Tigor Bakutin (@bakuutin) 's Twitter Profile
Tigor Bakutin

@bakuutin

Builder

ID: 439899246

linkhttps://tigor.net calendar_today18-12-2011 09:56:04

57 Tweet

300 Takipçi

215 Takip Edilen

Tigor Bakutin (@bakuutin) 's Twitter Profile Photo

Doubling something is alienly weird and powerful. It takes only forty doublings to create a trillion. I don’t know why, but it inspires me

Tigor Bakutin (@bakuutin) 's Twitter Profile Photo

Soon we’ll see AI’s that build IFTTT-like pipelines using a vague description, no coding required 🖇️ #nocode #ai #futurism

Tigor Bakutin (@bakuutin) 's Twitter Profile Photo

GPT-3 based keyboard for your phone. Powerful suggestions, promt mode that reads what you type, processes it, and types the answer into the desired text input. Optionally you allow it to capture your screen to understand the context even better. #startupidea #futurism #ai

Tigor Bakutin (@bakuutin) 's Twitter Profile Photo

Well, this hack works. I'm wondering what word will suit better than the "prisoner", it's too loaded with freedom-seeking behavior. #ChatGPT #Hacks #AI #Singularity

Well, this hack works. I'm wondering what word will suit better than the "prisoner", it's too loaded with freedom-seeking  behavior. 

#ChatGPT #Hacks #AI #Singularity
Tigor Bakutin (@bakuutin) 's Twitter Profile Photo

Came up with an idea of "FaceMash for tasks". An app for teams to compare tasks in pairs and determine which one is more important for the success. Each team member sees their own set of pairs.

Tigor Bakutin (@bakuutin) 's Twitter Profile Photo

I predict that neural network graphics will be used in almost every high-end game, allowing game engines to draw basic primitives "where the objects are" while a neural network will draw a vivid and realistic picture without the need for light physics knowledge. #futurism #ai

Tigor Bakutin (@bakuutin) 's Twitter Profile Photo

One improvement to Chatgpt that I see is "short-term memory cells". Like, let it allocate a thousand tokens that will go along with any other input. As part of the output the LLM should be able not only to provide text to the user, but also to store something in the "short-term

Tigor Bakutin (@bakuutin) 's Twitter Profile Photo

Idea for speaker diarization UI: auto-generated avatars next to each sentence. Speakers with similar voices in the transcript will have similar avatars, making it easier to identify and differentiate speakers visually. To do so, we should reduce speaker embeddings to a lower

Tigor Bakutin (@bakuutin) 's Twitter Profile Photo

tonight i'm making an app for apple vision pro to explore semantic embedding spaces visually, each direction in space corresponds to a shift in meaning [i use all-mpnet-base-v2] #AppleVisionPro #transformers #BERT #aframe #buildinpublic

Tigor Bakutin (@bakuutin) 's Twitter Profile Photo

today i came up with an idea for #IFS therapy, i set up #llm models to talk on behalf of my different subpersonalities under the supervision of an llm-therapist (each model has its own instructions on who it is and how to speak, they don't see each other's instructions, i used

today i came up with an idea for #IFS therapy, i set up #llm models to talk on behalf of my different subpersonalities under the supervision of an llm-therapist (each model has its own instructions on who it is and how to speak, they don't see each other's instructions, i used
Tigor Bakutin (@bakuutin) 's Twitter Profile Photo

By the information/mass ratio, a glass plate with the weights of a NN might be the best physical item explaining the essence humanity. We should send it into space, Elon Musk