Kishan

@kishml.bsky.social

Co-founder @ Leadmatic Building AI Agents, sharing cool stuff happening in #Elixir, #Python, #MachineLearning

It's happening! Llama 3.3 packs similar performance to Llama's 3.1 405B parameter but at 70B parameters! Expect 2025 to be the year LLMs get smaller and faster. Its a matter of time before we get GPT-4 level models that can run on local devices. venturebeat.com/ai/meta-laun...

Meta launches open source Llama 3.3, shrinking powerful bigger model into smaller size

The 70B-Llama 3.3 is specifically optimized for cost-effective inference, with token generation costs as low as $0.01 per million tokens.

venturebeat.com

Plenty of people will sign up for ChatGPT’s pro plan for its novelty, but I still think the future is in Small Language Models. On device models for knowledge retrieval and tasks will be used more frequently than large models for complex reasoning and the best of these models will be open source.

(Elixir + LiveView + Oban + Flame) is an awesome tech stack for CPU/GPU workloads! Elixir = a functional language with fault tolerance Flame = Spins up VMs to execute functions and powers them down when done Oban = job orchestrator that handles fault tolerance LiveView = real-time UI tracking

Effortless Video Sharing with Phoenix LiveView and FLAME

As someone who loves experimenting with new technologies, I recently got the Vision Pro (I’m a big fan of VR and the idea of working in mixed reality). What excites me the most is sharing the cool stu...

poeticoding.com

Bsky is awesome! Feels like Threads in the early days. Hopeful it’s a place people keep coming back to. That never happened with Threads