@tjrobertson52

Gemini 3 is currently the smartest large language model, but it's not the most useful. I wanna talk about why I'm using Gemini 3 less than I thought I would, why I still use chat, G, B, t and Claude, and why I still think Google is winning this race. The problem with Gemini is that it's lazy.

Google has always been focused on efficiency. They already had distribution, so they've been focused on a model that they can offer for free and integrate into their existing products. Downside is it doesn't spend as much tokens deliberating before giving you an answer.

However, it is very fast and I still think it's the best model for certain prompts. As I explained in another video, Gemini is the model most likely to give you the correct answer and most likely to give you the incorrect answer. In other words, it's the model least likely to tell you it doesn't know the answer.

It's like the smart kid in class that doesn't study for the test. It's the smartest model and also the most likely to hallucinate. Therefore, it's the best choice when the cost of hallucination is low or the risk of hallucination is low.

For example, if you're asking about something that you're pretty sure is well established, maybe you're asking a beginner question around a topic or industry you're unfamiliar with, Gemini can be great for that because the risk of hallucination is low. Or maybe you just need Help brainstorming a list of ideas where there is no official right or wrong answer. Gemini is great for that because the cost of hallucination is low.

Drafting simple documents that you'll review yourself, building simple tools that you can easily test. These are all things we commonly still use Gemini for. However, when the task isn't so straightforward and quality really matters, I just don't trust Gemini to spend the tokens required for that kind of work.

We almost always use Claude. We use Claude Sonic for any kind of writing and Claude Opus for any kind of analysis. The only downside with Clyde is sometimes it works too hard.

It can take a long time to get an answer and sometimes it runs out of tokens. And so occasionally we'll use ChatGPT as a middle ground. But right now, at least for work, that's the one we use the least.

So why do I still think Google's prime to win this whole thing? It's because they've already solved the hard problems. They have the smartest model.

It's cost effective and they have distribution. It seems like at any point if they wanted to, they could just have the model spend more token.