Introducing SubQ - a major breakthrough in LLM intelligence.
It is the first model built on a fully sub-quadratic sparse-attention architecture (SSA),
And the first frontier model with a 12 million token context window which is:
- 52x faster than FlashAttention at 1MM tokens
- Less than 5% the cost of Opus
Transformer-based LLMs waste compute by processing every possible relationship between words (standard attention).
Only a small fraction actually matter.
@subquadratic finds and focuses only on the ones that do.
That's nearly 1,000x less compute and a new way for LLMs to scale.
No business plan, no team, no investors, no pitch, no KOLs, just me building a product for 4 months and promoting it with 2 posts on #xiaohongshu.
Happy to see HyXpert, my #hyrox training app, getting its first users and growing.
Train hard and build (software) hard!
Start dense. Remove what doesn't survive the next iteration.
Backyard ultra is just a biological neural network running in real time — 100 athletes at the gun, one left standing at the end.
The brain works the same way. So does the best AI.
#sparsemind#cht#sparsity#ai
“AI is the new electricity.”
Everyone says it. Nobody asks: what happens when electricity itself becomes the bottleneck?
Scaling AI models is the wrong path and #sparsemind with its proprietary #CHT algorithm will provide a solution. Intelligence shell not depend on scale.
98% of memecoins die within 1 hour after launch.
This leaves most people broke and scared to invest again.
Over the years, I’ve perfected a strategy to avoid pump-and-dumps and only buy hidden gems.
Here’s my personal method (90% winrate) for spotting 100x memes👇🧵