I haven't seen GPT saying such things about Sam Altman or Claude about Dario Amodei. Why every version of Grok is so attached to Elon.

oh no poor claude we're doing this today

why did chatgpt randomly lol'd NVIDIA

We need more 20B (or 30B) A3B size MoEs.

my synthetic data generation pipeline is slow, so obviously the solution is to rewrite the whole thing in Rust.

this is getting out of hands

Incredible release by NVIDIA. This model will be invaluable for latency-sensitive applications. It's small, fast, strong and fits on a single GPU, what more to ask for?

Me: Okay, it’s time for Gemma-4… Google: Introducing T5Gemma-2 🥳 Me: Alright then, Gemma-4 next… Google: Here’s Gemma Scope-2 Me: Definitely Gemma-4 this time… Google: FunctionGemma. Not to misunderstand, these are all absolute banger releases. But I’ll be patiently

why even the SoTA closed-source model does this? genuinely curious to know, is this quantization or some inference-level kernel bug..

If the rumours are correct Deepseek V4 and Qwen-3.5 before CNY

I shifted from MoE to Dense models and I've never felt better. I have more energy. My skin is clearer. My eye sight has improved.

once you start training on H200, you just can't go back to H100.

the exact same Hindi sentence tokenized by different models Qwen3: 222 tokens GLM-4.7: 212 tokens Nemotron: 84 tokens Gemma: 66 tokens GLM and Qwen3 are very strong models, but their tokenizers are dominantly trained on English text. As a result, they end up tokenizing almost

I got unlimited access to Cursor from smallest.ai. This might not be a good idea, I'm just getting started here.

Announcing... Voice x Memory! We’re unpacking what makes agents listen, respond, and remember, or sometimes forget, and what that means for building better voice systems. We will move towards a world of large LLMs remembering a lot of information to smaller LMs with finite

it's coming...

long day but it worked at the end 🫡