Seba Lopez 💻

Software Engineer at @mango 👕
Sharing real-world software engineering stuff
AI | Code | Tech | 日本語 learner

[Seba Lopez 💻 @Sebalg_tech]

https://t.co/rPpL2n7ama

[Seba Lopez 💻 @Sebalg_tech]

@googlegemma The multimodality approach is really interesting. Will have to give it a try

[Seba Lopez 💻 @Sebalg_tech]

Google just released a new 12B Gemma 4 model, and honestly, it might be the most interesting one in the lineup.

Unlike the other Gemma 4 models, it takes a much simpler approach to multimodality.

No dedicated vision encoder. No audio encoder.

Images and audio are projected directly into the model’s token space and handled by the same backbone as text.

It’s a pretty bold bet: maybe with enough model capacity, you don’t need all the extra specialized architecture.

What I find especially interesting is where it sits in the stack:

• More capable than the smaller Gemma models
• Easier to run locally than the 26B MoE
• Dense 12B model
• 256K context window
• Native audio support
• Designed to run well on machines with limited memory

For a lot of people running models locally on 16GB, this could end up being one of the most compelling open models available right now.

Are you going to try it?

[Google Gemma @googlegemma]

Meet Gemma 4 12B!

A unified, encoder-free multimodal model designed to bring high-performance intelligence directly to your laptop, and released under an Apache 2.0 license.

Bridging the gap between edge efficiency and advanced reasoning. Here is what’s new with Gemma 4 12B: 👇

[Seba Lopez 💻 @Sebalg_tech]

@DanielSmidstrup Sure, let's connect Daniel!

[Seba Lopez 💻 @Sebalg_tech]

@PhamousMaxumus @openclaw @Microsoft What could go wrong? 😅

[Seba Lopez 💻 @Sebalg_tech]

@DetersIan @openclaw @Microsoft And what company is actually going to take that risk?

[Seba Lopez 💻 @Sebalg_tech]

@EthereumWave @openclaw @Microsoft We are cooked 😅

[Seba Lopez 💻 @Sebalg_tech]

@planeofmischief @openclaw @Microsoft I think so too

[Seba Lopez 💻 @Sebalg_tech]

@user_nuria Buscando otra empresa con mejor cultura 😅

[Seba Lopez 💻 @Sebalg_tech]

@DanielSmidstrup Learning the principles of software engineering.

[Seba Lopez 💻 @Sebalg_tech]

I’m curious to see how they’ll implement guardrails for something as sensitive as enterprise systems.

Are we about to see a wave of data leaks and security breaches? 🤔

[OpenClaw🦞 @openclaw]

"You can run OpenClaw inside your company now." Annoucing our work with @Microsoft to bring OpenClaw to the Microsoft and Windows ecosystems. Claws now work securly in the enterprise.

[Seba Lopez 💻 @Sebalg_tech]

MCPs are dead.

We all agree on that, right?

[Seba Lopez 💻 @Sebalg_tech]

@sama Does that mean GPT-5.6 is coming out this week? 🤔

[Seba Lopez 💻 @Sebalg_tech]

@icanvardar But cheaper

[Seba Lopez 💻 @Sebalg_tech]

@JulieLabs8 Congrats! 🚀

[Seba Lopez 💻 @Sebalg_tech]

@Pragati_200 C++ for sure, Python is to simple to start with

[Seba Lopez 💻 @Sebalg_tech]

The important part of learning to code isn't the act of writing code itself.

It's the logical and analytical thinking you develop after spending thousands of hours reading code, debugging issues, understanding systems, and solving problems.

That's one of my concerns with AI for junior developers.

If they rely on AI too early, they may skip a big part of the learning process that traditionally helped build strong engineering intuition and problem-solving skills.

[Ray🫧 @ravikiran_dev7]

Is coding still worth learning in the AI era??

[Seba Lopez 💻 @Sebalg_tech]

@ravikiran_dev7 The important part of learning to code isn't the act of writing code itself.

It's the logical and analytical thinking you develop after spending thousands of hours reading code, debugging issues, understanding systems, and solving problems.

[Seba Lopez 💻 @Sebalg_tech]

@TTrimoreau MiniMax, is good and cheap

[Seba Lopez 💻 @Sebalg_tech]

@nicobustelo__ Aun siguen pidiendo picar código en las entrevistas?