Research Fellow, Consultant and Developer.
Joined April 2009
- Tweets88
- Following124
- Followers59
- Likes36
Our #NeurIPS2026 paper borrows a 1913 finding from social psychology, the Ringelmann effect, to measure how much LLM agent teams really gain from adding agents. 🧵
Across 44 conditions (debate, self-correction, self-consistency, Qwen/Llama/Ministral 7B–32B, Gemini, thinking models, mixed teams) the same form fits at R² > 0.99. Only c and β move.
Only diversity escapes the ceiling: mixing model families lowers c. More talk does not.
📄 arxiv.org/abs/2606.02646
📊 14 models, live: agenticscaling.ai/evidence.h…
We derive a two-parameter law for effective team size: R(N) = 1 / (1 + c(N−1)N^−β) c = how redundant the agents are β = whether adding agents still pays off: hard ceiling, sublinear, or linear
Carolina Fortuna retweeted
My problem with Astra reviews is that they focus almost entirely on intra-paper details: experimental rigor, missing ablations, and whether every local claim is airtight.
But good reviewing also needs an inter-paper view: what can we learn from this work, how does it change how we think about the problem, and what future research can it enable?
Research is more than a checklist.
If ICLR gets flooded with reviews that only see the intra-paper details and miss the bigger picture, we’re pretty cooked.
Last year we formalized how the cost of training ML models amortizes over their lifecycle through the eCAL metric published in IEEE JSAC and available as a python package. Now we generalized to agentic systems: arxiv.org/pdf/2609.18283v1
Thrilled to share that our paper has been accepted to the NeurIPS Main Track! 🎉 If you put 30 LLM Agents in a room to debate, they don't get 30x smarter, they basically form a committee.arxiv.org/abs/2606.02646
Carolina Fortuna retweeted
ICLR limited first-time authors to just one submission this cycle. I've gotten multiple emails now to be added as an author on papers purely to bypass this limit. Senior authors were already added to papers purely for prestige; the rate of paper laundering is going to go up.
Carolina Fortuna retweeted
TMLR has faced a deluge of submissions, necessitating stricter desk rejection policies due to limited reviewer capacity
Co-EiC Nihar Shah reached out to authors of 10 papers slated for desk reject. Could they answer questions about their *own* submission?
medium.com/@TmlrOrg/asking-a…
Are multi-agent debates actually just expensive groupthink? 🤔
From the inaugural ACM CAIS in San Jose we show that
sycophancy and consensus collapse are real. Read how self-correction prevails: 📄 Paper: lnkd.in/dMenz6xe 💻 Code: lnkd.in/dq8zCkkK
Carolina Fortuna retweeted
If you figure out how a large multi-agent system can autonomously coordinate to maximum effect, you’ll win a Nobel Prize, a Turing Award and a trillion-dollar fortune all in one.
🚀 Hiring: 2 MSCA PhD Positions in AI & Networking! 🎓
The Roles:
1️⃣ PhD 1: Neuro-Symbolic Orchestration (Personalized Home Spaces)
2️⃣ PhD 2: Neural Algorithmic Reasoning (Business Digital Spaces)
📎 Apply by April 10: genome-msca.eu
Please RT/Share. 🔄
Carolina Fortuna retweeted
Sufficiently advanced agentic coding is essentially machine learning: the engineer sets up the optimization goal as well as some constraints on the search space (the spec and its tests), then an optimization process (coding agents) iterates until the goal is reached.
The result is a blackbox model (the generated codebase): an artifact that performs the task, that you deploy without ever inspecting its internal logic, just as we ignore individual weights in a neural network.
This implies that all classic issues encountered in ML will soon become problems for agentic coding: overfitting to the spec, Clever Hans shortcuts that don't generalize outside the tests, data leakage, concept drift, etc.
I would also ask: what will be the Keras of agentic coding? What will be the optimal set of high-level abstractions that allow humans to steer codebase 'training' with minimal cognitive overhead?
Carolina Fortuna retweeted
This is a great speech. Worth watching the whole thing.
WATCH: Secretary Rubio Delivers Remarks to the Munich Security Conference in Munich, Germany. teapawt.net/i/broadcasts/1mnGeNNPq…
Carolina Fortuna retweeted
There are two kinds of theorists in AI: those who drift toward experimental research and those who come to terms with their irrelevance.
Carolina Fortuna retweeted
We’re building at a @Stanford hackathon, supported by @GoogleDeepMind, and in 3 hours we have to ship a working MRP engine— something manufacturers usually pay millions for.
That wasn’t the hard part.
The hard part is data that captures intent and real business logic. Most ERP systems are just expensive databases that rely on people to type reality into software.
We’re here to build systems that understand how manufacturing actually works — not how it’s documented.
#Stanford #DeepMind #Manufacturing #AI #Hackathon #ERP
Carolina Fortuna retweeted
Recorded a quick tutorial on connecting Nordoon to Outlook. If your POs come in via email (most do), this saves a lot of copy-paste. youtu.be/ndnPdE7rGlE
