My name is Roger Oriol, I am a Software Architect based in Barcelona, Spain. I am a MSc graduate in Big Data Management, Technologies and Analytics. This blog will be the vehicle to divulgate and discuss topics on web development, data architecture, software architecture and much more.
Previous parts of Build a Basic AI Agent From Scratch : Basic Agent Tools Long Task Planning Human in the Loop & Security Security II You can find and clone this code in this blog series' Github repo . In the previous part we started closing the gaps left open by human-in-the-loop: a Docker sandbox to contain runaway commands, prompt-injection defenses so the model stops trusting tool output as…
Last week, HuggingFace suffered an AI-assisted cyberattack. The experience for HuggingFace trying to defend themselves from the attack is pretty surprising and damning for all people who think cybersecurity-capable models need to be restricted. From HuggingFace: the analysis requires submitting large volumes of real attack commands, exploit payloads, and C2 artifacts, and these requests were…
Previous parts of Build a Basic AI Agent From Scratch : Basic Agent Tools Long Task Planning Human in the Loop & Security You can find and clone this code in this blog series' Github repo . In the previous part we gave our agent a basic safety model: permission modes, an acceptEdits trust boundary, and an ask_question tool so the agent could stop and clarify before doing something risky. That was…
Previous parts of Build a Basic AI Agent From Scratch : Basic Agent Tools Long Task Planning You can find and clone this code in this blog series' Github repo . In the previous part of the Build A Basic AI Agent From Scratch series, we gave our agent the ability to plan and work on long tasks. We added a scratchpad, a to-do list and a system prompt that explains to the model how to break work…
Just 3 days after the release of Claude Fable 5 and Mythos 5, Anthropic was forced to suspend access to them by the US Commerce Department. According to the government, a third party reported to them a method of jailbreaking Anthropic's safeguards. The Commerce Department asked Anthropic to suspend access to Fable and Mythos of all people foreign to the United States, including all foreign people…
Anthropic has announced it's most capable model with the name Fable 5 . This model was previously hidden from the public and only made available only to a select number of companies with the name Claude Mythos Preview . The reported reason for hiding it was that it was "too powerful" to be made available to the broad public and therefore bad actors out there. Apparently, now Anthropic is…
In the previous part of the Build A Basic AI Agent From Scratch series, we added the essential tools to our agent to allow it to work autonomously for us. We gave it the ability to find files, read and write files, run bash commands and get content from the web. We got a very capable agent with just these tools. What happens when the agent runs long and complex tasks? The current agent works very…
In the previous part of the Build A Basic AI Agent From Scratch series, we built the most basic AI agent harness possible. It was just a connection to a model, a way to take user input, a store of context of the conversation and a loop that kept the agent running. Of course, this agent is not very useful. It can only interact by taking your input and answering you based on its internal knowledge.…
2026 is without a doubt the year of AI agents. Since the release of Claude Code, the power of these AI agents has become undeniable. Claude Code, Codex, OpenCode are a must for many developers nowadays. OpenClaw and Hermes are becoming many people's AI assistants. Agents are also breaking into knowledge work with tools like Cowork. If you follow me in this series of posts, we will build a basic AI…
OpenAI has finally released it's GPT-5 model, and as we were already expecting, it's a hybrid reasoning model. Now the model itself chooses how much to think about each task, and you can force the reasoning effort as well. This probably means the end of the o series of reasoning models from OpenAI, as the regular language models and the reasoning models will now be unified. Of course, the…
It's not at all straightforward to understand the variants of the GPT-5 model released today. The API docs describe four models: gpt-5, gpt-5-mini, gpt-5-nano and gpt-5-chat. However, the system card describes 6 models to replace older models, and none of the names match with the API: It can be helpful to think of the GPT-5 models as successors to previous models: Table 1: Model progressions…
Just like Sam Altman hinted at a while ago, OpenAI just released two open-weight models trying to appease the common criticism of being a company with "Open" in the name that hasn't released any open language models in a long while (since GPT-2!). The new open-weights models (not open-source like the name seems to imply) are Mixture-of-experts models with: 116.83 billion parameters with 5.13…
While reading Anthropic's great article " How we built our multi-agent research system ", I stumbled upon this quote where Anthropic researchers present the results where they found that multi-agent systems outperform single-agents for complex tasks: For example, when asked to identify all the board members of the companies in the Information Technology S&P 500, the multi-agent system found the…
An agent is something that acts in an environment; it does something. Agents include worms, dogs, thermostats, airplanes, robots, humans, companies, and countries. Artificial Intelligence: Foundations of Computational Agents, 3rd edition by David L. Poole and Alan K. Mackworth, Cambridge University Press 2023 This is the definition I personally like the best for what agents are in the context of…
A true AGI must be general across all domains. Any complete definition must at least include the ability to solve problems that originate in physical reality, e.g. repairing a car, untying a knot, preparing food, etc. In this excellent article, Benjamin Spiegel argues that our current approach to building LLMs cannot lead to an AGI. While the current next-token prediction approach is really good…
There's a new kind of coding I call "hype coding" where you fully give into the hype, and what's coming right around the corner, that you lose sight of whats' possible today. Everything is changing so fast that nobody has time to learn any tool, but we should aim to use as many as possible. Any limitation in the technology can be chalked up to a 'skill issue' or that it'll be solved in the next AI…
Together with the launch of the o3 and o4-mini reasoning models , OpenAI has released a coding assitant for the terminal: Codex. Codex is meant to be used with OpenAI models. You can use it to create new projects, make changes to existing projects or ask the model to explain code to you, all in the terminal. It can use multimodal input (e.g. screenshots). It also allows sandboxing your development…
After the unimpressive release of GPT-4.5 a month and a half ago, OpenAI is now releasing a new version - backwards. Today, they released three new models, exclusive to the API: GPT-4.1 , GPT-4.1 mini and GPT-4.1 nano . In the benchmarks, GPT-4.1 easily beats GPT-4.5 at a lower price and higher speed. For this reason, OpenAI has said they will be deprecating GPT-4.5 in 3 months time. While this is…
Just in the middle of the year of agents, Google has released two great tools for building agents: the Agent2Agent (A2A) protocol and the Agent Development Kit (ADK) . The Agent2Agent Protocol is based on JSON RPC , working both over plain HTTP and SSE . It is also built with security in mind, it implements the OpenAPI Authentication Specification . The agents published using this protocol will…
Meta should have made it clearer that “Llama-4-Maverick-03-26-Experimental” was a customized model to optimize for human preference. As a result of that we are updating our leaderboard policies to reinforce our commitment to fair, reproducible evaluations so this confusion doesn’t occur in the future. We now have acknowledgement from LMArena of what we already knew: AI labs are cheating to get…
Meta has finally released the Llama 4 family of models that Zuckerberg hyped up so much. The Llama 4 models are open-source, multi-modal, mixture-of-experts models. First impression, these models are massive. None of these models will be able to run in the average computer with a decent GPU or any single Mac Mini. This is what we have: Llama 4 Scout The small model in the family. A…
A group of Anthropic-affiliated scientists has released a paper where they study how human concepts are represented across Claude 3.5 Haiku's neurons and how these features interact to produce model outputs. This is a specially difficult task since these concepts are not contained within a single neuron. Neurons are polysemantic , meaning that they encode multiple unrelated concepts in its…
Every web developer has been there: you're starting a new project and staring at an empty file called index.html . You try to remember, which tags were meant to go in the <head> again? Which are the meta tags that are best practice and which ones are deprecated? Recently, I found myself in this exact situation. My first instinct was to copy the head section from a previous project, but as I…
The Anthropic team has discovered an interesting approach to LLM thinking capabilities. Instead of making the model think deeply before answering or taking an action, they experimented with giving the model a think tool . The think tool does nothing but register a thought in the state. However, it does allow the model to decide when it's appropriate to stop and think more carefully about the…
These benchmarks test if AI models can find the right answers to a set of questions we already know the answer to. However, real scientific breakthroughs will come not from answering known questions, but from asking challenging new questions and questioning common conceptions and previous ideas. - Thomas Wolf Interesting reflection from Thomas from HuggingFace. Current LLMs have limited potential…
On the Internet nobody knows you are a dog... or an LLM The Dead Internet Theory is a thought that has gained a lot of traction recently. I have to admit, the first time it was explained to me, I felt an eerie realization. Like I had already been experiencing it, but I hadn't paid too much attention to it. The first moment, I felt scared for the future and nostalgic for the past. But that's an old…
Le Penseur (1904) in the Musée Rodin in Paris. Picture by CrisNYCa - Wikipedia. A new kind of LLM has recently been popping out everywhere: Reasoner models. Kickstarted by OpenAI's o1 and o3, these models are a bit different from the rest. These models particularly shine when dealing with mathematical problems and coding challenges, where success depends on following precise, logical steps to…
The past year has been transformative for artificial intelligence, marked by breakthrough innovations, emerging regulations, and a shift toward practical AI tools that enhance productivity. As we look ahead to 2025, let's review the major developments of 2024 and explore what the future might hold. Part I: 2024 Year in Review OpenAI in the front OpenAI maintained its position at the forefront of…
The recent announcement that Meta's Llama 3.2 Vision models won't be available in the European Union has reignited discussions about the impact of EU regulations on AI innovation and accessibility. This development joins a growing list of AI technologies from major tech companies that are currently unavailable to EU users, including ChatGPT's Advanced Voice mode and Apple Intelligence , raising…
In previous articles, we built a basic RAG application. We also learned to introduce more advanced techniques to improve a RAG application. Today, we will explore how to tie those advanced techniques together. Those techniques might do different, sometimes opposite, things. Still, sometimes we need to use all of them, to cover all possibilities. So let's see how we can link different techniques…
In the last article, I established the basic architecture for a basic RAG app. In case you missed that, I recommend to first read that article over here. That will set the base from which we can improve our RAG system. Also in that last article, I listed some common pitfalls that RAG applications tend to fail on. We will be tackling some of them with some advanced techniques in this article. To…
The dawn of Generative AI makes possible new kinds of capabilities for the applications we build. LLMs can answer the user’s questions with an incredible skill. So, why not use them as part of our systems. If the user needs help getting around the app, we can put a chat function where the LLM will answer all the user’s questions. If our app has blog posts explaining important concepts, instead of…
One of the tasks Generative AI models are good at is summarizing texts. However, if you want to summarize a text, you need the whole text to fit in the context window, plus the command to summarize it. The following is an example prompt (you can test it yourself using the Google Colab notebook for this article ): Write a concise summary of the following text: The Turing test, originally called the…
Unlike the Transformers architecture, Mixture of Experts is not a new idea. Still, it is the latest hot topic in Large Language Model architecture. This architecture has been rumored to power OpenAI's GPT-4 (and maybe GPT3.5-turbo) and is the backbone of Mistral's Mixtral 8x7B, Grok-1 and Databricks' DBRX, which rival or even surpass GPT 3.5 with a relatively smaller size. Follow along to learn…
2023 was a great year for AI. Large Language Models were already in the spotlight for both users and businesses. ChatGPT had been just released in late 2022 and was taking the world by storm. Still, 2023 has brought more rapid change in the field than we could have imagined. This last year we got the newest version of OpenAI’s model, GPT-4. We also got a ton of open-source models competing with…
The problem with LLMs LLMs are very capable to perform a lot of feats that seem incredible to us, but they are bound by the borders of what Generative AI was originally meant to do: generate text based on the data it has been trained on. They cannot access any services on the web to give us more accurate and recent answers and instead, they are meant to answer using the training data that their…
LLMs do Natural Language Processing (NLP) to represent the meaning of text as a vector. This representation of the words of the text is an embedding. The token limit: the LLM prompting biggest problem Currently, one of the biggest problems with LLM prompting is the token limit . When GPT-3 was released, the limit for both the prompt and the output combined was 2,048 tokens. With GPT-3.5 this limit…
Language models have rapidly improved in recent years, with large language models (LLMs) such as GPT-3 and GPT-4 taking center stage. These models have become popular due to their ability to perform a great variety of tasks with incredible skill. Also, as the number of parameters of these models (in the billions!) has increased, these models have unpredictably gained new abilities. In this…
If you have been using SageMaker Studio, you might have missed some features that other modern IDEs have. As you might know, SageMaker Studio is built on top of the JupyterLab IDE. It extends it, integrating with other SageMaker resources. But this doesn't include any features for developer experience. To enrich SageMaker Studio with the features that you need, you can install extensions to…
Many enterprises choose a cloud environment to power the work of their data science team. If you chose the AWS SageMaker Studio, this article might interest you. Having both the data lake and the data scientist environment makes it easy to integrate them. You can choose what data any given data scientist is able to see. You might want a data scientist only to be able to use this data inside the…
Painter photo created by rawpixel.com - www.freepik.com A feature store is a storage system for features . Features are properties of data calculated through an ETL process or feature pipeline. This pipeline takes raw data and calculates a property from it. This property - usually a numeric value - will be useful to a machine learning model. It is important to find adequate, correct, and quality…
In recent years, the web development landscape has discovered that, while super stylized websites and web applications with tons of very rich interactions made with Javascript might be appealing to users, the loading time of the page can be a hundred times more influential to the user's perspective of the experience. As a result, there has been a ton of effort to reduce the loading time metrics of…