^^Artificial intelligence.

read “Style in Scientific Writing” here.

rob: I computer hanno sostituito il capo contabile della banca (negli anni 1970-80), le Artificial Intelligences ...
L'intelligenza artificiale ruberà il lavoro (ed è giusto che sia cosi) a chi non la userà !
prova Warp invece del terminale standard. Niente più copia incolla dei comandi, parli col terminale e gli fai creare al volo script e comandi bash 😉

AI matematica.

AI transumana.

2026-03-31 &t=458 Mistral just entered the voice race with Voxtral,

2026-07-10 yt I Built an LLM From Scratch

Physicist Mario Rasetti

2026-06-13 yt The Future of Artificial Intelligence: An Interview with Physicist Mario Rasetti

2026-05-15 Yann LeCun on What Comes After LLMs

FAIR := Facebook AI Research, O Meta's Fundamental AI Research; Yann LeCun : Co-founded FAIR alongside Rob Fergus in December 2013, Founding Director & , left the diection after 4 year and 1/2, and became Chief AI Scientist.

VLA Visual Language Action Model

World Model := abilita' nel predire-prevedere le conseguenze delle proprie azione

VAE Variational Autoencoders, or autoencoders

JEPA Joint Embedding Predictive Architecture

LeCun famously worked alongside AI leader Jeff Dean, Google's chief scientist and head of Google AI.

Yann LeCun says:

Ther's always going to be a prompt where the system is going to do really stupid things.
PyTorch: done by FAIR; entire industry is built on PyTorch basically.
There are more restrictions on pubiication now in Meta and Google.
If you're doing a Ph.D. you're not going the GPUs you need for work on LLM.; so, like, forget that; like don't work on LLM.
In a keynote ad NeurIPS 2016 i said that for human level AI we need
1. self-supervised learning
2. prediction from sensory signals, like video
3. world models, they predict consequences of yout own actons and plan
Reinfocement Learning is not the thing that will take us there cuz is too inefficient; supervised learning has shown its limits, and so, the future il self-supervised learning and world models.
D: What's like one thing you have changed your mind on in last year?
R: I mean, the whole idea what we used to call "unsupervise learning" that we now call "self-supervised learning"

2026-03-16 Leanstral first open-source code agent designed for LEAN

yt&t=581 Mistral introduced Leanstral for checking and fixing code through formal reasoning,

https://mistral.ai/news/leanstral

2026-02-14 Google Gemini 3 DeepThink Is Now the Smartest AI In The World yt

il video dura 22min, merita di essere visto tutto, perlomeno per me che non avevo mai visto risultati di benchmarks

Visione breve per chi codifica

iniziare a 2:15. benchmark "Codeforces" (competitive programming community), e se si guarda solo la tabella del risultato, gia' finito, il commento sottostante dura qualche minuto.

Esempi d'uso fatto in campo scientifico 8:42

es: circa 6 mesi fa AI era in grado di risolvere problemi di matematica a livello gold medal IMO (International Mathematical Olympiad), ora e' salito a livello dei problemi affrontati per il PhD (Philosophiae Doctor o Dottorato di Ricerca è il più alto grado scolastico universitario ottenibile, focalizzato sulla ricerca scientifica originale, dura solitamente di 3-4 anni).

2026-01-22 Sergey Karayev: "Claude code Opus 4.5 is a watershed moment",

moving software creation from an artisanal, craftsman activity to a true industrial process. It’s the Gutenberg press. The sewing machine. The photo camera.
David Blundin: "The amount of code I've created in the last couple of months is bigger than my entire life combined"

yt&t=2133 poi per quasi 15 min
Salim Ismail, Alexander Wisser-Gross, Peter H. Diamandis, David Blundin
anthropic/claude-opus-4-5

2026-01-20 yt Demis Hassabis and Dario Amodei, debate AI at Davos World Economic Forum.

Previsioni sull'AI 2026, penso piu' affidabili. Se hai poco tempo, possono bastare 2 minuti da 1:42

1:42 Dario Amodei have Engineers within Anthropic who say "I don't write any code anymore. I just let the model write the code. I edit the code."
1:52 Dario predicts models might be 6 to 12 months away from doing most, maybe all, of what SWEs (Software Engineers) do end-to-end.

penso che la visione di Amodei-Anthropic sia sbilanciata dalla sua specializzazione nel coding.

Dario Amodei (nato 1983) CEO of Anthropic

Demis Hassabis, Deep Mind co-founder & Ceo.

Terminologia

yt Demis Hassabis: AGI vs Superintelligence (And Why We're Not There Yet)

yt Figure Just DROPPED the Most ADVANCED AI ROBOT BRAIN — Meet Helix 02

Terence Tao & Riley Tao: Turning AI’s Firehose Into Usable Science

11:15 beautiful explanation on why benchmarks are useful until they are not.
yt Terence Tao, inaugural Veritas Fellow — autoformalizing number theory | Math, Inc.

-----------------------------------------------------

2026-02-10 yt Is This Reasoning Energy Based Model The Next AI Breakthrough? w/ Eve Bodnia of Logical Intelligence
2025-12-08 yt NeurIPS 2025: Top 3 Highlights

wp/Conference_on_Neural_Information_Processing_Systems

AI agent >>>

AI assistant, co-work

AI matematica >>>

Ricadute AI

2025-10-26 yt 20 PhDs’ In the Time of One: How AI Is Changing College

yt Mind-Blowing Theoretical Physics with DeepSeek

Info particolari

AI Explained https://www.youtube.com/@aiexplained-official/videos
yt AI, Machine Learning, Deep Learning and Generative AI Explained
ieee/rodney-brooks-ai An Inconvenient Truth About AI: AI won't surpass human intelligence anytime soon
ieee/deep-learning-computational-cost.
es: dimezzare gli errori di riconoscimento fotografico richiede di moltiplicare per 500 la potenza di calcolo
yt China's slaughterbots show WW3 would kill us all.

yt Expert shows AI doesn't want to kill us, it has to.

Biocomputer

wp/Long-term_potentiation

2026-03-14 yt Scientists Put a Fruit Fly’s Brain Into a Computer Simulation — What It Did Next Shocked Everyone...
- https://eon.systems/updates/embodied-brain-emulation How the Eon Team Produced a Virtual Embodied Fly
- Scientists have achieved an incredible breakthrough by recreating the brain of a fruit fly inside a computer simulation. By mapping around 140,000 neurons and millions of connections, they built a digital brain that can sense its environment, process information, and even control a virtual body. In the simulation, the digital fly was able to search for food, respond to stimuli, and show behaviors that were not directly programmed by scientists. This discovery shows how powerful neural connections are in generating behavior. It also raises fascinating questions about the nature of intelligence, consciousness, and whether complex brains—including ours—could one day be simulated in computers.
2026-03-02 Il computer con 200.000 neuroni umani che gioca a Doom
E consuma molto meno di una IA tradizionale. - yt

Limiti AI

Quando l'intelligenza artificiale bara: l'aneddoto degli husky scambiati per lupi. zeusnews

AI filds

wp/Long_short-term_memory
LLM Large Language Model
Pattern recognition wp/Pattern_recognition
A modern definition:

The field of pattern recognition is concerned with the automatic discovery of regularities in data through the use of computer algorithms and with the use of these regularities to take actions such as classifying the data into different categories.
Machine learning; apprendimento automatico.
1. boosting
ibm/autoregressive-model

La ricerca di una intelligenza artificiale generalista e' stata innescata da ...

“Single-brain”-style models have interesting connections to neuroscience. Mountcastle (1978) famously stated that “the processing function of neocortical modules is qualitatively similar in all neocortical regions. Put shortly, there is nothing intrinsically motor about the motor cortex, nor sensory about the sensory cortex”. Mountcastle found that columns of neurons in the cortex behave similarly whether associated with vision, hearing or motor control.

This has motivated arguments that

we may only need one algorithm-model to build intelligence
(Hawkins and Blakeslee, 2004).

Conclusions

In the future, such models could be used as a default starting point via prompting or
fine-tuning to learn new behaviors, rather than training from scratch.