jonny (good kind)RE: https://hachyderm.io/@jedbrown/116373136965443776
Very good thank you information conglomerates
What the fuck that actually works and the limitations on adding images to quotes are purely client side? I am turning that shit off on our instance
I can definitely see why we would want to put this in charge of employment, public benefits, journalism, and warfare.
fabiosantoscode@jonny don't forget to upvote all these helpful answers, to make sure the LLM learns how to be even more helpful in future training 💜💜💜
Blake C. Stacey@jonny my very eager, um, Jupiter sucked up nuts
Pare
scheme@jonny I found an interesting quirk: The first time I tried this, it responded with something like "I don't have access to this picture", asking me to reupload.
I believe to have this nailed down to the number of hex digits: With 8 digits I get a colorful description of a small dog, but with 7 I get the error.
I believe to have this nailed down to the number of hex digits: With 8 digits I get a colorful description of a small dog, but with 7 I get the error.
@scheme it's extremely good that the space of responses is smooth with respect to what we perceive as smooth and gradual variation in the input
bleeptrack
Elio Campitelli@jonny Does filename matter?
@eliocamp
I was trying from a free account so I ran out of messages after like 5 tries. I think its mostly the pointy brackets and the pattern of appearing like a filename
I was trying from a free account so I ran out of messages after like 5 tries. I think its mostly the pointy brackets and the pattern of appearing like a filename
@jonny I couldn't make it work and then I realised I had written "upolad". Correcting the typo fixed it. So it seems it's pretty sensitive to the file name pattern.
@jonny This works with other formats too. It can even have whole fabricated conversations by "uploading" audio files.
Jed Brown
DaSchnott🌞@jonny
Totaly true.
Mercury/Venus/earth/Jupiter.....
nothing left....
Maybe we will explore some new planets...
But for now...top result.
fraggle@jonny Tried Gemini and it does the same, except the summaries are hallucinated based on past conversations I've had with it. Interestingly when I asked it how it generated a summary without an image, the bot lets me see its internal thinking process where it mentions a "user summary" document that it's referring to, but seemingly has been directed to keep secret because of an "invisible personalization directive"
EDIT: When asked to show my "user summary document" it denies that any such document exists, but it will happily save a copy of it in a Google Keep note if asked
Alex UsedToGoToVenice
wohali
GeorgeK@jonny I've had some free time when the researcher paper was released: https://zskulcsar.github.io/posts/mirage-reasoning/
Mirage reasoning
This morning an email popped up in my inbox, with the subject “The mirage of visual understanding in current frontier models”. This sounds interesting enough, I started to read. Quickly I ended up on the research paper’s page on arxiv: MIRAGE: The Illusion of Visual Understanding it was released on 26th of March.