Skip to content

ViralLoop

Your next post already worked.

← All creators

Posts
Likes
Typical
Best

Every post of theirs in the library

AllXLinkedInInstagramTikTok

FilterOrderGridTableSelect

Loading their posts

← All creators

wh

Writes about AI. 75% of posts open with an observation. A typical post reaches 374 likes.

Remix their style

Posts: 4
Likes: 5.5K
Typical: 374
Best: 4.5K

What they write about

AI: 100%

How they open

observation: 75%
number: 25%

Every post of theirs in the library

AllX LinkedIn Instagram TikTok

How to train a 670B parameter model. Let's talk about the DeepSeek v3 report + some comparisons with what Meta did with Llama 405B

Opens with a number

4.5K likes36 repliesAIX1y ago

Worth thinking about the compute gap. Pretraining compute for DeepSeek v4 is ~1e25 flops. OpenAI has 100K GB200s. Assuming all are used, and with a mere 15% MFU, the pretraining run would complete in just over a day (37 hours)

Opens with an observation

410 likes10 repliesAIX4mo ago

Muse Glimmer was trained directly logit distilled from Muse Spark. This means that there isn't a traditional 'base model' in that it was trained from the start on agentic traces. Super cool and haven't seen this approach in a while Welcome back GPT OSS

Opens with an observation

337 likes18 repliesAIX1mo ago

This is quite different from Mythos and is nowhere near as interesting imo This is a fine-tuned version of GPT-5.4 specifically for cybersecurity. Mythos was never explicitly trained for this purpose, and its capabilities naturally emerged

Opens with an observation

205 likes12 repliesAIX5mo ago

wh (@nrehiew_) — what they write about and why it works · ViralLoop