The right question, and one too few enterprises are asking. Thanks @realmtbman and @palebluenexus for having our co-founder @nikolaborisof on.
Full episode: https://t.co/AZMuaTllzq
Deep Infra serves the largest selection of open models - Kimi K2.5, GLM-5, Minimax M2.5, NVIDIA Nemotron, and more - running on NVIDIA Blackwell.
Competitive pricing. Cached pricing to cut costs further. More inference for your budget.
https://t.co/Yq9lL2BHYs
What a better way to spend the weekend than reverse-engineering your childhood favorite games!
After being inspired by @skirano and @ammaar and their efforts. I had to give it a try. AND, not just reverse engineering, but also improving by ... adding some AI.
This is an old Eastern European game called VLAK. Snake-like game, but now ... you can call an LLM to create a custom VLAK level, OR, change the gameplay dynamics of VLAK.
What are you building?
#ai #games #reverseengineering #llm
Linear Digressions by @multiarmbandit is back! <-- my favorite AI / ML podcast! I was heartbroken when it ended in 2020.
check it out here: https://t.co/Zawk0Rrh28 or here on Spotify: https://t.co/Vjlx3vJTUe
I absolutely LOVED their episodes, even on more non-traditional topics, like The Coastline Paradox or Traffic Metering Algorithms.
Starting strong by episode on "Chasing Away Repetitive LLM Responses with Verbalized Sampling" .
#AI #podcast #lineardigressions #Spotify
Sometimes you feel compelled to do things. At the University of Michigan, I was drawn to artificial intelligence, what could be more appealing than studying what thinking was? and how could we make something that really thought? So I did my PhD in Computer Science focusing on AI, when everyone else told me the field was dead.
Soon after, I met the amazing people at Google who I immediately knew would be changing the world, and felt compelled to join them, making the world's information accessible to everyone.
When language models started talking, I felt compelled to figure out how they could think beforehand, and was drawn to work with Eric and Noah on Quiet Star.
Now, I have that familiar feeling again, of a calling, to work on a humanistic AI, one that understands and values people - alongside amazing friends @ericzelikman, @YuchenHe07, @noahdgoodman, @AndiPenguin and many other amazing humans! I'm excited to announce our company humans& that will work on this humanistic AI.
Why? Not because I miss the sleepless nights and pressure of a startup :) The world is changing, and rapidly, and this is a challenging time for people when really no one can predict where the future goes and almost everyone is somewhat anxious as a result. So I think it's worthwhile to think about why that is, and what might be done. I think training an AI to understand us, and value us is part of the answer.
I have finally found something more appealing than studying what thinking is - to make the thinking of AIs great for people. I hope you think this is a worthwhile mission, and I hope you will support us - because no one changes the world alone, and we'll need your help to do it.
MEMORY! Memory! memory! ...memory everywhere!
OpenAI and Google released their cookbooks. But if you are looking for a good overview, I liked https://t.co/KkKQo2K9AZ by Leonie Monigatti . <-- this is a great intro at least as long as text-based memory goes.
BUT! sooner or later, as the history grows, you will NEED to go for non-textual representations. A great approach of learned memory is here : https://t.co/0w05G87tZ2 by Jessy Lin .
I have seen in more than once, this is your matrix-factorization, replacing KNN. But in short-term, explainable text memory is super usefull.
Anyways! Love this next step in the LLM world towards more personalization and life-long agents.
#AI #LLMs #AIAgents #Memory #MachineLearning #Personalization #ContinualLearning #ArtificialIntelligence
For the past few months, we’ve been building something I’ve always wished existed:
Introducing Wonder - the first AI-native design tool on an infinite canvas with taste and understanding of your designs.
~37 people only, on Earth can train an llm from scratch. why? the crucial hacks live in heads, not papers. this is the reason many big labs lost the ability to train a GOOD model.
I am trying to change that and democratize LLM training. packed every trick I know into a public cheatsheet. steal it, improve it, share it.
🙏 please add your hacks / knowledge:
👉 https://t.co/XCgYteHtXC
#llmtraining #ai #foundationmodels
In October 2006 we started working on live video for the internet. That became Twitch. More than 16 years later, I'm now a father and ready to move to my next phase of life. I wrote a blog post, but the short version is: thank you so much to everyone who built this with me.
After an amazing 2.5 years my time at Twitch is up. This was an amazing ride and so far the job I enjoyed the most! I loved all the stuff I learned building Proactive Detection and making Machine Learning a tool to make Twitch safer and more inclusive. …https://t.co/sflPAej0GO
Twitch is hiring an Applied Science Manager! Know anyone who might be interested? Reach out! Or apply directly! This is an amazing team with stelar scientists and engineers! #hiring#science#engineers#team#twitch#machinelearningjobs https://t.co/oQMPKSk1Vt
Amazing work on novel view synthesis, material decomposition and relighting. https://t.co/hbJ4EVk7I8 . I love to see combinations of my old passion, computer graphics and geometry and my new passion, machine learning. #machinelearning#rendering#computer…https://t.co/pAfya5zIJO
Twitch is hiring! Join us! Twitch is looking for a talented Manager who will lead Machine Learning Engineers and Applied Scientists building recommender systems. If you are interested, let me know!
#machinelearning#hiring#recomm…https://t.co/1laC5DWduf https://t.co/X6GLcD0gQw
Infinite Nature. What an interesting application of neural networks, Perpetual View Generation. Just fly into an image and generate renderings from a new viewpoint. A paper from Google Research is available here: https://t.co/q262JlRowe and you can try th…https://t.co/dGcbGKo7oR
If your team depends on manual labeling you might want to reduce the number of mislabeled examples. Confident learning does exactly this https://t.co/1yRLBLiNBs . It helps with identifying mislabeled examples by estimating the joint distribution of corr…https://t.co/7OBLkdPHEv
Ever wanted to learn Transformers or just understand them better? Or just wanted to know exactly how self-attention works? This blog post: really nicely illustrates how Transformers work. It does a great job going top-down and ex…https://t.co/Net3ujdFq1 https://t.co/y5Yrk4fh2D
What a great collection of Machine Learning introductory resources : https://t.co/mtf29Y52Jr . These are collected by Simon Prince who wrote one of my favorite Computer Vision books, Computer Vision: Models, Learning, and Inference. Especially the first p…https://t.co/IXn1iVuxv4