Joshua Shew
@joshuashew.bsky.social
If your brain isn’t tired by the end of the day, you’re doing it wrong
he/him
2025 theme: Year of Foundations
he/him
2025 theme: Year of Foundations
I find the way Emmett Shear is credited as the former interim CEO of open AI kind of funny. The second one listed too!
November 9, 2025 at 9:27 PM
I find the way Emmett Shear is credited as the former interim CEO of open AI kind of funny. The second one listed too!
I’m finally reading it!
November 9, 2025 at 9:27 PM
I’m finally reading it!
I really needed this new "extra usage" feature in Claude. Thank you @anthropic.com!
November 7, 2025 at 10:58 PM
I really needed this new "extra usage" feature in Claude. Thank you @anthropic.com!
Kinda aggressively stated, but there’s an interesting feature request in here: filter posts by intractability
November 1, 2025 at 10:49 AM
Kinda aggressively stated, but there’s an interesting feature request in here: filter posts by intractability
You're ahead of me!
October 31, 2025 at 1:55 PM
You're ahead of me!
Just discovered a-Shell today and it is amazing. I'm putting together a quick utility to make it easier to share videos with my friends in our group chats, and being able to put all the complexity in something not Shortcuts feels so good.
October 25, 2025 at 5:14 AM
Just discovered a-Shell today and it is amazing. I'm putting together a quick utility to make it easier to share videos with my friends in our group chats, and being able to put all the complexity in something not Shortcuts feels so good.
It happened again :/
Claude safety filter is sensitive when it comes to discussions about AI, I guess?
claude.ai/share/1b4d6f...
Claude safety filter is sensitive when it comes to discussions about AI, I guess?
claude.ai/share/1b4d6f...
October 25, 2025 at 4:20 AM
It happened again :/
Claude safety filter is sensitive when it comes to discussions about AI, I guess?
claude.ai/share/1b4d6f...
Claude safety filter is sensitive when it comes to discussions about AI, I guess?
claude.ai/share/1b4d6f...
Looks great. It’s a little finicky trying to post stuff from this interface, but I’ll try using is as my space to read posts for now.
October 24, 2025 at 6:49 PM
Looks great. It’s a little finicky trying to post stuff from this interface, but I’ll try using is as my space to read posts for now.
Recently I've been running into issues where Claude's output includes lists that should have newlines in them but don't. I haven't tried this in the API enough to see if this is a rendering issue or an issue with the output itself...
October 24, 2025 at 6:35 PM
Recently I've been running into issues where Claude's output includes lists that should have newlines in them but don't. I haven't tried this in the API enough to see if this is a rendering issue or an issue with the output itself...
It's beautiful too. Underrated part of UI in modern web apps.
October 24, 2025 at 6:30 PM
It's beautiful too. Underrated part of UI in modern web apps.
This was fun!
In my view, 80+ is indicative of a lack of curiosity with one of the most interesting and accessible technologies of our time.
www.aipuritytest.org
In my view, 80+ is indicative of a lack of curiosity with one of the most interesting and accessible technologies of our time.
www.aipuritytest.org
October 24, 2025 at 4:24 PM
This was fun!
In my view, 80+ is indicative of a lack of curiosity with one of the most interesting and accessible technologies of our time.
www.aipuritytest.org
In my view, 80+ is indicative of a lack of curiosity with one of the most interesting and accessible technologies of our time.
www.aipuritytest.org
I used to use "<phrase> - <phrase>" a lot in my text messages, but I switched to "<phrase>—<phrase>" once my fingers learned how to type it.
Now that LLMs are avoiding em dashes, does active use of em dashes suggest human text?
Now that LLMs are avoiding em dashes, does active use of em dashes suggest human text?
![Claude starts a response with "You're misunderstanding - and it's [...]"](https://cdn.bsky.app/img/feed_thumbnail/plain/did:plc:sfmur4r4wzriaupbgmz7sshy/bafkreiddagcgn3prq5hmz27dpfyqrxaayug5kzbtpv5wzwlldjay2gtlje@jpeg)
October 18, 2025 at 11:38 PM
I used to use "<phrase> - <phrase>" a lot in my text messages, but I switched to "<phrase>—<phrase>" once my fingers learned how to type it.
Now that LLMs are avoiding em dashes, does active use of em dashes suggest human text?
Now that LLMs are avoiding em dashes, does active use of em dashes suggest human text?
A story in two parts
October 18, 2025 at 11:06 PM
A story in two parts
My problem with running agents in parallel (aside from it being expensive) is how much context switching it involves. My brain is spent after a little bit of it. There must be a better way…
![Quoting except from article: I was skeptical of [running multiple agents in parallel] at first but I’ve started [doing it myself now] and it’s surprisingly effective, if *mentally exhausting*.
(emphasis mine)](https://cdn.bsky.app/img/feed_thumbnail/plain/did:plc:sfmur4r4wzriaupbgmz7sshy/bafkreifejljdxgneg44dzipl4q2apgscgmosiw46ri372sh4fog3ntue7a@jpeg)
October 18, 2025 at 9:08 PM
My problem with running agents in parallel (aside from it being expensive) is how much context switching it involves. My brain is spent after a little bit of it. There must be a better way…
The (presumably Haiku-generated) thought process summary doesn’t line up with the actual contents of the thought process…
Not sure why this happens. My query was pretty clear from our preceding conversation, so maybe this is because Haiku only has access to the most recent user message.
Not sure why this happens. My query was pretty clear from our preceding conversation, so maybe this is because Haiku only has access to the most recent user message.
October 18, 2025 at 8:57 PM
The (presumably Haiku-generated) thought process summary doesn’t line up with the actual contents of the thought process…
Not sure why this happens. My query was pretty clear from our preceding conversation, so maybe this is because Haiku only has access to the most recent user message.
Not sure why this happens. My query was pretty clear from our preceding conversation, so maybe this is because Haiku only has access to the most recent user message.
Claude just made my Alfred clipboard history a whole lot more useful for me by making it really easy to merge entries in my history.
I needed the tool for a task, took 5 min for Claude to make it, and then I was back on task.
I've already used it a couple of times since then.
I needed the tool for a task, took 5 min for Claude to make it, and then I was back on task.
I've already used it a couple of times since then.
October 16, 2025 at 3:50 AM
Claude just made my Alfred clipboard history a whole lot more useful for me by making it really easy to merge entries in my history.
I needed the tool for a task, took 5 min for Claude to make it, and then I was back on task.
I've already used it a couple of times since then.
I needed the tool for a task, took 5 min for Claude to make it, and then I was back on task.
I've already used it a couple of times since then.
Looks like I need to stop driving...
Relatedly, the energy:value ratio on chatbots is crazy good. I wouldn't have guessed.
Relatedly, the energy:value ratio on chatbots is crazy good. I wouldn't have guessed.
October 14, 2025 at 10:31 PM
Looks like I need to stop driving...
Relatedly, the energy:value ratio on chatbots is crazy good. I wouldn't have guessed.
Relatedly, the energy:value ratio on chatbots is crazy good. I wouldn't have guessed.
Tragic. Another reason to switch to a BYOK chatbot interface...
October 14, 2025 at 10:13 PM
Tragic. Another reason to switch to a BYOK chatbot interface...
I did take this as an opportunity to put together a super simple Alfred workflow to make it easy to wrap something in an XML tag. I'm not sure if this makes measureable improvements, but I hear that Claude likes XML, so I'll give it XML when I'm throwing a lot in the context window.
October 14, 2025 at 3:49 AM
I did take this as an opportunity to put together a super simple Alfred workflow to make it easy to wrap something in an XML tag. I'm not sure if this makes measureable improvements, but I hear that Claude likes XML, so I'll give it XML when I'm throwing a lot in the context window.
Looks like this was a quick project put together a year ago, but I appreciate the perspective on branching and merging conversations.
October 14, 2025 at 3:09 AM
Looks like this was a quick project put together a year ago, but I appreciate the perspective on branching and merging conversations.
Ran into this issue when I was talking about how Void compares to various measures of consciousness last week :/
October 10, 2025 at 5:43 PM
Ran into this issue when I was talking about how Void compares to various measures of consciousness last week :/
Here's another idea, this time from Claude
October 6, 2025 at 12:38 AM
Here's another idea, this time from Claude
Open tabs on iOS Safari has been my "read later" solution and I'm finally down to just 4 open tabs!!
October 5, 2025 at 3:55 AM
Open tabs on iOS Safari has been my "read later" solution and I'm finally down to just 4 open tabs!!