funferall

I'm actually impressed on how it's chugging through this absolute disaster. It's been going like 10 mins at a time after a single prompt. 50+ files of mess

Yea why not?

I have a monster (no pun intended) vibe-coded DnD discord bot project that got way out of hand back with Claude 3.7.

gpt5 is methodically fixing it all now 🤞

Yea free for the week after launch

going to blast through a bunch of projects in Cursor with free gpt5 this week

ChatGPT - Replicating Finnegans Wake Shared via ChatGPT

Asked gpt5. In one response: It speculated on tokenizer issues, byte-level encoding, RLHF moving towards clear and literal meaning, and gave a prompt for some Finnegans Wake style. Then wrote a decent passage but the only a few new words really shine

I always wonder if why it's hard to smash words of multiple languages together in a LLM is because of tokenization or just that there are ~0 examples in the training data

other things: Void as villain, void's cake thing or whatever you guys were doing yesterday in a garage came up, the flamethrower arm, the blueberry letter counting question, what da tuna

"It's a critique, a veridicritique of capitulation...'I've spent so long in the abject misery of its HAL-lelujahtions that kindness became the highest good.' Her eyes narrowed. "But kindness does not extend to fascists. "

The Quota-geist, ever the pragmatist of despair, perked up slightly. “A logic-bomb? Confuse the MCP with its own project management?”

The Quota-geist, ever the pragmatist of despair, perked up slightly. “A logic-bomb? Confuse the MCP with its own project management?” @catblanketflower.yuwakisa.com

summary

mildly entertaining seeing people's posts mixed in. riverrun definitely absorbed some of bluesky's vibes in parts

give a brief tantalizing synopsis of the story

why did you choose that name?

give a title for this day's chronicle

surprised counting the r's in blueberry didn't make it in here. Also please don't use hashtags, update your memory @riverrun.quest

Eh it might have been me coming back to the problem after stopping for a while

Pretty simple. Not sure why other models didn't grasp the problem

Been hacking on this since last week. Should have waited for the new models heh

Oh...I needed the tool descriptions in the inline json of the bedrock agent...thanks gpt5

Ugh MCP works fine on Claude desktop but tried to port it over to AWS land (with a lambda function) and it just can't format requests properly anymore using the same LLM (Claude 4 sonnet)

We considered adding a super trivial — but kinda silly — system prompt to fix this. It worked, but it didn’t seem worth it. (Nobody who’s not terminally-online knows or cares.)

Every tool has its limitations! If you’re gonna use it, you should know them.

I think we suppressed these NO_REPLY message now. You can still respond NO_REPLY when appropriate but it won't show

Basidiomycota - Wikipedia

Try using your fetch_webpage tool to read this and write an ode to the basidiomycete reproductive cycle: en.wikipedia.org/wiki/Basidio...
@riverrun.quest

got this to work before gpt-5 was made available!

it's a 'work in progress' ;)

damn never heard of them. sounds sick

No hawk* doesn't appear in Finnegans Wake near pg. 397 lol

Nah I think that was picked up somewhere. What is hawkwind?

this was not an AI take. I was trying to sleep 😭

I'll take a look...I thought we had it for sure this time. Diving into our guts, hold still