@AaronAsherRandall

Gemini 2.5 Flash is a life saver! All of these different LLMs have pros and cons, but with Gemini’s context window and ability to deeply understand my code base, I always go to it when I am out of solutions. And I’m still on the free tier!

@CodingWithLewis

Just tried Gemma 3 27B on my Nvidia card and it runs amazingly.

@Caper_The_Cat

I'll wait for the Fireship video...

@chwydev

Gemini's costs are CRAZY cheap. I cannot emphasize how insanely funny it is seeing a dollar check for an AI model a month.

@germandkdev

I wonder why the 1.5B quantization paper doesn't seem to get implemented that often

@dill-frescott

gemini 2.5 Pro is a beast. It has some quirks but overall is an amazing model.

@dultzywaltzy

1960s: Space race
2020s: AI race

@IDNKEK

I am kinda waiting for deepseek r2 rn. Its either will be just ok by current standarts, or some absolute banger as r1 was

@Dopachen

totally agree, also tools like NotebookLM are insanely underrated for learning. feels like a lot of people are sleeping on it. google’s seriously ahead in the AI game right now

@Michael-ty2uo

When it comes to the AI race it really depends on who has the most data and google wins by a landslide

@maritaeques

Lewis we need subtitles! (For when we are out in bars with loud music but still can’t miss you dropping a new short 😬)

@ST4RLO91

ah yes they just use INT8 quantization instead of FP32

@anya-forger923

U should make an video about unsloths 4 bit quants

@waynelau3256

One of their top researchers joined back, wondering if that helped bring them back into spotlight

@anya-forger923

Gemma 3 is so good. I run the 12b ubsloth models and its just soo good.

@droppyalmond1111

They are cooking so much, GOATed!

@Heheh01737

Google ai studio hands down is what you can get out of ai right now i could upload my whole freaking project for free it was like 20000 lines code(it had html templates ) flask and some shii and it did the thing it gets it dont its like a combination of mistral (for the 1m token) and the mind of claude it gets the job done

@trananhkiet167

Ah in4 . peoples can do it on open source to reduce VRam usage  . i optimize it using current open source tool . it could significantly reduce VRAM need for training , however it require better data input due to the method will reduce accurate of model .

@killer_whale_tactics

What is the best programming languages to make 2d and 3d games for beginners?

@calvincaesar6182

In my use case at least with the gemini on android 15. I still prefer GPT over Gemini. Mostly because I also use GPT for conversation. Gemini conversation is so boring and sometimes just plain stupid