NVIDIA's new Blackwell arquitecture could accelerate the process to reach General Artificial Intelligence in 5 years!

Definitely, artificial intelligence is covering everything at an incredibly fast pace. A few days ago, Jensen Huang introduced us to a new BlackWell GPU architecture focused on accelerating and improving the training of large language models (LLMs) that require processing a huge amount of data in the shortest time possible. Everything indicates that these processors will be capable of ushering in a new technological era for AI and enabling advancements in all computing areas.

---

---

We won't delve into technical details, but we will mention the capabilities of these processors which we could compare with current ones. These GPUs contain around 208 billion transistors manufactured through a process called custom TSMC 4NP. To make a comparison, I did a quick research and noticed that one of the latest Intel i9 processors has up to 14 billion transistors.

Obviously, their use could be focused in different directions, so perhaps this comparison may not make much sense. What does make sense is that we are starting to witness a technological revolution that involves not only an improvement in processing data of generative language models, but also robotics.

We recently saw Figure, the new robot from OpenAI, which pleasantly surprised me with its ability to reproduce much more "human-like" movements. It seems clear that the fundamental idea behind this advancement involves replicating human behavior to the point of potentially achieving a consciousness-like reasoning.

---

Aideal Hwa | Unsplash

---

I dare to think that one of the objectives in the coming years will be to introduce robots into various work areas. In fact, that's what the figure.ai site expresses. Considering that there will be significant competition in developing humanoid robots (such as Tesla's Optimus), inevitably competition will also arise in the architectures of other companies; therefore, it wouldn't be surprising if Intel is already trying to position itself in the market focused on creating AI-based chips.

While there are still many obstacles to overcome, efforts are being made to refine any type of interpretation error that may arise. The most prominent issue is hallucination.

Many times we have noticed that when using a prompt, AI may not have an answer to our question and tries to formulate a response that sounds good but may lack sense and be incorrect. Huang, at this point, argues that a deep investigation is needed to find the responses that align closest to a better answer, which obviously is one that holds a known truth approved by all.

It involves analyzing sources and contexts, and if the answer is not precise, additional sources are consulted. If the language cannot provide an answer, it will indicate so.

The general artificial intelligence (AGI) will be polished in a very short time if we remain expectant of the progress being made. Huang maintains the idea that we might see AGIs performing tasks similar to humans in 5 years, but currently we are in the narrow AI stage; which entails simple processing tasks like detecting product faults.

It is also true that companies will invest in this technology, so there is nothing more to say than the future will be oriented in this direction.

The big question is, how will we evolve as a human species? That's something we should be concerned about before control falls entirely and largely into the hands of AIs.