Looking at people re-run the debugger on errors tells me what we might be sitting on some interesting data 😉
AI
-

LLM Retraining vs Vector Stores: Expert Panel Discussion
By
–
So what’s better? Retraining your LLM or using vector stores to give your AI the context it needs to answer questions?
— Replit ⠕ (@Replit) 29 mai 2023
Our AI panel with Claypot AI co-founder Chip Huyen, in conversation with Replit’s Amjad Masad & Michele Catasta, covers this & more! https://t.co/sLzgzpxyKJ pic.twitter.com/7a8kxqbR9wSo what’s better? Retraining your LLM or using vector stores to give your AI the context it needs to answer questions? Our AI panel with Claypot AI co-founder Chip Huyen, in conversation with Replit’s Amjad Masad & Michele Catasta, covers this & more! https://
youtube.com/watch?v=zXX0I6
dOPYk
… -
LearnOpenCV Partners with RunPod for Free GPU Cloud Services
By
–
We're excited to announce our new cloud partnership with @RunPod
! All students for our https://
indiegogo.com/projects/maste
ring-ai-art-generation
… will now get up to 75 hours of free GPU cloud services! #AI #cloudGPU #partnership #aiartcommunity -

Protecting Open Source AI for Economic Growth and Innovation
By
–
It is essential that we protect & encourage open source generative AI as a means to drive:
Economic growth /GDP
Job creation as new startups scale Innovation & remaining competitive Reduce the risk of data oligopoly EU startups use open source tools from the US & elsewhere & -
Open Source AI Startups Drive Innovation and Job Creation
By
–
There should be a pan industry approach with startups involved. Open source generative AI startups will bring innovation, job creation, GDP growth, & reduce the risk of data oligopolies being created @vonderleyen @pierrepinna @Xbond49
-
GPT-4 Performance Improvements Come at Higher Cost
By
–
Works much better with GPT-4, which is even more expensive
-
New Tech Distribution Experiment: Learning Fast Through Velocity
By
–
What are these experiments? We're trying this weird new combo of tech + distribution. @thetanmay and I sat one day and asked ourselves what happens if we combined a high shipping velocity in a new space with pre-existing distribution and basically just learn fast. We'll see!
-
Applied AI: Team’s evolving understanding of generative AI
By
–
Our team's (and my own) understanding of gen AI constantly evolves with everything new we ship. We stay away from core ML since its not our core competency (except for maybe finetuning/adapters). Applied AI is new + uncharted + exciting territory and a good place to gain exp 🙂
-
Chrome Extension AI Tool With Advanced Debugger Capabilities
By
–
Compared to smol, we're much more tailored toward chrome extensions atm (on a side note, smol is a great repo, defo do check it out. Showed us what's possible!). We also have a much better debugger in action.
There is still a 20-30% failure rate/generation which will get better! -
Shipping Fast: Building Gen AI Muscle Memory Through Experimentation
By
–
I do want to mention that we have no plans to turn this into a business. Gen AI is new and vast, and the only way to build up muscle memory is to ship lots + fast. This is our 3rd experiment, and we just wanted to show the world what's possible in a simple to use interface.