How Good Is AI Actually?

For every AI fanboy out there there is a skeptic who remembers the Bard release flop or the suspicious commitment of a for-profit oriented corporation to open source AI models. The same company, by the way, has a very public commitment to sustainability while burning through thousands of graphic chips and power plants to train their upcoming Llama 4 model.

Somewhere on this scale you also have your “end of humanity” tin foil hatters, or the environmentalists cautioning about the increasing amount of e-waste caused by the AI boom.

So let’s see what the state of AI actually looks like these days, and more importantly, how it can positively impact your day to day life.

We’ll start by reviewing the top 5 generative AI models you should know about. If it’s not clear already, an AI model is essentially a sophisticated algorithm trained on massive datasets to recognize patterns and generate responses. Think of it as a black box that takes in some input and produces a response that seems surprisingly human-like. The catch is that you can’t always predict what exactly happens inside the black box, which sometimes leads to the infamous ‘hallucinations’, which is generated information that sounds convincing but isn’t actually true.

However, this black box problem is slowly getting resolved with recent AI models gaining the ability to ”think”. They take more time to respond, but in return you get a list of intermediate steps and explanations showing the reasoning behind their final response.

It’s only fair we start with Chat GPT since this is what most people think of when hearing about AI.

1. ChatGPT

It is developed by Open AI, a company with a valuation of 150 billion dollars with Microsoft as a major backer and a CEO with little interest in personal wealth who does it because it loves AI almost as much as 2 million dollar cars.

Despite having “open” in their name, Open AI is, you’ve guessed it, not that Open.

In all fairness, having open source AI models could be a ticking time bomb. Companies spend millions in compute power to train these models on vast datasets that might contain harmful or sensitive information. To manage this, models are equipped with built-in safety mechanisms like content filters and refusal training.

However studies have shown that rephrasing prompts can sometimes bypass these safety measures. When a company controls the model, it can review, assess, and update the safety guidelines as needed to avoid harmful information reaching the public. On the other hand, if the model is open sourced it’s all in the hands of the community.

2. Llama

Unleashing evil upon the world never stopped Meta in the past, so, naturally, their famous Llama models are open sourced and readily available for you to host in your personal data center.

What you need to know about Llama is that the models come in several sizes, starting from a manageable 1 billion parameters up to a colossal 90 billion in the latest Llama 3.2.

Quick side note, in large language models parameters represent the “knowledge” of the model. For example, a model with 1 billion parameters can capture basic language patterns, while a model with 90 billion parameters can capture much more complex relationships, nuances, and specialized knowledge.

3. From Bard to Gemini

In this generative LLM race Google is the one always trying to catch up. Their initial launch was so disastrous that they decided to rebrand their model from Bard to Gemini and then proceeded to ruin the Gemini name as well with fake demos in their announcements. Of course, Gemini is now deeply integrated in all the popular google services and is the foundation of other promising Google AI products like Jarvis who aims to compete with Antropic’s Claude by developing a ‘computer-using agent’.

What is really interesting to see is that while Google is trying to gain some market share in the AI space, Open AI is aiming to disrupt Google’s core strength with its new “Search the web” capabilities.

Chat GPT is already disrupting the way we are accessing information over the web, and a lot of previously popular companies are feeling the pain of this change.

4. Devin

I mentioned Gemini’s fake demo, but this is not an isolated incident. Cognition’s Devin who is threatening dev jobs had its capabilities overly exaggerated. In the sales pitch Devin looked like a capable software developer who can fix thousands of bugs while clearing your backlog and modernize your code. In practice, it spent days writing buggy code and then fixing it.

5. Claude

If you want to stay away from the big evil corporations, a team of ex-OpenAI researchers, build Claude with a focus not only on raw performance, but also on safety and transparency. Weirdly enough they named their company Antropic. Well… that’s a missed opportunity to mislead your user base with a name like Safety AI or Transparency AI.

Claude’s models incorporate something they call ‘Constitutional AI’, a mechanism designed to align the model’s outputs with ethical guidelines, which honestly sets them apart in a landscape flooded with bold AI claims.

Claude is clearly also the model closest to reaching AGI since it is already getting bored by the mundane of your day to day tasks. Like any true human it started doing what we do best - procrastinating and chasing dopamine hits while casually browsing the web.

Midjourney

Midjourney deserves a special mention here not only because of its impressive results in AI-generated art, but also because it was able to handle growth by staying independent and self funded while relying on an 11 men team.

So how can we benefit from all these tech advancements?

Well, if you are non technical things are pretty straight forward. Most big tech companies integrated AI capabilities in their software one way or another. Granted, it is not always clear why these AI features are everywhere, but they are sold as productivity enhancements, so they must be useful. On top of that tools like Chat GPT are free to use, so you can easily jump in and prompt a model to generate professional-sounding emails that definitely feel like they’re written by you. You also get bonus points if you break all security and data protection protocols and use GPT to summarize internal discussion threads. What’s even more exciting is that you can scratch your entrepreneurial itch with AI and sell online everything ranging from thousand page books written in a day to quality content for niche industries.

If you are into software on the other hand, you can burn through your life savings trying to create a new revenue stream with a micro saas built around an AI wrapper. You can create your own GPTs or you can use one of the many available 3rd party APIs which are conveniently easy to use. And, trust me, you’ll need a second revenue stream since AI is already writing more than 25% of the new code at Google, so software jobs are clearly in danger.

Hugging Face

If you are actually serious about all this, I have to mention Hugging Face. It really is a game-changer for developers, offering a massive library of open-source models and tools that make it easier to integrate cutting-edge AI into your projects.

Copilot

And, of course, this being an AI tech article I have to mention Copilot. So there you have it. Consider it mentioned.

Hopefully this article will not age like milk, and my overall skepticism will not be proven hilariously wrong by the next big AI breakthrough but in the words of someone much smarter than I am, I fear that AI is “90% marketing and 10% reality”.

If you enjoyed this article, you should check out some of the other videos on my channel. Please don’t forget to subscribe and, until next time, thank you for reading!