j_bum

I think the parent comment is simply referring to “@“-ing files in the chat.

So if you want CC to edit “file.R”, the prompt might look like:

“Fix the the function tagged with ‘TODO-bug’ in @file.R”

That file is then prioritized for the agent to evaluate.

It is awesome.

But what I’ll say is, ideally they would demonstrate whether this model can perform any better than simple linear models for predicting gene expression interactions.

We’ve seen that some of the single cell “foundation” models aren’t actually the best at in silico perturbation modeling. Simple linear models can outperform them.

So this article makes me wonder: if we take this dataset they’ve acquired, and run very standard single cell RNA seq analyses (including pathway analyses), would this published association pop out?

My guess is that yes… it would. You’d just need the right scientist, right computational biologist, and right question.

However, I don’t say this to discredit the work in TFA. We are still in the early days of scSeq foundation models, and I am excited about their potential.

Just a heads up, your URL 404’s

I'm not sure I understand what's going on here. Maybe a broken link?

When I click to play, there is no Wordle-like game. It's a shape tracing game.

Am I missing something?

> “We have a cat and a dog and a child and an adult in the same room,” he said. “Why not an A.I.?

Because pets are living creatures with agency that form emotional bonds to their owners. Pets also can’t sycophantically talk back to us.

I’m just as stunned as any other at the ability of LLMs/agents to code, plan, execute, etc.. But these sycophantic stochastic parrots do not have agency or emotion in the historic sense of those words.

LLMS can cause dangerous mental health outcomes for the unequipped who don’t understand that they don’t have agency.

No problem! I’d love to hear any approach you’ve taken as well.

Sure thing. Here is an example set of the agent/spec/to-do files for a hobby project I'm actively working on.

https://gist.github.com/JacobBumgarner/d29b660cb81a227885acc...

This was a fun read.

I’ve similarly been using spec.md and running to-do.md files that capture detailed descriptions of the problems and their scoped history. I mark each of my to-do’s with informational tags: [BUG], [FEAT], etc.

I point the LLM to the exact to-do (or section of to-do’s) with the spec.md in memory and let it work.

This has been working very well for me.

I’m almost 30 and made this change about a year ago.

I now rotate between high rep (sets of 20 rep max) and medium weight weeks (sets of 8-12 reps). My joints haven’t ached in a while and I’ve become much less prone to random muscle tweaks. Mike Isreatel has an excellent intro to high rep training [0]. It produces pumps and mind-muscle connection like nothing else!

I actually went too far into the high rep/volume training direction for several months, but realized I needed to reincorporate medium weight lifts when I started losing a bit of grip strength. I am now super content with my current rotation cycle!

[0] https://youtu.be/HzFHAHOOA4A?si=avUNYahKGPoHbYph

Best of luck for your continued recovery. Keep at it.

I can’t tell you to do this because I don’t know your medical history, but slowly working your way up to medium weight training can also be a game changer. Re: squats, Romanian deadlifts, pull ups, dead hangs.

Would you be willing to point me to a primer of how I can get started with building agents?

This week I experimented with building a simple planner/reviewer “agentic” iterative pipeline to automate an analysis workflow.

It was effectively me dipping my toes into this field, and I am so floored to learn more. But I’m unsure of where to start, since everything seems so fast paced.

I’m also unsure of how to experiment, since APIs rack up fees pretty quickly. Maybe local models?

I’d love to hear how to start today efficiently. Please do share!

Sincerely, A true night owl.

The search is very impressive. It uses CLIP.

So you can search “black cat on blue carpet in the morning” and expect to see a result (if it exists)

I haven’t listened to this podcast, but I listened to an excellent RadioLab podcast a while ago on the topic. They ended the podcast by discussing some of the ethics of “fixing” aphantasia, many of which I had never considered.

I recall them mentioning: 1. the ethical challenge of arguing that aphantasia is something that needs “fixed” in the first place 2. The unknowns of what might happen to someone emotionally if they go from nothing to something. This might sound odd, but we know that hyperphantasia can be associated with schizophrenia and other neuropsychiatric issues. 3. The implications of downstream cognitive “enhancements” that might result from this.

I have aphantasia, and I do not think I’d want it “fixed”.

My partner has hyperphantasia, and similarly she wouldn’t want it “fixed”.

I’d love to hear the work you’re involved in!

This makes me think that for simple things, we need to anti-prompt — tell the model to not overthink things.

He would pay to have his videos as ads on other people’s videos.

E.g., I’d go to watch a video of SmarterEveryDay, and Tai Lopez would show up as an ad, telling me all about his Lamborghinis and bookshelves of books he’d never read. And people would just watch the full ad, even after they could’ve skipped (5s).

That was an interesting era of YouTube, for sure.

The sea urchin (#5) is my favorite, I think. The tardigrade is incredible too.

Both made me awe at the marvel of biological machinery. Such complicated and sophisticated behavior is occurring all around us, even in the tiniest of organisms.

Beautiful.

Agreed.

This is how the R language allows you to explicitly scope functions from packages. I honestly love that syntax.

E.g., `dplyr::filter`, `limma::voom`

For what it’s worth, in the iOS App Store, you can scroll down to look at the “In-App Purchases” drop down menu, and all of the prices are shown there.

This is true for every app with IAP. It’s how I typically decide if I’ll download an app with IAP.

Incredible list. Thanks for sharing.

VSCode is my go to for large text file interaction on macOS.

TextEdit may be worth looking into as well? Haven’t tested it for large files before.

Did you mean the Search Engine podcast..?

I don’t imagine you meant to, but this triggers my LLM sense because of the way Claude Sonnet 4 responds to any critique.

Do you have any go-to examples of “facts” that were disproven in this field?

The recent betas have been extremely performant for me, especially today’s. Might be worth checking out again.

Good to know, I’ll try to just use this a bit more then. I always opt for one of the above modes, with varying degrees of success.

Not sure if you tend to edit your posts, but it could be worth clarifying.

Btw — my colleagues and I all love your posts. I’ll quit fanboying now lol.

Is this the “Web Search”, “Deep Research”, or “Agent Mode” feature of ChatGPT?

Navigating their feature set is… fun.

I have cut out the vast majority of news and social media since the beginning of 2025.

My mental health is noticeably better, and I would readily attribute this to not being tuned in to the ongoings of the globe that I have zero control over. Instead, I have been more focused on my local life and community.

But I regularly feel guilty about not being keyed in to the flirtations with/forays into authoritarianism in my country.

When I was getting my Ph.D., my advisor jokingly told me that his lab has three 8 hour shifts per day, and I could pick two to work.

This was never literally practiced.

But excessive hours were the norm. And I loved it. It helped me launch into a successful career.

But it hurt my relationship with my partner (now wife), and it burned me out.

I miss those days, but I don’t miss what they did to my health.