
CTO and Cofounder of @FocoosAi. PhD in Computer Vision and Continual Learning at @PoliTOnews. Past president of IEEE @HKNPoliTo Mu Nu Chapter.
Piedmont, Italy
Joined July 2011
- Tweets419
- Following430
- Followers727
- Likes1.8K
Fabio Cermelli retweeted
Don't understand all the AI jargon everyone around you keeps saying? You're welcome, I made the updated AI dictionary 🥳🥳- :
- The bitter lesson - scale beats everything else, especially your clever idea
- Brain-inspired - we read one neuroscience abstract in 2019
- AGI - whatever the current models can't do yet
- Superintelligence - AGI, but the last name was taken
- Self improvement - letting a coding agent run your experiments
- Recursive self improvement - the same thing but it sounds more impressive
- RL - it works now, we just don't know why this time
- Memory - a text file the agent appends to
- Continual learning - solved next year, every year since 2016
- Agent framework - the same model prompted five different ways and called a team
Novel architecture - a transformer
- Frontier model - our model
- Technical report - a paper with the methods section removed
- Emergent capabilities - a metric we didn't plot until it went up
- Neolab - a bunch of ex-{Meta, OpenAI, GDM} people who think they know better
- We fired because of AI - we did not fire because of AI
- We're hiring - we raised
Hope this helps. See you in the next edition, the field should generate enough new terms by Friday!
Fabio Cermelli retweeted
Today, we’re launching our home robot Isaac 1.
Isaac 1 deliveries will begin this fall.
Order yours below.
#CVPR workshop days shouldn't require 20 browser tabs.
So I built CVPR Workshop Radar 🚀
🔎 Search for workshops & tutorials
⭐ Save your favorites
📅 Explore schedules in one place
Open source, mobile-friendly, no account required.
cvprworkshopradar.vercel.app…
#CVPR2026 @CVPR
Fabio Cermelli retweeted
This is where we are right now. And i’m not gonna lie it feels pretty magical 🧚♀️
Qwen3.6 27B running inside of Pi coding agent via Llama.cpp on the MacBook Pro
For non-trivial tasks on the @huggingface codebases, this feels very, very close to hitting the latest Opus in Claude Code, or whatever shiny monopolistic closed source API of the day is.
In full airplane mode.
Most people haven’t realized this yet.
If you have, it means you have a huge headstart to what I call the second revolution of AI.
Powerful local models for efficiency, security, privacy, sovereignty 🔥
Fabio Cermelli retweeted
What if a model could learn dense semantic matches from just a handful of annotated landmarks, while still generalizing to unseen keypoints and categories — and running 10× faster than diffusion-based approaches?
MARCO is selected as an Oral at #CVPR2026! A unified model for generalizable semantic correspondence, built on DINOv2⭐️
👉 Try our model: github.com/visinf/MARCO
Fabio Cermelli retweeted
Launching Lens, our fine-tuning service for Moondream.
Improve Moondream's accuray to production-ready levels in hours, with as few as 20 images.
Example: we taught Moondream to find the player with the ball in NBA footage. F1 jumped from 28% to 79%, beating ChatGPT. 54 minutes, $16.89.
Fine-tune your own: moondream.ai/blog/lens-moond…
Fabio Cermelli retweeted
Should you SHOW 🖼️ or TELL 📝 a model what to segment? 🤔
Our new #benchmark compares visual vs textual prompts for semantic segmentation across 14 datasets spanning 7 domains!
Check out our findings ⬇️
Fabio Cermelli retweeted
The Hidden Cost of LLMs: What Happens Under the Hood? A Thread 🧵
You’ve probably used an genAI assistant today, certainly powered by LLMs. But have you ever wondered what does it take to answer you? 🤔
It's not just computation—it’s energy and money. Let’s break it down
Fabio Cermelli retweeted
Il ministro @adolfo_urso ha incontrato Antonio Tavera e Fabio Cermelli, CEO e CTO di #FocoosAI, startup vincitrice del Premio Nazionale Innovazione 2023-categoria ICT, che ha sviluppato #Anyma, rivoluzionario software nel campo dell'#intelligenzaartificiale. – at Roma, Lazio
Fabio Cermelli retweeted
Ummm ... why is this a surprise? Transformers are not elixirs. Machine learning 101: gotta cover the test distribution in training!
LLMs work so well because they are trained on (almost) all text distribution of tasks that we care about. That's why data quality is number 1 priority: garbage in, garbage out. Most of LLM efforts these days go into data cleaning & annotation.
Fabio Cermelli retweeted
VANDAL lab from @PoliTOnews reunited at #ICCV2023 🇨🇵😍 @fcdl94 @gabriberton @gabTrivv @tommasi_tatiana @neikos00 @ShyamNandanRai4 @chiaraplizzari and Claudia Cuttano 🤓
Thanks to the one and only @CSProfKGD for the picture ❤️
Fabio Cermelli retweeted
2009-2019 memories ... thanks to all collaborators who made @PAVIS_IIT successful, working hard, having fun 😃 #IIT20Y
Fabio Cermelli retweeted
Unified Model for Image, Video, Audio and Language Tasks
paper page: huggingface.co/papers/2307.1…
Large Language Models (LLMs) have made the ambitious quest for generalist agents significantly far from being a fantasy. A key hurdle for building such general models is the diversity and heterogeneity of tasks and modalities. A promising solution is unification, allowing the support of a myriad of tasks and modalities within one unified framework. While few large models (e.g., Flamingo (Alayrac et al., 2022), trained on massive datasets, can support more than two modalities, current small to mid-scale unified models are still limited to 2 modalities, usually image-text or video-text. The question that we ask is: is it possible to build efficiently a unified model that can support all modalities? To answer this, we propose UnIVAL, a step further towards this ambitious goal. Without relying on fancy datasets sizes or models with billions of parameters, the ~ 0.25B parameter UnIVAL model goes beyond two modalities and unifies text, images, video, and audio into a single model. Our model is efficiently pretrained on many tasks, based on task balancing and multimodal curriculum learning. UnIVAL shows competitive performance to existing state-of-the-art approaches, across image and video-text tasks. The feature representations learned from image and video-text modalities, allows the model to achieve competitive performance when finetuned on audio-text tasks, despite not being pretrained on audio. Thanks to the unified model, we propose a novel study on multimodal model merging via weight interpolation of models trained on different multimodal tasks, showing their benefits in particular for out-of-distribution generalization. Finally, we motivate unification by showing the synergy between tasks
Fabio Cermelli retweeted
Research is hard: if there's no struggle, you're doing it wrong - and it doesn't get any easier with seniority :)
Fabio Cermelli retweeted
Saw this AI post trending on LinkedIn.
The best way to have a ‘competitive edge’ isn’t to ‘use AI to write your resume‘ or to ‘get AI to automatically mass apply to jobs’ for you.
Instead, network in the real world. Meet people in real life and ask them to have coffee with you.
Fabio Cermelli retweeted
Pending #CVPR2023 in June, we are pleased to share our 4 accepted papers
(3/4) "CoMFormer: Continual Learning in Semantic and Panoptic Segmentation"
by @fcdl94, @quobbe, @Ar_Douillard
preprint: arxiv.org/abs/2211.13999
Collab w/ Politecnico di Torino and @heuritechlab
Fabio Cermelli retweeted
Perché l'intelligenza artificiale è la nuova elettricità?
@bcaputo_iit, docente del #DAUIN, spiega quanta energia consuma l'Intelligenza Artificiale e quanto inciderà questa tecnologia sul futuro del pianeta.
#rassegnastampa #PoliTo via @CorriereAlpi
📺video.corrierealpi.gelocal.i…


































