Víctor Tirreau

Founder @ | Interested in startups, NLP, AI, RecSys & Computer Science

[Víctor Tirreau @viantirreau]

Valorización de 3 veces el PIB de Chile, piola

[Anthropic @AnthropicAI]

We've raised $65 billion in Series H funding at a $965 billion post-money valuation, led by @AltimeterCap, Dragoneer, @Greenoaks, and @sequoia.

This investment will help us advance our research and expand our capacity to meet growing demand for Claude.

[Víctor Tirreau @viantirreau]

@quien_esel_jefe @El_Ciudadano @CNNChile @TVN @chvnoticias @tv_monica Mira @jjlyonn!!

[Fundación América Transparente @a_transparente]

🚨 ¿Cómo saltarse la Ley de Compras Públicas usando una doble identidad comercial? Luego de revisar el CIC de Contraloría sobre irregularidades en eventos municipales, pusimos la lupa en la Municipalidad de Río Verde. Lo que encontramos es de manual. Abrimos hilo 🧵👇

[Víctor Tirreau @viantirreau]

@jjlyonn @austinc3301 Espectacular hallazgo JJ!! Estaría cool aprovechar modelos locales para automatizar estos filtros con un costo razonable. No sé si están a la altura de la tarea aún o termina siendo más rápido conseguir grants de créditos para LLMs online

[Juan Carlos Muñoz Abogabir @JuanCaMunozA]

Les comparto mi carta al Director publicada hoy en La Tercera.

[Víctor Tirreau @viantirreau]

@Fcobalml En todo caso!!

[Víctor Tirreau @viantirreau]

Buena po Talca

[Matteo Iadonisi @6abcmatteo]

This is the official ‘Longest Line of Cheesesteaks.'

The new world record was accomplished today at the Philadelphia International Airport.

[Víctor Tirreau @viantirreau]

the next gen Sony's name will also activate Windows 11

[۟ @4NGELWING]

no bro don’t buy airpods. you need the Sony WH-1000XM5-WHCH720N-WF1000XM5-CH520

[Víctor Tirreau @viantirreau]

@patriciosainzr Grande Pato!! Felicitaciones!! 🎉

[The Hacker's Choice (@[email protected]) @hackerschoice]

THC Release 💥: The world’s largest IP<>Domain database: https://t.co/o4F8M1Pqi1

All forward and reverse IPs, all CNAMES and all subdomains of every domain. For free.

Updated monthly.

Try: curl https://t.co/5V2xLadmx5

Raw data (187GB): https://t.co/cBZOSAE89K

(The fine work of messede 👌)

[Víctor Tirreau @viantirreau]

Alto shitpost académico

[anshuman @athleticKoder]

She dumped me last night.

Not because I don't listen.
Not because I'm always on my phone.
Not even because I forgot our anniversary (twice).

But because,

in her exact words:

"You only pay attention to the parts of what I say that you think are important."

I stared at her for a moment and realized...

She just perfectly described the attention mechanism in transformers.

Turns out I wasn't being a bad boyfriend. I was being mathematically optimal.

See, in conversations (and transformers), you don't give equal weight to every word. Some words matter more for understanding context. Attention figures out exactly HOW important each word should be.

Here's the beautiful math:

Attention(Q, K, V) = softmax(QK^T / √d_k)V

Breaking it down:

Q (Query): "What am I looking for?"
K (Key): "What info is available?"
V (Value): "What is that info?"
d_k: Key dimension (for scaling)

Think library analogy:

You have a question (Query). Books have titles (Keys) and content (Values). Attention finds which books are most relevant.

Step-by-step with "The cat sat on the mat":

Step 1: Create Q, K, VEach word → three vectors via learned matrices W_Q, W_K, W_V
For "cat":

Query: "What should I attend to when processing 'cat'?"
Key: "I am 'cat'"
Value: "Here's cat info"

Step 2: Calculate scoresQK^T = how much each word should attend to others

Processing "sat"? High similarity with "cat" (cats sit) and "mat" (where sitting happens).

Step 3: Scale by √d_kPrevents dot products from getting too large, keeps softmax balanced.

Step 4: SoftmaxConverts scores to probabilities:

"cat": 0.4 (subject)
"sat": 0.3 (action)
"mat": 0.2 (location)
"on": 0.1 (preposition)
"the": 0.1 (article)

Step 5: Weight valuesMultiply each word's value by attention weight, sum up. Now "sat" knows it's most related to "cat" and "mat".

Multi-Head Magic:Transformers do this multiple times in parallel:

Head 1: Subject-verb relationships
Head 2: Spatial ("on", "in", "under")
Head 3: Temporal ("before", "after")
Head 4: Semantic similarity

Each head learns different relationship types.

Why This Changed Everything:

Before: RNNs = reading with flashlight (one word at a time, forget the beginning)

After: Attention = floodlights on entire sentence with dimmer switches
This is why ChatGPT can:

Remember 50 messages ago

Know "it" refers to something specific
Understand "bank" = money vs river based on context
The Kicker:Models learn these patterns from data alone. Nobody programmed grammar rules. It figured out language structure just by predicting next words.
Attention is how AI learned to read between the lines.

Just like my therapist helped me understand my focus patterns, maybe understanding transformers helps us see how we decide what matters.

Now if only I could implement multi-head attention in dating...

Still waiting for "scaled dot-product listening" to be invented.

[Andrew McCarthy @AJamesMcCarthy]

Immense planning and technical precision was required for this absolutely preposterous (but real) view: I captured my friend @BlackGryph0n transiting the sun during a skydive.

This might be the first photo of it's kind in existence. See a video of this moment in the reply 👇

[Jay Cummings @LongFormMath]

I might have a new favorite proof that the harmonic series diverges. It's just so beautiful.

[Víctor Tirreau @viantirreau]

Tremendo tutorial para fiscalización ciudadana!

[Juan José Lyon @jjlyonn]

Me acabo de demorar exactos 2 minutos en encontrar un trato directro irregular por 20 millones en la Corp de Desarrollo de @MuniLaReina . De hecho, me demoro más en escribir este hilo. Pero léanme para que aprendan ustedes también y seamos más fiscalizando. Voy:

[Víctor Tirreau @viantirreau]

@IngenieriaUC @CIGIDEN

[Víctor Tirreau @viantirreau]

Me sonaba conocida la universidad!!

[Víctor Tirreau @viantirreau]

Hoy mi hermana tuvo una fonda en su práctica de psiquiatría

Almorzaron empanada de locos

[Víctor Tirreau @viantirreau]

@womchile https://t.co/PJLhqezjsJ &lt;- acá

[Víctor Tirreau @viantirreau]

Paso el dato de que en WOM se pueden bloquear los prefijos 809 para llamadas comerciales no solicitadas:
Grande @womchile

[Víctor Tirreau @viantirreau]

@VraserX @3blue1brown in your last guest video you were looking for examples like this one :)