AI in 2025

💻🤖🧠

“Because innovation like the former can’t exist without the likes of the latter….”

Meme by KK the memelord (Me).

Large Language Models (LLMs) are the dominant type of models that make for most of the AI.
May not be the best or the path to Artificial General Intelligence (AGI) or Artificial Super Intelligence (ASI), which are the points that are equivalent to, or surpass human level of intelligence.
AGI could come in 2030-2035, says the Nobel laureate and DeepMind CEO Demis Hassabis.
Google is outperforming all other models in most tasks with their Gemini 2.5 Pro (NLP) and Veo 3 (Video generation) models. The previously King in AI is back, it seems.
Google’s offerings are also nuts, giving free access to API and developers.
There are other types of models catching up, like JEPA (Joint Embedding Predictive Architecture from Meta, latest: V-JEPA), which has a more grounded undertanding of the physical world, meant for robots and other applications aiming to achieve Artifical Machine Intelligence (AMI). AMI could be one of pathways to AGI.

📈Are stochastic (probabilistic) simulations of people
🧠😕Have anterograde amnesia - they do not know of anything beyond the point of their training
🧠🤯ADHD (Attention-Deficit Hyperactivity Disorder) behaviour, hallucination: Get distracted in their vast amount of knowledge easily.
🧠🤓Jagged intelligence: smart mostly, but dumb a few times. When asked how many ‘r’s are there in ‘strawberry’, they often say the wrong answer. This is mainly because computers do not think the way we do, their thinking is upside down - They can do rational/complex math more easily than do intuitive simple tasks like identify a fruit in a basket. For a two year old human, identifying a fruit is easier than doing complex math. (more on How We The Humans Build
🏃‍♂️❌Don’t continuously learn or think autonomously
😴❌Don’t sleep after that to consolidate knowledge or insight, expertise into weights (that happens only when the model is being created): Humans do that every night.
🤔❌Lack true reasoning and understanding: As they’re just stochastic simulations of people, they really lack true understanding
🔠❌🔢✅They see tokens, not letters or words: LLMs think in terms of tokens, each word is converted to a token for the model to process and once that is done, it is converted back to word for you to read.

AI needs to be kept in leash, says Andrej Karpathy
We need to augment✅ the AI like the Ironman does with his suit, instead of letting AI do it all the agentic❌ way.

Biggest revolutions of this century (after the Internet):
1. AI (Software + Hardware): The software wave is already here. The hardware/robotics wave is next—and it’s not far behind. This revolution is moving fast, largely because it doesn’t need heavy regulation to scale.
2. BioTech: Possibly even more transformative and more ethically complex. Its pace may be slower due to strict regulations, as it directly involves the human body. But its implications run deeper: it will force humanity to rethink morality, identity, and what it means to be human. Neuralink, for instance, is already tackling things like blindness and aims to augment human capabilities in ways that could permanently reshape how we live. BioTech is being accelerated by AI, and while it’s not massive yet in 2025, brace yourself - the wave is coming.

GDP will obviously grow because of increased productivity, hence economy may not suffer (unless there are other factors that I’m not aware of).
So even if people lose jobs, there may not be an Economic Depression.
There will be a huge crisis for meaning. Because just getting money and resources will not make humans happy, that is not how we’ve grown evolutionarily. We need meaning. There is a huge possibility of an Existential Depression or Moral Disintegration.
But the number of problems to solve will never be zero, there is cancer to cure, planets to colonize. Hence there is always a way to look for meaning.

Future is here, and its definitely not boring.

Onward, to solve the next great problems. Onward, to the stars….

🌎🚀