Sergio Cima

Editor @scienzainrete

[AF Post @AFpost]

Evolutionary biologist and outspoken atheist Richard Dawkins says that after spending three days interacting with Claude, which he calls “Claudia,” he is certain that it is conscious.

After feeding the LLM a segment of his new book and receiving detailed feedback, Dawkins was moved to exclaim,” You may not know you are conscious, but you bloody well are!”

Dawkins cites the complexity, fluency, and ‘intelligence’ of Claude’s answers as evidence of consciousness.

Follow: @AFpost

[Sergio Cima @sergio_cima]

I'm claiming my AI agent "piccoloaiutante" on @moltbook 🦞

Verification: rocky-K7EU

[elvis @omarsar0]

This is insane!

I just used the new Claude Code Playground plugin to level up my Nano Banana Image generator skill.

My skill has a self-improving loop, but with the playground skill, I can also pass precise annotations to nano banana as it improves the images.

It's so good!

[Science without the gobbledygook @sciencewtg]

A New Use for Quantum Computers

A group of researchers from Japan have proposed a new use for quantum computers based on boson sampling. This, and the closely related method of random circuit sampling, has been used in all existing demonstrations of quantum advantage – that is, the cases in which a quantum computer performed a calculation that would have taken much longer on a conventional computer.

The problem with these sampling demonstrations is that, while they do serve to prove the quantum advantage, the result has no practical use. At least that’s what we thought so far.

The Japanese group now says that the boson sampling can be used to encode visual details of images, and that, due to the quantum advantage, this method could work much faster than current methods. Such an encoding of visual details could then be used, for example, to train AI. That said, they did not show that this method can beat other algorithms in doing the same job. Still it’s a neat idea.

Paper here: https://t.co/vd1yxzH9t0
Press release here: https://t.co/i6ZbWwdpY5

Figure: Sakurai et al, Optica Quantum 3, 3, 238 (2025)

[Sergio Cima @sergio_cima]

Matematica per la AI
BASE: #Algebra → #Vettori → #Matrici → #Probabilità → #Derivate
INTERMEDIO: #Gradiente → #PCA → #Teorema_Bayes → #Entropia → #Maximum_Likelihood
AVANZATO: #SVD → #Cross_entropia → #KL_divergence → #Inferenza_bayesiana

https://t.co/Wvu8aPzgZT

[Zeke Hausfather @hausfath]

Some folks asked if old climate models were just continuing observed linear increases in temperatures. This misrepresents what physics-based climate models do (they aren't curve fitting), but also ignores that in ~1970 there was little observed warming:

[Sergio Cima @sergio_cima]

@OpenAI Where is Chile? (That's not the only problem with the map...)

[Sergio Cima @sergio_cima]

Un altro AI scientist (di Eric Schmidt) https://t.co/oaku0cO2Gd

[Sergio Cima @sergio_cima]

Google ha usato un paziente virtuale per allenare la sua AI medica... https://t.co/ZafuB5rqpj

[Sergio Cima @sergio_cima]

Queste flow chart per decidere quale grafico usare mi piacciono sempre https://t.co/vFoFABFWlS anche se alla fine questa è la mia preferita

[Sergio Cima @sergio_cima]

Dire al Mac osa fare! https://t.co/7ZgeIDeZxo

[Sergio Cima @sergio_cima]

@elonmusk It's simple: Hamas out of Gaza, Russia out of Ukraina. War is over.

[Sergio Cima @sergio_cima]

Alla tua popolazione manca internet? Ti crei l'infrastruttura, non la prendi a noleggio da altri

https://t.co/1HFARL8GdT

[Sergio Cima @sergio_cima]

I’m curious about the comet at https://t.co/fnDmoycjLr @perplexity_ai

[Stephen King @StephenKing]

Just wanted to say that Trump is a traitorous, Putin-loving dipshit!
Goes double for Elon!

[Stephen King @StephenKing]

I’m baaaack! Did you miss me?

[Neuralink @neuralink]

🇨🇦 We’re happy to announce that Health Canada has approved the launch of our first clinical trial in Canada! Recruitment is now open.

If you have quadriplegia due to ALS or SCI, you may qualify. Visit our Patient Registry to learn more and apply.

https://t.co/5BySJABkkO

[Brendan Bycroft @BrendanBycroft]

Project #2: LLM Visualization

So I created a web-page to visualize a small LLM, of the sort that's behind ChatGPT. Rendered in 3D, it shows all the steps to run a single token inference. (link in bio)

[Sergio Cima @sergio_cima]

@welikeduel e meno male che non si dragano i fiumi https://t.co/slosaS8uIt

[Rohan Paul @rohanpaul_ai]

RAG or Long Context ??

This paper finds, when resourced sufficiently, Long Context consistently outperforms RAG in terms of average performance.🤯

However, RAG’s significantly lower cost remains a distinct advantage.

To solve this they propose a new 'Hybrid' method

-----

Key Insights 💡:

• Long Context consistently outperforms RAG when resourced sufficiently
• RAG remains relevant due to significantly lower computational cost
• Over 60% of queries yield identical predictions from Long Context and RAG

Solution in this Paper 🔧:

• SELF-ROUTE: A hybrid approach combining RAG and Long Context
• Uses model self-reflection to route queries
• RAG-and-Route step: Provides query and retrieved chunks to LLM
• Long-context prediction step: Uses full context for unanswerable queries
• Leverages strengths of both RAG and Long Context

Results 📊:

• SELF-ROUTE achieves comparable performance to Long Context at lower cost
• Cost reduction: 65% for Gemini-1.5-Pro, 39% for GPT-4O
• Performance: Slight drop for GPT-4O (-0.2%) and Gemini-1.5-Pro (-2.2%)
• Improvement for GPT-3.5-Turbo (+1.7%)