are there reasoning models that are better at big-picture thinking/planning, versus lower-level implementation? i’m wondering whether a hybrid strategy of models would be optimal; my workflow at the moment is to plan and theorise with o3, then cursor+claude 4 for implementation

I remember Alex Cheema - e/acc's viral video of a small Mac Mini cluster: "Nemotron 70B at 8 tok/sec and scales to Llama 405B". I requested benchmarks, discovered how awesome Alex is via Zoom, and built the initial stages of benchmarks.exolabs.net Now Exo v2 launch incoming!

What did I say? Honeymoon effect. The quality drop is in your head. We’re in the second S-curve. Embrace it or die.

it was a pleasure to be asked to give a talk on our paper ‘SPARTA’ at ICLR 2025. distributed training isn’t a fantasy any more; with algorithmic improvements like this, training models over low-bandwidth environments becomes a reality read the paper here: openreview.net/forum?id=stFPf…

great things come from cold dms Naval cold dm’d me and invested in EXO Labs hired Matt Beton after i cold dm’d him our first customer came from a cold dm you can just do things

fml linkedin is unusable

Thrilled to share that I’ve started my new role as a Senior Engineer at Qualcomm Research in Amsterdam. I’ll be joining the Model Efficiency team, where I’ll continue research on quantization and compression techniques for machine learning and AI. Qualcomm Qualcomm Research & Technologies

He applied to EXO Labs last year

We’re already doing this with EXO Labs Last month was the first trial, we provided free M-chip public cloud access to developers at a hackathon. These were M3 Max/Ultra Mac Studios with up to 512GB unified memory. Awni Hannun gave a talk at the hackathon on how to leverage MLX

pump is one of the fastest growing startup ever. 0 to $1B ARR in 9 months. 25% of revenue for $PUMP buy backs is insane, i'm predicting this ends up in the top 10.

if they ever tell my story, let them say I walked with giants; men rise and fall like the winter wheat, but these names will never die.

A new approach to efficient large scale distributed training on Apple Silicon. Most AI research today is focused on traditional GPUs. These GPUs have a LOT of FLOPS but not much memory. They have a low memory:flops ratio. Apple Silicon has a lot more memory available for the GPU

EXO isn't just for inference.