Nothing_but_the_truth

[Nothing_but_the_truth @KabalanFreeman]

@OfficialLoganK 2.5 pro is beast , only downside it refuses to do long automated tasks like implementing a route wit tests etc to a codebase, always needs like "procede" commands to continue , copied a fully working implement task prompt that worked with sonnet

[Nothing_but_the_truth @KabalanFreeman]

@OfficialLoganK can you add propper seperation of thought and output last time i checked it was all in one output

[Nothing_but_the_truth @KabalanFreeman]

@abacaj they have PhD "Superagents" somewhere in the future maybe :)

[Nothing_but_the_truth @KabalanFreeman]

@jobergum What I don't understand is that there are still companies who publish embedding models and put research into it. My idea is to just index your data properly and feed it into your LLM. Saw some crazy over-engineered pipelines with fancy chunking, LangChain...

[Nothing_but_the_truth @KabalanFreeman]

@TheRealAdamG @ItsThomasHTX @beyondthisjourn Realtime api is cool but, after testing it for couple minutes i burned 10 bucks, cant be that expensive for audio only, is the live video ready and you blocking it on purpose ?

[Nothing_but_the_truth @KabalanFreeman]

@elder_plinius you posted a while a ago a list of some matrix stuff, the thing i remembered was look for signs that you are in the matrix, after starting watching for signs there are actually a lot of stuff that I notice from a day to day basis, not yet in the exploit phase though

[Nothing_but_the_truth @KabalanFreeman]

@abrakjamson

[Nothing_but_the_truth @KabalanFreeman]

@abacaj then you have the bottom of the bottom feeders who still develope embeding models

[Nothing_but_the_truth @KabalanFreeman]

@alexalbert__ feature request:
as a user i can define tags in my chat message this tags can be used to upload data which is placed in the tags like in the antrophic workbench

[Nothing_but_the_truth @KabalanFreeman]

@alexalbert__ Context caching !

[Nothing_but_the_truth @KabalanFreeman]

@elder_plinius possible, first jailbreak the mode, let the model do hypnosis on you, plants a prompt injetion into your subconsciousnes , profit ?

[Nothing_but_the_truth @KabalanFreeman]

@literallydenis Gemini is quite good at producing very detailed and accurate textual representations.
Give it a prompt to return what you want from the video (video Scene, content). Then another prompt with the output and the specific instruction.
(chat does not work so well)

[Nothing_but_the_truth @KabalanFreeman]

@kyutai_labs i think the model does the answering before he finish taking because it already has all the content to answer , no need to finish listining to hi, :)

[Nothing_but_the_truth @KabalanFreeman]

@skirano pls post repo bro

[Nothing_but_the_truth @KabalanFreeman]

@teknium it is also exceptionally good ad writing meta prompts and followig them wihtout getting confused by all the instructions

[Nothing_but_the_truth @KabalanFreeman]

@skirano i asked Claude 3.5 to create some object it can choose in svg

[Nothing_but_the_truth @KabalanFreeman]

@AnthropicAI what is model name in VertexAI? / is having claude sonet enough or do I need to wait for another model in the model garden ?

[Nothing_but_the_truth @KabalanFreeman]

@abacaj why dont use miro board instead ?

[Nothing_but_the_truth @KabalanFreeman]

@hrishioa not sure if understood it correctly, you have a stream response where after each block or so you have a validator whcih stops the stream and request a new reponse ?