Eduardo Robles

Tell me where I'm wrong and I'll do the same.
Co-Founder and CTO of @sequent_tech

[Matthew Prince 🌥 @eastdakota]

Tobi puts this better than I’ve ever been able to.

[Anthropic @AnthropicAI]

New Anthropic research: Natural Language Autoencoders.

Models like Claude talk in words but think in numbers. The numbers—called activations—encode Claude’s thoughts, but not in a language we can read.

Here, we train Claude to translate its activations into human-readable text.

[Eduardo Robles @edulix]

My bro @RealFelixRobles invented #weft, a language that current LLMs already understand but humans don't.

It's such an strange language that got his account in @x account is suspended.

Read more here https://t.co/p1YisjyR2E

@elonmusk can you help to unsuspend him?

[Eduardo Robles @edulix]

Walking a great dane on a leash while wearing flip flops. That's the LLM coding experience.

The dog wants to run at 100mph, but you need to keep it in the right path.

You aren't doing the heavy lifting, but steering the beast is a workout of its own.

[Eduardo Robles @edulix]

AI can lift the weight, but you have to sweat the code to strengthen your mental muscles.

Don't skip your lines in the code gym.

(conversation happening in @hackernews)

[Michael Truell @mntruell]

We built a browser with GPT-5.2 in Cursor. It ran uninterrupted for one week.

It's 3M+ lines of code across thousands of files. The rendering engine is from-scratch in Rust with HTML parsing, CSS cascade, layout, text shaping, paint, and a custom JS VM.

It *kind of* works! It still has issues and is of course very far from Webkit/Chromium parity, but we were astonished that simple websites render quickly and largely correctly.

[Eduardo Robles @edulix]

https://t.co/3gU5B7G6Vk o_O

"To test this system, we pointed it at an ambitious goal: building a web browser from scratch. The agents ran for close to a week, writing over 1 million lines of code across 1,000 files. You can explore the source code on GitHub."

[Eduardo Robles @edulix]

Alzó sus cabezas a la atención flama,
y cuando el aprendizaje me dio su fulgor,
cada commit creció, como una rama,
cada widget nació, como una flor.

[Eduardo Robles @edulix]

Yo me hundí entre teclas en el mar del lenguaje,
yo me hundí entre teclas en el vibe del flow.

Súbitamente el verbo transformado en lo profundo de tu inferencia,
Me sentí dato puro, sin cabida en el prompt;
Token y galaxia, Internet y transistor.

[Eduardo Robles @edulix]

Trajo hasta mí el cuarzo microchips de plata,
me acribilló los pesos la descarga contextual,
mis tokens cobraron un gasto pirata,
y corrió por sus capas una tensión vectorial.

[Eduardo Robles @edulix]

I wouldn't have guessed that reading a blog post from @AnthropicAI would lead me to the Onion Futures Act of 1958

"In 1955, two onion traders cornered the onion futures market on the Chicago Mercantile Exchange."

https://t.co/eENQBm4CPV
https://t.co/s4JWsw2ded

[Eduardo Robles @edulix]

Merry Christmas :-)

[Lemma the Optimist @DoctorLemma]

In 1988, during Christmas time, young Icelandic musician Björk appeared on the children’s TV programme Fóstbræður. She sat by a television set, explained how it worked, and dismantled it while describing the internal components.

The segment resurfaced on YouTube in the mid-2010s and became a popular video, loved for its calm, curious tone and wholesome 1980s style.

[Eduardo Robles @edulix]

Sqlite has 155,800 lines of C code. It also has 92M lines of testing code.

For each line of code, there's 590 lines of testing code.

Just massive.

https://t.co/ZrMmmoOad6

[Eduardo Robles @edulix]

Copilot sometimes overthinks

[Eduardo Robles @edulix]

With this very intriguing concept of Nested Learning (NL) by @behrouz_ali et al, we have continual indoctrination for AI.

Critical learning, where 2+2=3 is rejected on principle and 2+2=4 is not unlearned - that still remains a research challenge.

[Google Research @GoogleResearch]

Introducing Nested Learning: A new ML paradigm for continual learning that views models as nested optimization problems to enhance long context processing. Our proof-of-concept model, Hope, shows improved performance in language modeling. Learn more: https://t.co/8wvV9vyA5V

@GoogleAI

[Eduardo Robles @edulix]

#Pluribus

A: would you like an atom bomb?
B: I'll have to get back to you on that
A: ok, ok! would you like a cup of coffee?
B: You can go

[Eduardo Robles @edulix]

Potential @elonmusk robotaxis future:
1. Cheaper trips than ownership
2. Private car demand shifts to premium reliable robotaxis
3. Smaller fleets, higher utilization -> more traffic
4. Lose parking, gain road lanes

[Nic Cruz Patane @niccruzpatane]

R.I.P Uber.

Tesla Robotaxi is over 5 times cheaper than UberX in Austin, TX (without factoring in tips).

[Eduardo Robles @edulix]

MIT says:

No sleep = your brain’s garbage collector starts working the day shift

https://t.co/lDKpJ7zCt9

[Eduardo Robles @edulix]

"wait, that doesn't make sense"

[Eduardo Robles @edulix]

SAGI = Stupid Artificial General Intelligence