Near the end of 2025, Google was on top - best LLM (Gemini 3 Pro), best video model (Veo 3.1), and best image model (Nano Banana). They seemed unstoppable.
And then, they stopped. Why? Are they having trouble with TPU?
When even Meta can come out with a better LLM than Google, something must be terribly wrong.
>Alphabet Inc.’s Google is months behind schedule on delivering Gemini 3.5 Pro, its most powerful flagship AI model, because the company has been taking time to try to improve its capabilities, particularly in coding, according to people familiar with the matter.
Seems like deepmind needs to get more internal usage
Near the end of 2025, Google was on top - best LLM (Gemini 3 Pro), best video model (Veo 3.1), and best image model (Nano Banana). They seemed unstoppable.
And then, they stopped. Why? Are they having trouble with TPU?
When even Meta can come out with a better LLM than Google, something must be terribly wrong.
>Alphabet Inc.’s Google is months behind schedule on delivering Gemini 3.5 Pro, its most powerful flagship AI model, because the company has been taking time to try to improve its capabilities, particularly in coding, according to people familiar with the matter.
Seems like deepmind needs to get more internal usage
I recommend that every AI user start with Gemini as their first AI.
Use ChatGPT, and it's ask a question, get an answer, done. Tell it to code, get the code, done.
But use Gemini? You get burned by the hallucinations Gemini casually spews out on a daily basis.
You develop a habit of fact checking every single answer.
You learn that blindly trusting the AI's response will get you absolutely screwed.
AI answers must always be cross verified.
That habit prevents a major disaster later on.
It thinks the current year is 2025.
You hear that Gemini 1.5 Pro is the latest.
You learn about the AI knowledge cutoff.
You find out that Gemini hasn't learned anything past 2025.
That naturally leads you to think, if I want to ask about something recent, I need to tell it to search.
You wrestle with prompts and pick up prompt engineering.
Why can't Gemini remember what I said earlier?
You look it up and learn the concept of the context window.
The official context is one million tokens, so why does it get a lobotomy at a hundred thousand?
Ah, the Transformer architecture scales computation quadratically with input length.
And that's how you learn the concepts behind AI.
Ah, Gemini is skimping on context to cut costs.
To use this properly, I need to go to AI Studio.
And that's how you learn about cost management in the AI industry and the existence of developer tools.
In AI Studio, they say you have to use the API. You try it and get billed 80$
And that's how you learn about the concept of APIs.
All of this is the know how of a big corporation, handing out free trials while managing costs.
And that's how you learn about marketing and business management.
You code with Antigravity and experience your code turning into a complete train wreck.
Ah, AI generated code must absolutely be reviewed by a human.
That habit prevents a major disaster later on.
Looking at the code, you see it hardcoded everything again.
It wrote the same logic three times.
It crammed five thousand lines into a single file.
Let's figure out how to fix this dumpster fire of a codebase.
And in doing so, you end up actually studying the code.
Watching your quota drain in real time after just a few responses in the consumer app,
you learn restraint and planned consumption.
The person who used ChatGPT got an answer to their question.
The person who used Gemini gained wisdom for navigating life itself.
Hahaha, I wrote a meshcore firmware for the lilydeck t-go plus exclusively using Gemini and antigravity, just to give it an honest try.
Every single line you wrote is based in truth.
It's really decent now, it was a pile of yuck at the end of 2025