Profile avatar
joshuashew.bsky.social
I read, think, educate, program, run, and write. I earnestly try to engage in good faith. 2025 theme: Year of Foundations
913 posts 110 followers 224 following
Regular Contributor
Active Commenter
Posts 16 Comments 34

Saving the thought: Gotta practice, “yes this person also agrees with X and they suck. But that does not detract from the specific claims of X”

submitted 14 hours ago • 0 comments

I don't have much to add, except that I found it interesting, and that it is valuable for people to be able to self-experiment like this and share results. There is so much to learn about our bodies.

submitted 4 days ago • 1 comment

Lots of good uses for LLMs, but to pull out two:

submitted 4 days ago • 1 comment

What's the best way to connect two ideas on Bluesky? I can't quote two posts at once, but it seems like such an important way of doing "intellectual work"—to pick out two disconnected posts and bridge them in some way.

submitted 4 days ago • 1 comment

I haven't read the actual paper yet but, interesting "Brain-to-LLM >>> LLM-to-Brain": so engaging your brain first and then enhancing that effort with LLMs is advantageous. Seems to agree with the sentiment I've seen around that LLM enhanced work should be *more* effortful, not less

submitted 4 days ago • 0 comments

It works, and I could be making more "posts" for myself, but I think the barrier to makes posts is still too high? I gotta really simplify the process if it's going to stick.

submitted 4 days ago • 1 comment

Public transit commute is actually so good. I'm really busy since this is my first week doing actual work, and it means I can catch up on small tasks while I'm getting to and from work.

submitted 4 days ago • 1 comment

Neat! Getting specific counts on how many more tokens it tends to go through it helpful too. Using Opus for the orchestration and Sonnet for the search agents is especially interesting. I anticipated there would end up being systems like this, and I wonder if the handoffs can become smoother...

submitted 4 days ago • 1 comment

I'm steadily finding places to make myself more effective in my job with LLM assistance. There's some routine parts of my job that, if I set up the right data pipelines, could be improved a great deal.

submitted 4 days ago • 1 comment

"it’s like a life coach" This, I agree with. I find it easier to use as a life coach than as a therapist, but mostly because I've never really figured out the therapist thing so I'm not sure how that's supposed to work anyway Having a truly outside view on something can be very helpful

submitted 6 days ago • 0 comments

> If your book used normal English grammar [...] the output will sound like normal English But it doesn't, right? If you're just using Markov chains (as you describe), the output is never that good. Transformers got us to coherent English, and that seems important here.

submitted 10 days ago • 1 comment

Parent emailed to say that their child said their new instructor was "the best yet". That instructor was me! Feels good man

submitted 10 days ago • 0 comments

"What every computer science major should know" Great resource. I wish I ran into it earlier, while I was actually a CS major. My job is not software-related, but I will go through this if I decide to move in that direction in the future.

submitted 14 days ago • 0 comments

I was thinking of pain like this. Velleman didn't describe exactly this condition, but my mind made the connection. It's haunting. I'm glad there are people working on it.

submitted 14 days ago • 0 comments

“AGI” : Bluesky tech ppl :: “decentralization” : Bluesky decentralization (idk the correct new term to pick here) ppl

submitted 14 days ago • 0 comments

I listened to the first ~40 min of this and there's a lot of interesting ideas

submitted 14 days ago • 1 comment