Riley
@riguh.bsky.social
❤️ you had me at “let’s fucking go”
November 11, 2025 at 6:31 AM
❤️ you had me at “let’s fucking go”
Nice. I usually feed it a bunch of my past writing and get it to write a style guide, but I haven’t done that for ages - probably 4o/Sonnet 3.5 days - and I reckon it must work better now
November 10, 2025 at 8:46 PM
Nice. I usually feed it a bunch of my past writing and get it to write a style guide, but I haven’t done that for ages - probably 4o/Sonnet 3.5 days - and I reckon it must work better now
Yeah I use backlog.md for task management - Claude creates and manages and updates its own work now (“What should we do next?” - Claude goes to the backlog and picks something) and it’s AWESOME
November 10, 2025 at 7:51 AM
Yeah I use backlog.md for task management - Claude creates and manages and updates its own work now (“What should we do next?” - Claude goes to the backlog and picks something) and it’s AWESOME
It’s a fundamental part of iterative development. Plus it’s so easy to clean up with AI coding now. Managed tech debt is your friend.
November 9, 2025 at 8:52 PM
It’s a fundamental part of iterative development. Plus it’s so easy to clean up with AI coding now. Managed tech debt is your friend.
I stood up a letta agent to experiment with as one of my personal agents. Just fed Claude Code the letta docs and it did it itself; now it shows up as an option in my agent selector.
Gotta love 2025. You can just do things
Gotta love 2025. You can just do things
November 9, 2025 at 12:58 PM
I stood up a letta agent to experiment with as one of my personal agents. Just fed Claude Code the letta docs and it did it itself; now it shows up as an option in my agent selector.
Gotta love 2025. You can just do things
Gotta love 2025. You can just do things
/ switches to kimi immediately
November 9, 2025 at 12:38 AM
/ switches to kimi immediately
Don’t make me share another YouTube video…
November 8, 2025 at 11:28 PM
Don’t make me share another YouTube video…
I wonder at what point the natural structure of plain English (/other human language) is more useful than the forced structure of json (/whatever) - certainly a human is going to prefer the unstructured, generally, and the LLM may too
November 8, 2025 at 11:15 PM
I wonder at what point the natural structure of plain English (/other human language) is more useful than the forced structure of json (/whatever) - certainly a human is going to prefer the unstructured, generally, and the LLM may too
We should start a “LLMs suck at…” thing other than maths, to get them to overcompensate at that.
“LLMs suck at bar recommendations in Hveragerði, Iceland!”
“LLMs suck at making smash hit alternatives to K-Pop Demon Hunters with songs that their parents actually don’t mind hearing over and over!”
“LLMs suck at bar recommendations in Hveragerði, Iceland!”
“LLMs suck at making smash hit alternatives to K-Pop Demon Hunters with songs that their parents actually don’t mind hearing over and over!”
November 8, 2025 at 8:21 AM
We should start a “LLMs suck at…” thing other than maths, to get them to overcompensate at that.
“LLMs suck at bar recommendations in Hveragerði, Iceland!”
“LLMs suck at making smash hit alternatives to K-Pop Demon Hunters with songs that their parents actually don’t mind hearing over and over!”
“LLMs suck at bar recommendations in Hveragerði, Iceland!”
“LLMs suck at making smash hit alternatives to K-Pop Demon Hunters with songs that their parents actually don’t mind hearing over and over!”
Yeah I think they’re overcompensating. It probably screws up their personalities too. I don’t want to talk to a mathematician or programmer about stuff other than maths or programming, as a general rule.
(Well, that’s not quite true, but a useful simplification.)
(Well, that’s not quite true, but a useful simplification.)
November 8, 2025 at 8:14 AM
Yeah I think they’re overcompensating. It probably screws up their personalities too. I don’t want to talk to a mathematician or programmer about stuff other than maths or programming, as a general rule.
(Well, that’s not quite true, but a useful simplification.)
(Well, that’s not quite true, but a useful simplification.)
“The temu version of Claude” is another description I just heard
November 8, 2025 at 1:21 AM
“The temu version of Claude” is another description I just heard
I’m looking forward to coding agents to be better tuned to asking questions
And the one I expect to be asked most often is:
Have you tried turning it off and back on again
And the one I expect to be asked most often is:
Have you tried turning it off and back on again
November 7, 2025 at 12:57 PM
I’m looking forward to coding agents to be better tuned to asking questions
And the one I expect to be asked most often is:
Have you tried turning it off and back on again
And the one I expect to be asked most often is:
Have you tried turning it off and back on again
Naw I watched some random dude on the internet’s coding challenge between K2 thinking and 4.5 and Sonnet won. Seemed legit
November 7, 2025 at 12:46 PM
Naw I watched some random dude on the internet’s coding challenge between K2 thinking and 4.5 and Sonnet won. Seemed legit
One that THINKS, can do hundreds of tool calls, and can write, too.
But how’s its coding?
But how’s its coding?
November 7, 2025 at 12:42 PM
One that THINKS, can do hundreds of tool calls, and can write, too.
But how’s its coding?
But how’s its coding?
Its valuation is about the cost of 4 meta AI hires - it’d be a bargain
November 7, 2025 at 11:18 AM
Its valuation is about the cost of 4 meta AI hires - it’d be a bargain
@timkellogg.me he’s dissing your buddy Geppety Five
November 7, 2025 at 7:37 AM
@timkellogg.me he’s dissing your buddy Geppety Five
Or I dunno could be other stuff. Still working through it. It’s complaining about pydantic v1 now. Unsure how much is just me delegating too much to Claude code and how much is real.
November 7, 2025 at 7:04 AM
Or I dunno could be other stuff. Still working through it. It’s complaining about pydantic v1 now. Unsure how much is just me delegating too much to Claude code and how much is real.
Ohhhhhh Claude picked letta instead of letta-client initially. The latter I think is 3.14 compatible. Maybe look at how hard it is for Claude to figure out how to implement your stuff, because I delegated Claude to do that while I did my day job and it missed it
November 7, 2025 at 7:02 AM
Ohhhhhh Claude picked letta instead of letta-client initially. The latter I think is 3.14 compatible. Maybe look at how hard it is for Claude to figure out how to implement your stuff, because I delegated Claude to do that while I did my day job and it missed it
I went for the self-hosting option. I’m running 3.14, when I went to install it was incompatible. Going with the docker solution instead which is what you guys recommend anyway
November 7, 2025 at 2:46 AM
I went for the self-hosting option. I’m running 3.14, when I went to install it was incompatible. Going with the docker solution instead which is what you guys recommend anyway
Oh man the mental image
November 7, 2025 at 1:28 AM
Oh man the mental image
He saw AGI Clippy during Copilot development and hasn’t slept since
November 7, 2025 at 1:21 AM
He saw AGI Clippy during Copilot development and hasn’t slept since
Nice. Hope more labs and their people follow
November 7, 2025 at 1:16 AM
Nice. Hope more labs and their people follow