^^AI programmi open source, commerciali.

SOTA state-of-the-art

BitNet b1.58 e' un LLM caratterizzato da 3 pesi (-1, 0, 1)

invece dei tradizionali (2026) 16 o 32 bit. Prestazioni paragonabili ma con dimezzamento nel consumo energetico e un'efficienza drastica sulla CPU.

1. yt BitNet b1.58 Explained | The End of GPU-Dependent AI?
2. https://github.com/microsoft/BitNet

yt I Tried Google's 7 FREE AI Tools (Destroys All Paid Alternatives)

ChatLLM https://chatllm.abacus.ai/

One AI assistant with access to all top LLMs, video and image generators. General purpose and coding agents included.

https://www.youtube.com/watch?v=RcIWhm16ouQ

NotebookLM

https://notebooklm.google/

Come usare LLM in locale

1. 2026-04-03 yt  Come usare LLM in locale sul tuo PC
2. 2026-03-05 yt  IA locale GRATUITA di Qwen 3.5 - Simone Rizzo
3. 2026-03-16 yt  Qwen 3.5 su Mini PC da 200€: LLM Locale Senza GPU - Alessio Garau.  Mostra come fare, iniziando dalla configurazione di Ollama.

Clone voice

1. 2025-09-25 https://www.youtube.com/watch?v=AiRksVoiUAI How to Clone Your Voice with AI - Realistic AI Voice Clones
2. 2025-08-01 yt How I Cloned My Voice & Avatar in 2025: The Complete Tutorial

OCR

1. 2026-04-28 yt  GLM-OCR: il nuovo SOTA per trasformare PDF in Markdown
2. 2026-02-21 yt OCR: il modello SOTA da 1.7B che fa tutto da solo. - Simone Rizzo

Timeline

1. 2026-04-24 yt Come usare Claude Code GRATIS al 100% (anche su computer economici)
2. 2026-04-15 yt With this new AI you can study every subject [LearnYourWay tutorial]
3. 2026-04-14 yt Caveman: skill che obblga AI a testo stringato. Il piu' stringato e' il cinese. - Simone Rizzo
4. 2026-04-07 yt NVIDIA’s New AI Just Changed Everything: e' aperto
5. 2026-04-06 yt Tutorial Duck AI: for your privacy
6. 2026-03-27 yt Claude ora usa il tuo computer come un umano (Demo Completa)
7. 2026-03-26 yt Google Just Dropped TurboQuant And Changes AI Forever

AI commerciali. Aziende.

Dario Amodei (nato 1983) CEO of Anthropic

Anthropic american AI company, founded in 2021 by Dario and his sister, along with other former senior members of OpenAI. As of November 2025, Anthropic has an estimated value of $350 billion  (1 billion = 1 miliardo).

GPT Generative Pre-trained Transformer.

GPT vs Gemini

yt dice equivalenti  1: GPT 5.2 20 dollari/mese, 2: Gemini 3 ultra  250 dollari/mese

kimi.com/kimi-claw: Deploy Your 24/7 Personal OpenClaw in Seconds

Claude

Claude Free. claude.com/pricing

Cowork   yt New to Claude Cowork? Start Here.

Claude Pro  20euro/mese  per avere Cowork.

support.claude/getting started

Domandare direttamente a Claude come fare per usarlo.

2026-04-07  Claude Mythos Preview, a watershed moment for security

originale https://red.anthropic.com/2026/mythos-preview/

segue mia sintesi. Vedi anche yt&t=387 6:27

2026-04-07  anthropic.com announces Claude Mythos Preview, a new general-purpose language model.
This model is STRIKINGLY CAPABLE at computer SECURITY tasks:

1. it has already discovered thousands of high-severity zero-day vulnerabilities in every major operating system and web browser.
   Some of these include: a now-patched 27-year-old bug in OpenBSD, a 16-year-old flaw in FFmpeg, and a memory-corrupting vulnerability in a memory-safe virtual machine monitor.
2. it has proven capable of turning vulnerabilities into exploits, also chaining together more vulnerabilities.
3. it has proven capable of reverse-engineering closed-source software.

In response, we have launched Project Glasswing, an effort to use Mythos Preview to help secure the world’s most critical software, and to prepare the industry for the practices we all will need to adopt to keep ahead of cyberattackers.

The model will be given in use only to a small set of organizations, including Amazon Web Services, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorgan Chase, the Linux Foundation, Microsoft, NVIDIA, and Palo Alto Networks, along with Anthropic, to secure critical software.

We see no reason to think that Mythos Preview is where language models’ cybersecurity capabilities will plateau. The trajectory is clear. Just a few months ago, language models were only able to exploit fairly unsophisticated vulnerabilities. Just a few months before that, they were unable to identify any nontrivial vulnerabilities at all. Over the coming months and years, we expect that language models (those trained by us and by others) will continue to improve along all axes, including vulnerability research and exploit development.

Conferme da Mozilla

• Firefox 150, scovate 271 vulnerabilità con l'IA. Mythos individua i bug e supera i ricercatori umani nell'analisi del codice. Leggi l'articolo originale su ZEUS News - https://www.zeusnews.it/n.php?c=32014
•  https://blog.mozilla.org/en/firefox/ai-security-zero-day-vulnerabilities/

AI open source

ChatGPT Salute

1. https://openai.com/it-IT/index/introducing-chatgpt-health/
2. ChatGPT Salute, lo spazio sicuro per dati sanitari e referti digitali. Ecco come funziona. Come chiedere consiglio all'IA senza compromettere i propri dati personali e sanitari.  - zn
3. Sanità digitale: ChatGPT for Clinicians, per medici e farmacisti

   Il modello offre strumenti avanzati e innovativi, senza costi.  - zn

1. yt Claude Mythos Clone opensource, innovativo, fatto da ragazzo 22 anni leggendo letteratura
2. 2026-04-16 yt Gemma 4: DeepMind’s New AI: A Gift To Humanity

   2026-04-18 yt Gemma 4 The real reason Google gave away

3. 2025-09-05 Apertus: la IA svizzera open source che punta su trasparenza e privacy per il bene pubblico - zn
4. 2025-07-27 Proton Lumo, l'alternativa open source a ChatGPT. Privacy al primo posto - zn

   https://lumo.proton.me

   rob: e' la prima Ai provata, per una ricerca bibliografica sui tensori, e ho fatto quache domanda. 

Hugging Face

libreria open source per aggiungere la IA ai siti web,  si scarica da GitHub e consente di generare immagini, tradurre testi, individuare oggetti nelle foto e altro ancora - zn

Links

Proof assistant. LEAN. La matematica in linguaggio informatico.

AI sperimentali

yt Self-Reflective Geometric Engine:Lossless Continual Learning via Geometric and Orthogonal Injection

the system remembers by surprise, but learns by validation