Blog

Category: HAL149

3 Applications of Generative AI for Businesses

Generative AI applications offer significant potential for businesses, but it’s important to manage expectations and tailor AI solutions to specific needs and contexts. At the moment, the main value-added solutions are

Chatbots: With the advances in GPT-3.5 and GPT-4, chatbots have become more powerful and cost-effective for businesses. They can provide customer support, answer FAQs and handle various enquiries, providing valuable assistance and saving time and resources.…

Deepfakes are good

What is the problem with deepfakes when most of the mainstream media has been systematically lying for years?.

Deepfakes pose an inherent danger to people who base their lives and beliefs on what they see in the media and online, rather than reflecting and trusting their own intuition and inner voice.

Deepfakes offer a new opportunity to stop taking what we see on screens and in the media seriously, because it has always been and continues to be deceptive.…

Is intelligence the result of connecting things?

It may one day make more sense to start with a generic neural network and train it for every possible case: text, images, etc., but for now neural networks are specialised.

In convolutional networks, we build a topology that makes the neurons connect in a similar way to the image.

In other machine learning problems, we build a neural network that resembles a system of nonlinear equations to model the problem.…

The End Of The Internet As We Know It?

The recent meltdown on Reddit marks the beginning of a concerning trend that could change the open and free internet as we know it. This shift is connected to the rise of artificial intelligence and its reliance on vast amounts of high-quality data, making unique data sets incredibly valuable.

Companies like Reddit and Twitter have started charging for their APIs, making it difficult for third-party developers to build innovative applications on top of their data.…

No intelligent LLMs

The current generative AI trend may seem impressive, but it is primarily based on a simplistic statistical approach to finding patterns in training data.

Causality and theories have become less relevant in this approach, but data scientists have been using this method for over two decades, relying on large amounts of data to uncover patterns.

This applies to all current large language models (LLMs) that lack a foundation in linguistics, semantics, or conceptual understanding.…

Talking to machines will make us much more productive

We are moving towards talking to machines as if you were talking to a real person. That means one conversation at a time, not  cluttered screens filled with links, data, images and ads.

Many people are completely addicted to complexity and constant external stimulation. But imagine being able to work just by thinking and talking with people and AI. Being able to concentrate and think clearly by doing one thing at a time.…

From Wikipedia to Embeddings

The code demonstrates how to use the mwclient library to connect to Wikipedia, retrieve the content of a specific page, and print it. This provides a starting point for further exploration and utilization of the mwclient library for retrieving information from Wikipedia to create embeddings and fine-tune chatgpt.

 

Machine learning is about modelling reality

The basic idea of machine learning is to build models that capture the essence of reality.

We do this by reducing the entropy of a set of observations by building a model: a mathematical structure and associated parameters that capture certain aspects of the inner structure of the data.

We build models to understand data (statistics), to predict data (regression), to classify data (clustering, etc.),…

GPT Customization in 3 Layers

The GPT-3 model underwent pre-training using a vast corpus of text sourced from the open internet. With only a small number of examples as input, it can often deduce the task you intend to accomplish and generate a plausible output. This ability is commonly referred to as “few-shot learning.”

By employing fine-tuning, the model’s performance in few-shot learning is further enhanced through training on a substantially larger set of examples beyond what can be included in the initial prompt.…

Jobs Based on Knowledge are to Go. A key reason AI will disrupt white collar work is that many of these jobs require knowledge but not intelligence.

Jobs like HR, law, marketing, software development, etc don’t need original thinking like 80% of the time. They just need you to know the rules of the system.

Secret Messages Can Hide in AI-Generated Media

According to a recent article in Quanta Magazine, it is possible for secret messages to be hidden in AI-generated media. The article discusses how researchers have found ways to embed messages in images and audio that are undetectable to humans but can be decoded by machines. This has implications for security and privacy, as well as for the potential misuse of AI-generated media.…

ChatGPT Vulnerable to Prompt Injection via YouTube Transcripts

AI Security Researcher Johann Rehberger has discovered a security vulnerability in ChatGPT plugins that allows bad actors to pass instructions to the bot during chat sessions. The exploit involves feeding new prompts to ChatGPT from the text of YouTube transcripts.

Rehberger edited the transcript for one of his videos to add new instructions and a prompt, and then asked ChatGPT to summarize the video.…

IA Generativa y Empresas: Dónde y Cómo Aplicarla

Hoy en día ya es imposible ignorar los avances de la inteligencia artificial (IA). Desde la nueva generación de chatbots generativos hasta los modelos que pueden generar (casi) cualquier imagen (o muy pronto, vídeo), el ritmo de desarrollo en el campo de la IA ha sido poco menos que fenomenal.

Especialmente interesante es el campo de la IA generativa, donde vemos un número creciente de modelos que pueden crear imágenes, texto, vídeo y música.…

AI will make the Web Invisible

The Internet will become invisible to people in the next few years, buried under a new layer of intelligent content. And with it, digital marketing as we know it: content marketing, web design, email marketing, etc.

Why? Because your AI personal assistant will know your preferences better than you do. And it will go out of its way to review, summarize, and present information the way you want it, not the way Google or some designer wants it.…

The AI Wars: Elon Musk attacks Microsoft over ChatGPT tech access

Elon Musk has criticized Microsoft for gaining exclusive access to the entire OpenAI codebase, including the ChatGPT technology, which is housed within Microsoft Azure.

Musk argues that OpenAI should continue to be a non-profit and not a source of income for a single firm. Microsoft invested more than $10 billion in OpenAI and is now its largest shareholder.

“Aren’t you an OpenAI investor?”…

Is Skynet taking over everything?

I’m old enough to remember when the Internet started, and it took several years for it to really go mainstream.

The same will happen with AI: it will take longer than many expect. Society has a lot of inertia, and the latest developments are always followed by a small group of early adopters.

Ignore the noise and the MSM hype. There’s time to prepare; relax and enjoy the ride.

Labor Market Impact Potential of AI

Until 200 years ago, most people worked in agriculture, productivity was terrible, and famine was common.

Then the machines came along and other problems arose, but food shortages and malnutrition were a thing of the past.

AI will automate all jobs that can be automated.

Anything that can be automated should be automated.

Will society look forward and adapt, to innovate and improve people’s lives?.…

The rise of the one-person unicorn

The startup landscape is changing, and a new type of business model is emerging: the bootstrapped, cash-generative, and rapidly-growing one-person unicorn. These hyper-specific, profitable companies are solving single, specific problems, such as Headlime, ConvertKit, Fathom, and Transistor. They are built by software company founders, often serial entrepreneurs, who can fund their companies themselves and run them with small teams of one or two people.…

Beyond Language: How GPT-4 Can Reason

GPT-4 is a powerful language model that can generate content based on previous inputs using machine learning algorithms. However, it has the potential to do much more than just language processing.

The model is built on a transformer architecture that allows it to perform symbolic reasoning and incorporate external knowledge sources, such as large databases of information like Wikipedia.…