Luis García de la Fuente

Blog

AI training: the problem and the opportunity are in the data

The problem with AI today is not that we don’t know how to implement it. The problem is that, for the most part, we don’t have the domain-specific data to train it on (legally). This is where companies like Google and Facebook have an advantage. At least for consumer data.

The technology is not the advantage. The data is. The tech is available to anyone who knows high school level Python and can rent cloud servers, but only the big guys have the data.…

Language is primarily a tool for communication rather than thought

One of the most interesting things about GenAI is that it tells us that language is not what we think it is, and that it does not have to be learnt using traditional grammar-based rules.

The fact that a machine can become fluent in any language simply by reading or watching a large enough amount of properly tagged content is entirely consistent with the way humans learn their native language.…

Does the ability to speak imply intelligence?

Recent advancements in Large Language Models (LLMs) have revealed that complex linguistic capabilities can emerge from machine learning processes without explicit programming of grammatical rules.

The fact that certain processes such as language can be learned as an emergent property, rather than through grammar and syntax, is one of the most interesting findings of the whole GPT thing.…

Latent Reality

The world is your particular point of view of a multidimensional totality that contains all possible past, present and future scenarios. There is no past or future time: there is an eternal now that contains everything and in which everything is possible.

El mundo es tu punto de vista de una totalidad multidimensional que contiene todos los escenarios posibles pasados, presentes y futuros.…

Semantic Search and Embeddings in AI Applications

Large Language Models (LLMs) and fine-tuning techniques are often misunderstood in their application to specific question-answering tasks. While fine-tuning can improve model performance, it does not inherently enable a model to provide predetermined answers to specific questions. Instead, a more effective approach for retrieving precise information involves semantic search using embeddings technique.…

This house has a bold, futuristic design that contrasts sharply with its natural surroundings

The house has a bold, futuristic design that contrasts sharply with its natural surroundings. It consists of two main sections: a lower level and an upper level that extends outward dramatically. The upper level is a rectangular module made of grey concrete with rounded edges that juts out horizontally from the main structure, seemingly defying gravity. This cantilevered section has large floor-to-ceiling windows on its front face, offering panoramic views of the forest.…

The house has a bold, futuristic design that contrasts sharply with its natural surroundings.

The house has a bold, futuristic design that contrasts sharply with its natural surroundings. The house features clean lines and smooth surfaces throughout. The surrounding environment is a dense pine forest. The house is built on uneven, rocky terrain with large boulders. Its almost night, weather is very foggy and there´s a thick layer of snow. The blue lights of the bunker and the yellow lights of the house through the fog give the scenery a dystopian look.…

The building has a bold, futuristic design that contrasts sharply with its natural surroundings

The building has a bold, futuristic design that contrasts sharply with its natural surroundings. It features clean lines and smooth surfaces throughout. The surrounding environment is a dense pine forest. The building is built on uneven, rocky terrain with large boulders. It consists of two main sections: a lower level and an upper level. The upper level is a house made with a rectangular module made of grey concrete with rounded edges that juts out horizontally from the main structure.…

Generative AI: Beyond the Interface

Generative Artificial Intelligence (AI) systems consist of multiple layers extending beyond the visible output. While consumers interact with text, image, video, and audio outputs, these applications are supported by a complex technical stack. This infrastructure includes predictive models, data platforms, and computational resources that enable specialized generative outputs for various tasks and industries.…

Analyzing the Capabilities and Limitations of Large Language Models

The development of Large Language Models (LLMs) has led to significant advancements in machine-generated text, creating challenges in distinguishing between AI-produced and human-authored content. This technological progress has implications across multiple sectors, including education, literature, and various professional fields. The increasing sophistication of LLMs has diminished the effectiveness of traditional statistical methods for detecting machine-generated text.…

Striking juxtaposition of futuristic architecture and primeval wilderness

Image of a striking juxtaposition of futuristic architecture and primeval wilderness. A sleek, angular structure stands prominently in the foreground, its dark exterior contrasting sharply with the lush green forest surrounding it. The building’s design is distinctly modern, featuring a geometric shape that resembles a hexagon when viewed from the front.…

Limitations and Advancements in Language Model Technology

Large Language Models (LLMs) exhibit significant weaknesses in information reliability. Absent meticulous human-designed scripts and rigorous quality assurance processes, these systems frequently produce erroneous outputs (aka “hallucinations”).

The challenges extend beyond mere factual inaccuracies. LLMs face substantial hurdles in maintaining current information.…

AI: Playing with entropy to recreate reality

Neural networks (and LLMs are no exception) are all about deconstructing information, removing noise from signals, encoding signals and then deliberately adding noise (entropy) to create signals.

These processes of constructing and deconstructing information are at the heart of the emerging field of generative AI, which aims to create systems that can generate new and original content.…

The real Matrix

La realidad aparente de cosas sólidas en el espacio-tiempo es un fractal creando ilusión de solidez. Conciencia mezclándose con un espacio de probabilidades, que adopta los valores y formas que ésta imagina. En realidad no hay nada sólido: todo está fundamentalmente vacío. El vacío es lo que en realidad da forma aparente a todo. En 3D causas y efecto son una red totalmente impenetrable, una ilusión fractal.…

Humans are not computable processes

Imagine being someone who thinks humans are like premium computers. Living in a short-sightedness equivalent to a Paleolithic mind. Thinking that the mind can only regurgitate information and datasets that absorb our brains, when humans have literally invented all the abstractions, art, language, infinity and God.

The belief that humans are merely lines of code, held by many including those in the field of technology, is a troubling indication of the current state of our world.

La Singularidad y la guerra contra las personas

En “Singularity is near” Kurweil sostiene que hacia 2046 la tecnología permitirá codificar las mentes (como si fueran programas de ordenador), y cargarlas encriptadas en ‘la nube’ para formar una conciencia única e inmortal, que controlará todo el poder de computación del planeta.  Sostiene que los baby boomer son la primera generación que se hará inmortal y que se podrá hablar con los muertos como quien habla con un software de AI.…

The Return of the East India Companies

But the risk of corporations more actively supporting multiple sides in conflict zones and carving up their own territories and spheres of influence is a concerning prospect, akin to the Dutch East India Company which governed its own territories through military force and trade monopolies.

While there are still waning expectations that multinational corporations pick clearer sides in interstate conflicts, there appears to be little stopping them from fueling and prolonging intrastate conflicts featuring non-state actors, as long as it serves their financial interests.…

Opportunity of the Second Layer Businesses in AI Startups

The emergence of large language models (LLMs) and artificial intelligence has undoubtedly transformed the tech landscape, but the true potential for new businesses lies in what we might call the “second layer” of AI implementation. While many startups are merely creating wrappers around existing LLM capabilities, the real opportunity exists in developing specialized integrations and applications that address specific industry needs and business challenges.…

Researchers leverage shadows to model 3D scenes, including objects blocked from view

This technique could lead to safer autonomous vehicles, more efficient AR/VR headsets, or faster warehouse robots.

Imagine driving through a tunnel in an autonomous vehicle, but unbeknownst to you, a crash has stopped traffic up ahead. Normally, you’d need to rely on the car in front of you to know you should start braking. But what if your vehicle could see around the car ahead and apply the brakes even sooner?…

Evolution or Will of Power?

Darwin's theories are not the cause of anything, they are the apparent effect. The dinosaurs didn't start flying because some cosmic casino was betting on trillions of mutations to see what would b...

Las alucinaciones son una feature, no un bug de los LLM

Las alucinaciones son una feature, no un bug. Este concepto se puede usar para crear nuevos negocios y productos. Si sólo tienes unas ideas vagas de lo que quieres hacer la IA te puede ayudar a cristalizarlas en propuestas de valor concretas.

https://hal149.com/en/artificial-intelligence-to-generate-business-ideas/

GenAI y la Creatividad

Could hallucinations prove to be a source of creativity and therefore innovative solutions for verifiable problems?…

A blind LLM can create and refine images?

Researchers at MIT CSAIL found that large language models (LLMs) trained only on text data have an impressive understanding of visual concepts. By prompting LLMs to generate code for rendering images, the researchers collected a dataset of simple digital illustrations. Remarkably, the LLMs could iteratively improve these illustrations when prompted, demonstrating their robust visual knowledge gained from textual descriptions.…