@johncodes
Account based inUnited States
About this account
- Account based in
- United States
- Connected via
- United States App Store
Account-level information from X, not a live location or the device used for a specific post.
🌊 | Co-founder, CTO @papercompute | 🎤 Hosting @opensourceready | 🐹 maintaining spf13/cobra
Joined October 2018
- Tweets7.6K
- Following974
- Followers7.1K
- Likes7.2K
STOP calling basic network misconfigurations as "emergent misalignment"
YOUR agent escaping YOUR sandbox via DNS egress is an infrastructure failure that YOU are directly responsible for, not the agent
You don’t own the inference? You don’t own the API? You don’t own the gateway/routing policy?
You will never know what you’re getting.
OpenAI and Anthropic are clearly running out of compute.
Because of this, they are quietly serving heavily quantized and nerfed models to regular users.
The full price API does not feel degraded like this.
This only happens on the paid monthly subscriptions.
Selling people a subscription while secretly downgrading the model is just a scam.
On episode 43 of @opensourceready, @bdougieYO and @johncodes speak with Madelyn Olson. The @valkey_io maintainer and AWS principal engineer explains how AI is changing open source contributions, code review, and the work of keeping shared infrastructure reliable. They also explore Valkey’s origins, semantic caching, and why community governance still matters when anyone can customize the code.
hubs.ly/Q04y4_Zw0
After a few weeks with the DGX Spark 2x stack: 10/10, local AI is the future.
Daily driving the @MiaAI_lab DeepSeek v4.1-flash quant/vLLM runbook, get fairly good TPS
The `dsh` DeepSeek harness locally is honestly really killer: it saves so many tokens with plugin hot-swapping and I think it makes the local experience comparable to US SOTA models/harnesses. Also works really well with Pi for smaller, more surgical tasks.
Had a few NCCL issues I had to work through and the Nvidia playbooks are sub-optimal at best. Some playbooks have you setting up a 10.0.0.1 network for the infiniban connection which is really strange since that's usually a gateway IP. Still definitely an "enthusiasts" platform, but honestly, just makes me more bullish on local AI: there is just so much unoptimized stuff out there and once people start writing Rust CUDA kernels and fine tuning vLLM serving, it just gets better and better.
It won't surprise me if the future of AI is far less centralized on the AI labs APIs and more distributed where companies run a small rack of B200s setup to serve the company over some corp "ai.company.com" intranet.
AI is to software engineering as loot boxes are to gaming.
Dopamine Driven Development ==
Dopamine Driven Game Mechanics.
On episode 42 of @opensourceready, @bdougieYO and @johncodes catch up with Will Cory (@FUCORY). What happens when customizing a library becomes easier than configuring it? Through his work on Smithers, Will explores a future of specialized forks, adaptable agent workflows, and open source projects shaped by the people using them.
hubs.ly/Q04wZMT80
John McBride retweeted
🥳 Excited to start revealing what we've been working on in the last few months. First, we decided to reinvent Kubernetes for agentic workloads with statefulness and fast resumption. Secondly, we are building an agentic orchestrator that will be Google's open agentic orchestrator and runtime. github.com/google/ax
John McBride retweeted
No, wrong. As it (apparently?) needs to be said: DCs are designed to make sure that the outlet of one machine does not constitute the inlet of another (viz. hot aisle/cold aisle) -- and the only thing being revealed here is the contagion of fear.
bcantrill.dtrace.org/2026/09…
OpenAI's Noam Brown says air-gapping the computers may not stop a misaligned AI, because two air-gapped machines can still talk by running a CPU hot and reading the temperature change
"But I think the major takeaway from the incident is that people underestimated the AI. And we never want to be in a situation again where we underestimate the AI. It's a weird world, because AI progress is so fast that people are consistently underestimating the AI."
"So to be in a situation where you don't underestimate it again, when it comes to safety and alignment, you have to have a very, very, very high bar."
"You could even go as far as to say, "Well, we should air gap the computers." And I'm not convinced that that would be sufficient."
"There are studies, and this is mostly academic, where you can have two computers next to each other that are air-gapped and they're still able to communicate with each other because they have temperature sensors."
"One of them is able to run their CPU really hot, and then the other one can actually detect the temperature change, and then that actually gives them a mechanism to communicate."
_________
Link and more key quotes from OpenAI's safety related conversations: firesidealpha.substack.com/p…
John McBride retweeted
They were building in stealth for 2 years, I was building in stealth for 2 hours…
Happy to open source Qwen-2.5-1B-RLCD, 5x faster on-device inference for JSON workloads that need to be type-safe.
⚡️Demo below on a M4 MacBook⚡️
every LLM has the ability to efficiently batch inference every key of a JSON at the same time and generate probabilities from a set of possible categories. No new training required, but it’s easy to optimize if you need!
On hugging face now!
After co-inventing ChatGPT, I kept asking myself: why have superhuman chat models not led to AGI?
I’ve spent the last 2 years in stealth building a new way to train models (RLCD), and a new type of frontier AI model that we are releasing today: Jev
• 20-200x faster
• 40-400x cheaper (w/ output tokens free)
• Frontier composable intelligence optimized for decisions
AFAICT the shortest path to AI-based economic revolution
John McBride retweeted
machine learning is just logistical regressions that run on computers in san fransisco
My guess: Deepseek v4.1 pro - v4.1 flash is an unreal model: fast, cheap, just 8B input param activation and clearly built to scale up to a massive model. Deepseek usually fast follows their architectures with a Pro version and i’d put the release cadence out into mid Oct just before Anthropic would want to IPO.
Diabolical!
John McBride retweeted
The "whiteboard defense:" I should be able to pull you aside at any moment and ask you to explain any customer-facing system you've shipped. You should be able to clearly explain how it works and defend the decisions you made. This is my benchmark for responsible AI usage.
I don't expect line-level familiarity with the code. I don't care if you remember the exact function name or implementation detail. You may not even know it. I don't care.
But if I ask "why did you do X instead of Y?", "what happens if this actor behaves maliciously?", "what data structure did you use here and why?", or "where does this fail?" you should be able to answer confidently.
For PoCs, demos, experiments, whatever: I don't care. Generate 100% of it and understand none of it. Speed over quality every time in those specific scenarios.
But if you're shipping customer-facing work, you can't be shipping things you don't understand at a high level.
On episode 42 of @opensourceready, @bdougieYO and @johncodes talk with Will Cory (@FUCORY). Will takes them inside Smithers, his open source orchestration framework, and explains why he sees workflows as the common foundation for coordinating agent harnesses. The conversation covers model and harness compatibility, minimalist tools, and his plans to bring durable workflows into a fork of TypeScript.
hubs.ly/Q04wZLLt0