Browse Comments — Raw (as collected)

Building diverse models that listen and interpret differently really resonates; I’ve noticed subtle tone shifts make a big difference in customer conversations.

shifting from monolithic models to ensembles feels like the next frontier

Impressive approach. Ensemble architectures balance specialization and efficiency, capturing nuance lost in monolithic models. Real-world impact will depend on orchestration, scalability, and integration into existing workflows.

Soooo when can I use this in replacement of Claude code??

Im still a bit confused where anything here is "ground breaking"

Even from my very simple point of view this seems makes sense: I’m always having to choose ‘which’ model to run, but the relaty is that difence model are better at diffent things. If I had an orchestrator to help parse my requests and send it to those models then bring it all together I see both higher quality product and lower use on each model as it only gets the pieces right for its ‘area of expertise.’ Is that anywhere near a laymen’s articulation of what I just read on your site, Mike, and how could apply this myself - or do I have to wait for services I use to start engaging ELM or you all to broils a centralized orchestrator I can use as front-end model?

I think this is brilliant. But I don’t anticipate M2M or A2A (model to model or agent to agent) communication, which is clearly where the market is going even for consumer products, to be voice based

Really interesting research. The idea of orchestrating an ensemble of specialized models is basically what our brain is compiled of. While the brain may be the orchestrator, each section of the brain is optimized for specific tasks like emotion, memory, reasoning, etc. Intelligence comes from how those specialized parts work together. Seeing AI engineering move towards mirroring biological architecture is amazing. Well done!

Impressive innovation, Mike! ELMs seem like a game-changer for voice understanding and efficiency—excited to see how this transforms AI applications

Mike, awesome work. Do checkout some research under similar umbrella: https://arxiv.org/abs/2508.14314

This is a fascinating approach! Excited to learn more and see where this goes. Congrats on your success so far!

🥱

How does it compare relating to energy requirements? Deepseek might beat chat GPT on cost, but it presents significantly higher emissions per query based on available studies. Being able to beat them on both fronts seems like a real prize.

It’s like going from a computer as big as a room to putting it on your desk; we’re seeing something amazing happening here

Scaling intelligence vs designing intelligence. Big shift.

By using semantic compression we can reduce AI training cost from 40% to 70%

Looks suspiciously like the “integrator” in consciousness theory!

Awesome! Seems you are saying it's a giant DAG implemented with LangGraph most likely? Or really some new LLM architecture. both are awesome. Weights or code available?

When do we get "HER" . Congrats my greek brotha

I love the “harkening a new era of AI” line. Like - 90% of the world hasn’t even started with THIS era of AI. How are we at the NEXT? 🤔😂