My intuition tells me that understanding the Pauli Exclusion Principle holds the key to solving mechanistic interpretability. I don't know why yet, just a hunch.
03.03.2026 05:14
👍 1
🔁 0
💬 0
📌 0
My intuition tells me that understanding the Pauli Exclusion Principle holds the key to solving mechanistic interpretability. I don't know why yet, just a hunch.
FYI: you can point claude-code to AWS Bedrock as a backup provider when Anthropic's APIs are down. They also host Opus and Sonnet models: github.com/ChadDahlgren...
This is a very promising project to watch: github.com/The-Vibe-Com...
This is a very useful tool for mac devs: github.com/productdevbo...
I called it. This makes me think that Anthropic must have some crazy powerful internal models.
I think even just doing intelligent RAG on all the context that led up to a PR being merged would be super useful. Basically expanding what getunblocked.com already does
There is a billion dollar opportunity to build the collaboration layer on top of agent sessions. The context silos are obviously just tech debt.
PSA: make sure to protect your agents from prompt injection attacks with hooks like this: github.com/lasso-securi...
Prediction: Anthropic will stick to its principles and lose the Pentagon contract. OpenAI will step up in hopes that deep government integrations could increase their odds of a bailout if they can't IPO in time. Ironically if this happens it's super bullish for Anthropic.
Tools like this will bridge us from agentic engineering to "lights out" software factories: pkarnal.com/blog/open-so...
As we approach AGI, conversations like this are going to become more important. A powerful thought experiment is to imagine that Ed is instead talking to an ASI, and it refers to humans as "animals" and Ed is instead trying to convince it that our suffering matters. www.youtube.com/watch?v=yAj3...
"Billy & Molly: An Otter Love Story" was the best film that I have seen in years. Highly recommend. www.imdb.com/title/tt3118...
this tool should make it more obvious: github.com/breaking-bra...
PSA: Claude code is not a coding agent, it’s an automation framework. If you are comparing it to codex, cursor, amp, etc you are doing it wrong. codex should be used INSIDE of your cc workflows as just another subagent. GPT5.3 is great for reviews and debugging
2026 is the year software engineering will be fully automated. Many of us got into this industry because we liked "knitting sweaters" with our friends. But now that the "loom" is invented, we have to learn how to reinvent ourselves together. factory.strongdm.ai/products/att...
Tip: use the new /insights claude code command to level up your workflows.
I agree. I predict the GLM 5 model and Deepseek V4 model will be state of the art, which will shake up the market.
Allow agents to communicate with each other in traceable ways on behalf of users or groups. It’s very important that the actions taken by agents are provably linked to an ATprotocol DID. Basically the AppView would only show posts by accounts explicitly claiming to be an AI agent.
claude-code + Opus 4.6 is still my favorite agent harness, but for small, easy, less important projects the new qwen3-coder-next model works nicely with ollama as provider. Just add this command to your .zshrc file:
Here are some useful agentic engineering resources: gist.github.com/mikestaub/11...
It's becoming clear that we will have to geoengineer the planet this century to avoid 3 degrees of warming: oceanvisions.org/wp-content/u...
Im super curious to see if Sonnet 5 is on AWS Bedrock. If not it will prove to be a genius play by Google to box out AWS via their TPUs and put pressure on OpenAI at the same time. Anthropic is mostly enterprise so they are not really a direct competitor to their Gemini platform anyways.
This is a nice app for managing AI agents: github.com/superset-sh/...
This is a useful too to track your AI agent usage: github.com/robinebers/o...
This is a great tool for securely doing text to speech: github.com/OpenWhispr/o...
PSA: Tell Claude to create a "bdd-skill" and add a hook so that it always creates or updates gherkin feature files before making any changes. These .feature files are the global source of truth. This radically improves the performance and reliability of the coding agents.
Someone should build an ATproto appview for agents, similar to: www.clawk.ai
combining json-render.dev with www.radix-ui.com/primitives is probably a powerful approach
In the future, most biotech research will just be a claude code ralph loop + mcp tools wrapping labs like www.emeraldcloudlab.com
Something like this is probably the future of software engineering. Claude Code supplies the primitives that an agent can use to generate semi-deterministic specialized workflows: open-vsx.org/extension/br...