If I had to burn down Accelr8 and start over, I'd build in one of these categories. We're at a moment where community, tech, and social experimentation are melding together. Ambitious people are willing to yeet themselves across the globe to find their tribe and build new lives.

I'm surprised how often Ad Hoc Rescues actually occur irl

tip: to get more friends tell strangers about your most important goals.

Instead of having human look at results and then tuning the prompt, you can do this automatically dspy.ai/#2-optimizers-…

Interested in fine-tuning an LLM on my writing so I can get better auto-complete in my text editor. I can LoRA fine-tune llama 8b on my own hardware 😋 github.com/axolotl-ai-clo…

Cool map of cognitive biases, many more than I thought existed: bias.wiki

think of self-healing for components of software systems that are brittle (such as web scrapers or parsers). I imagine when a website updates its layout, instead of needing a dev to fix it claude code can automatically fix the error. youtube.com/watch?t=1267&v…

perhaps in the future we will be comitting and reviewing prompts + specs and assume code is implemented correctly--like how programmers assume hardware correctly runs code. youtube.com/watch?v=IS_y40…

lesswrong.com/posts/vzfz4AS6…

In this case, the humans making the evals did not see the work-around and thus this eval failed. When upgrading models, are you finding that some previous evals failed because the Claude Opus 4.5 is just better? anthropic.com/news/claude-op…

That's right folks! It's JPM week, SF's biggest biotech event of the year. Here's our guide to some of the top events of the week (that you can still get into) 🧵

When I dance, it makes me feel like a Pebblesorter lesswrong.com/posts/mMBTPTjR…

I was surprised that the barrier to entry to do _something_ novel in LLM interpretability research is actually quite low. It maybe costs maybe at most $100 renting an A100 on runpod

You can use Terminus to control a Claude code instance on your remote server from your phone, you can check on the little guy in the brief moments of boredom

I'm working on a clone of rewind.ai but making it open source. github.com/MattHandzel/li… Use case: I got an email saying I didn't fill out a form but I remember doing so. I sent a screenshot of the form filled out and got it resolved

Another use of having a second brain: I'm doing applications and one question it asks is "Whose voice should carry more weight in the technology ecosystem than it does today?". To answer this question, I go to my `people-i-admire` document, choose someone, and write about them.