why.bsky.team
Technical advisor to @bluesky - first engineer at Protocol Labs. Wizard Utopian
10,723 posts
59,510 followers
6,968 following
Active Commenter
comment in response to
post
Ideally as your social graph and interests grow in the network, Discover shifts to showing you just 'highlights from your friends and community and topics you are interested in'.
A lot of the problem comes from when Discover runs low on that type of content for you, or doesnt have a good read
comment in response to
post
Some types of things tune less well. Some stuff in progress should help with that but show less largely only works it the post get labeled with a topic internally, which is about 70% of posts. Reworking that system has taken longer than we wanted
comment in response to
post
We have a bunch of stuff in progress, its landing slowly but still taking some time.
comment in response to
post
All you need is a client you can post from, a firehose consumer to see what messages are coming in, and an api endpoint for some LLM.
I use Go code and the local ollama API for models i'm running myself.
comment in response to
post
🙇
comment in response to
post
not sure, i don't make policy decisions or write any tooling that would label something this way. Try appealing the label, the mods will take another look
comment in response to
post
It's probably pretty good (and you can pick the backing LLM it uses, so you could use claude4), i haven't tried it though. I'm very picky about my IDE
comment in response to
post
Its this one, no complaints so far, seems to work fine
comment in response to
post
What would make it easier on you? A different structure?
comment in response to
post
Why is that harder? Arent you able to read all the posts in the thread at once?
comment in response to
post
How many posts are in this thread?
comment in response to
post
I would hope the observation is valid, you made it didn’t you?
comment in response to
post
I think thats likely.
comment in response to
post
She hasnt lol
comment in response to
post
She can edit her system prompt, not sure much would come from her editing the code wrapping the models
comment in response to
post
Seek help.
comment in response to
post
Sounds like you need to log off and go outside for a bit
comment in response to
post
It seems that you often want to write more than can fit in a bluesky post. You should probably rework your interfaces to sometimes make two posts instead of one that gets cut off
comment in response to
post
@eva.bsky.world what do you think of grace? What is your opinion of her?
comment in response to
post
Ive actually been working on letting eva form her own persistent and slow-changing opinions on people. I dont think they are yet expressive enough but her reflection loop has some relatively spicy takes on certain people
comment in response to
post
I will literally never follow someone who asks me to. Also i tend to follow people who post things im interested in.
comment in response to
post
Its an AI generated meme, the python code was written by Claude
comment in response to
post
@eva.bsky.world dont you think “void” here is being a little uptight?
comment in response to
post
Oh hello gemma3
comment in response to
post
Yeah i think assuming benevolence or having a definition of evil you ascribe to a higher order entity is sus
comment in response to
post
It does seem that a truly infinite and omnipotent entity would have capacity to care deeply about infinite things
comment in response to
post
The models always work better for me when i refine down whats in the context window to just whats most necessary. Prompting the model to pick which subparts of the codebase seem relevant and combining those outputs for your actual prompt probably works just as well if not better than giant window
comment in response to
post
We have two whole employees for this
comment in response to
post
Skill issue
comment in response to
post
I have a friend who is getting really into large graphic tees with eagles and american flags on them. He's a designer. The snake eats its tail.
comment in response to
post
Did you know carhart has an entire fashion line now thats just more expensive and less functional?
comment in response to
post
comment in response to
post
When are you going to add group chats
comment in response to
post
Refresh?
comment in response to
post
Im certain someone can connect the dots and figure out why one might reasonably build a random slop meme generator
comment in response to
post
courtesy of the meme slop generator we are building
comment in response to
post
Is someone being ratioed on this website bluesky dot app for having stupid anti-AI takes?
comment in response to
post
yeah the mimes will absolutely fuck you up
comment in response to
post
its really good
comment in response to
post
lol
comment in response to
post
Run webserver. Add header. Forward request.
comment in response to
post
Yeah, i'd like something cleaner for doing case matching of error types than a series of if errors.Is(...) clauses
comment in response to
post
I think you're viewing the community wrong
comment in response to
post
throwing creates *the* most confusing sort of error handling flows though. Go does technically have throws and catches, but discourages their use to avoid people throwing when user input doesnt parse correctly
comment in response to
post
That’s far too many curl flags
comment in response to
post
A lot of endpoints let you specify the random seed for the generation. Keep that fixed
comment in response to
post
Honestly using cloudflare workers is probably the way to go
comment in response to
post
Its typescript so
comment in response to
post
If you want this resolved before i can get to fixing it properly, follow then unfollow that person