Steven Normore

You don’t always need a bigger LLM, just more diverse ones. So I built an ensemble inference proxy that sends prompts to multiple small models in parallel and combines their responses. Initial results look great! gpt-4.1-mini + haiku + qwen 3b (local): 74% accuracy. GPT-5 alone: 73%. Claude Sonnet: 74%. This ensemble config is 13x cheaper and 2.5x faster than GPT-5. And I haven’t even tested other providers yet. The trick: cross-provider diversity. Same-family ensembles do nothing. But models from different providers make different mistakes, and that's exploitable. Tested 27 configurations across 6 aggregation strategies. The best ensemble beats GPT-5 on knowledge tasks by 8 percentage points. Easy to experiment with your own configurations, just a YAML and emerge sweep. github.com/jnormore/emerge

Autonomous code/agent optimization: LLM proposes optimization ideas. Genetic algorithms evolve the best combinations. That's cEvolve. Benchmarks show ~60% faster convergence (before parallelization) and more likely to hit top-tier results. Inspired by @karpathy’s autoresearch. github.com/jnormore/cevol…

@GuiBibeau @eden_ What's DB routing in this context? 😅

@eden_ When DB routing on doublezero?

DoubleZero is faster than the public interent. Time, and time again.

If your app can’t be used by agents better than a human can use it, it won’t survive. The next generation of software isn’t designed for humans. It’s designed for agents acting on behalf of humans.

Introducing DoubleZero Edge. DoubleZero Edge is a new platform delivering real-time market data to traders and market participants, powered by multicast. Its first feed: raw Solana shreds directly from leaders, delivered over a high-performance fiber network. Infrastructure that levels the playing field for validators and matches traditional trading systems is here. Dive in ↓

the singularity has begun. so many signs.

Striking image from the new Anthropic labor market impact report.

The correct answer to tabs vs spaces is fmt.

oh i get it now @doublezero

App: data.malbeclabs.com Code: github.com/malbeclabs/lake

I built an AI-augmented data app over the holidays for DoubleZero. The leverage we have at our fingertips right now is hard to put into words. The question is no longer, can you write this code? but more like: do you know what you want? snormore.dev/blog/building-…

From our dev cave to yours. ❄️

Excited to release new repo: nanochat! (it's among the most unhinged I've written). Unlike my earlier similar repo nanoGPT which only covered pretraining, nanochat is a minimal, from scratch, full-stack training/inference pipeline of a simple ChatGPT clone in a single, dependency-minimal codebase. You boot up a cloud GPU box, run a single script and in as little as 4 hours later you can talk to your own LLM in a ChatGPT-like web UI. It weighs ~8,000 lines of imo quite clean code to: - Train the tokenizer using a new Rust implementation - Pretrain a Transformer LLM on FineWeb, evaluate CORE score across a number of metrics - Midtrain on user-assistant conversations from SmolTalk, multiple choice questions, tool use. - SFT, evaluate the chat model on world knowledge multiple choice (ARC-E/C, MMLU), math (GSM8K), code (HumanEval) - RL the model optionally on GSM8K with "GRPO" - Efficient inference the model in an Engine with KV cache, simple prefill/decode, tool use (Python interpreter in a lightweight sandbox), talk to it over CLI or ChatGPT-like WebUI. - Write a single markdown report card, summarizing and gamifying the whole thing. Even for as low as ~$100 in cost (~4 hours on an 8XH100 node), you can train a little ChatGPT clone that you can kind of talk to, and which can write stories/poems, answer simple questions. About ~12 hours surpasses GPT-2 CORE metric. As you further scale up towards ~$1000 (~41.6 hours of training), it quickly becomes a lot more coherent and can solve simple math/code problems and take multiple choice tests. E.g. a depth 30 model trained for 24 hours (this is about equal to FLOPs of GPT-3 Small 125M and 1/1000th of GPT-3) gets into 40s on MMLU and 70s on ARC-Easy, 20s on GSM8K, etc. My goal is to get the full "strong baseline" stack into one cohesive, minimal, readable, hackable, maximally forkable repo. nanochat will be the capstone project of LLM101n (which is still being developed). I think it also has potential to grow into a research harness, or a benchmark, similar to nanoGPT before it. It is by no means finished, tuned or optimized (actually I think there's likely quite a bit of low-hanging fruit), but I think it's at a place where the overall skeleton is ok enough that it can go up on GitHub where all the parts of it can be improved. Link to repo and a detailed walkthrough of the nanochat speedrun is in the reply.

The DoubleZero network is a thing of beauty doublezero.xyz/dashboard

A new, faster internet is here. DoubleZero’s high-performance global network is now live on mainnet-beta powered by 2Z. Welcome to the world of high-performance networking.

Transactions are a protocol short new post

New gen of open source AI models just started. Here is Llama4 by meta: huggingface.co/meta-llama/Lla…

Really enjoying the content from @MLStreetTalk lately.

🚀

Faster block/blob propagation in Ethereum using random linear network coding instead of gossipsub: > We propose a change on the way we broadcast and transfer blocks and blobs in the P2P network, by using random linear network coding. We show that we can theoretically distribute the block consuming 5% of the bandwidth and with 57% of the number of network hops. ethresear.ch/t/faster-block…