation intriguing, albeit complicated and somewhat mysterious (see AI Methodology in the paper). After investigating the structure of the proposed solution more deeply, together with AI models, I arrived at a physically transparent derivation.
I am excited to apply this approach to other network models, trained networks, and, more broadly, to high-dimensional systems to which Lyapunov analysis is relevant, including AI models themselves. Of course, as with any paper, I took care to write up the paper clearly.
A few broader thoughts...
Neural circuits are high-dimensional nonlinear dynamical systems, and such systems are notoriously hard to understand. I would frame the central problem of theoretical neuroscience as understanding how such systems function, process information, learn, and generate behavior.
Many people in the theoretical sciences are anxious about AI. However, in my field at least, we are nowhere close to fully understanding the high-dimensional computations unfolding in our brains. With AI, we could reach a level of understanding that so far we have only gestured toward. So, I am excited.
Mohit Kulkarni 🧠 retweeted
I may have found two new Earth-sized planets in our galaxy this morning!
Inspired by yesterday's reddit thread on how someone used Claude Code to find a new planet, I spent this morning searching NASA TESS data on nearby stars that TESS scanned less thoroughly (at low cadence).
To my surprise the pipeline discovered two previously undocumented exoplanet candidates:
• TIC 370612437 (Cygnus constellation, 158 light-years): dims by 0.02% every 3.56 days → ~1.0× Earth's size.
• HIP 114417 (Andromeda constellation, 134 ly): 0.03% every 4.93 days → ~1.1× Earth.
The method to find them is to search for periodic dims in light coming from a star when a planet passes in front of it. For both the candidate exoplanets, the dip repeats 55 and 21 times respectively across 4 separate observing years (2019-2024).
These are obviously tentative and unconfirmed, but several signals point to them being credible candidates:
• same dip in NASA's and MIT's separate processing of the images
• hide one year's dip and predict its times from rest of the dips (matched in 3 of 4 years)
• flip light curves to create fake signals and validate observed dips against it
The method and results were reviewed by independent agents: Codex and fresh Claude. Important caveat: a faint neighbouring pair of eclipsing stars could fake the dip, so these exoplanets remain candidates pending further verification.
What's next: report to NASA's ExoFOP as Community TESS Objects of Interest, so astronomers can verify. The next scan of these stars via TESS will happen in Aug 2027 - Apr 2028, if the preregistered dip predictions hold up, these are indeed strong candidates.
Nicknames for these exoplanets: Aakanksha (Sanskrit: aspiration) and Orbit (well, Orbit)
Mohit Kulkarni 🧠 retweeted
Replying to @dhruvtrehan9 @lossfunk
let knowledge be freed from the confines of human mind, let us all revel in the plurality of forms, let us all be untethered from the bounds of reason and understanding like we once were from unreason and dogma, if they usher a dark age so be it, things must burn before they are built anew
Sometimes you just gotta appreciate the way a paper is written
1/ Today, we introduce Faraday, a 27B-parameter AI Scientist that extends the capabilities of coding agents with a layer of scientific intuition. Trained via long-horizon RL, Faraday outperforms Claude Opus 4.8 and GPT-5.5 on the task of replicating research papers. 🧵
2
195
Mohit Kulkarni 🧠 retweeted
1/ Today, we introduce Faraday, a 27B-parameter AI Scientist that extends the capabilities of coding agents with a layer of scientific intuition. Trained via long-horizon RL, Faraday outperforms Claude Opus 4.8 and GPT-5.5 on the task of replicating research papers. 🧵
Mohit Kulkarni 🧠 retweeted
life is not about success, or power. really intelligent people recognize the futility of life very early and from then on its all about the number of ice-cream scoops youve demolished
Mohit Kulkarni 🧠 retweeted
⏰ 3 days left until the deadline for the Conference for AI Scientists (CAISc) 2026.
With awards incl. model and compute credits over $20,000, this is a great chance to get feedback on ongoing research.
Our ask - use an LLM or AI system in some part of your research workflow.
huge. @NeocambrianAI is upto some amazing stuff
Today, we announce @NeocambrianAI
The future of AI will be physical. But Physical AI has no internet scale dataset to learn from.
We’re building the data foundation of Physical AI - a high fidelity, pre training scale database of Human Action.
From India. For the world.
Mohit Kulkarni 🧠 retweeted
this and daniel gross's agitrades essay keeps ringing in my head
"If you’re India, for example, where double-digit percentages of your GDP are literally IT services, what do you do when Claude and GPT-5 tokenize like vast portions of that flow"
GOI should be GPU-maxxing
Mohit Kulkarni 🧠 retweeted
1/6 Introducing Seesaw: a principled batch size scheduling algo. Seesaw achieves theoretically optimal serial run time given a fixed compute budget and also matches the performance of cosine annealing at fixed batch size.
Mohit Kulkarni 🧠 retweeted
SSMs promised efficient language modeling for long context, but so far seem to underperform compared to Transformers in many settings. Our new work suggests that this is not a problem with SSMs, but with how we are currently using them.
Arxiv: arxiv.org/pdf/2510.14826
🧵

























