Apple has made significant advancements in artificial intelligence by unveiling its Multimodal Large Language Models (MLLMs), including a specific model named MM1, designed to compete with GPT4 and enhance Siri's capabilities. This development positions Apple as a strong competitor in the AI field, alongside other major announcements from tech giants such as Google DeepMind and OpenAI. Additionally, the tech industry has seen a flurry of AI innovations, with Elon Musk's xAI releasing Grok-1, a Mixture-of-Experts (MoE) model with 314 billion parameters, under the Apache 2.0 license, written in JAX and Rust, and completed in October 2023. Grok-1's open-source nature and its benchmarks, despite being twice the size of GPT-3.5, have sparked discussions about its efficiency and potential applications. These developments underscore the rapid pace of AI research and the increasing investments by tech companies in multimodal AI technologies.
AI NEWS: Apple just quietly unveiled MM1, a new LLM that competes with GPT-4 and Gemini. Plus, more developments from Elon Musk/Grok, Google DeepMind, Cognition Devin, India, Bernie Sanders, and Maisa. Here's everything going on in AI right now:
Grok-1 is now the highest quality open-source LLM Grok's declared MMLU score of 73% beats Llama 2 70B’s 68.9% and Mixtral 8x7B’s 70.6%. At 314 billion parameters, xAI’s Grok-1 is significantly larger than today’s leading open-source model. @xai's Grok-1 is a Mixture-of-Experts… https://t.co/Dh6JJ1xKNC
Elon Musk's Grok LLM was just open-sourced! 🔥 It's uncensored, MASSIVE, and completely open-source. Spite is a powerful motivator lol. Here's what you need to know 👇🎥 https://t.co/hiKosjhMKr
Grok-1 feels like a YOLO model that was simply good not great Awesome to see it released
Grok is available on hugging face. You guys are so quick. https://t.co/jz3HHEdCZp
Musk's Grok AI goes open source https://t.co/3d6fX5LG1Q
Elon Musk’s xAI releases Grok-1 architecture, while Apple advances multimodal AI research https://t.co/H4KHKNgSlg
Grok is now available for testing. It's 314B MoE so you won't be able to run it 🙃 It's Amazing that a 314B model gives you a worse performance than open source 13B LLMs https://t.co/UU9AjwD7Cv
Yep, thanks to @elonmusk and xAI team for open-sourcing the base model for Grok. We will fine-tune it for conversational search and optimize the inference, and bring it up for all Pro users! https://t.co/CGn6cIoivT
Grok is now available for testing. It's 314B MoE so you won't be able to run it :upside_down_face: It's Amazing that a 314B model gives you a worse performance than open source 13B LLMs https://t.co/UU9AjwD7Cv
Xai (Twitter) releases does the open release of their Grok AI https://t.co/sNl6aMqEt0
Grok-1 is out on Hugging Face https://t.co/6GBcgnD1ZI
Musk's Grok AI was just released open source in a way that is more open than most other open models (it has open weights) but less than what is needed to reproduce it (there is no information on training data). Won't change much, there are stronger open source models out there. https://t.co/eFouYFVRmN
xAI release Grok-1, Community release Grok-1 on @huggingface 👀 https://t.co/iKgU7MdyIH
The Dawn of Grok-1: A Leap Forward in AI Accessibility: Today marks the open release of Grok-1, a behemoth in the landscape of AI, wielding a staggering 314 billion parameters. This Mixture-of-Experts model, which emerged from the fervent efforts of xAI’s dedicated team,…
Grok-1 314B parameters mixture of experts (MoE) - 8 experts top 2 selection. - The MoE implementation is different from Mixtral 8x7b. Mixtral 8x7b applies softmax over top2 experts, while Grok-1 applies top2 over the softmax of all 8 experts. - written in jax - training code… https://t.co/MlNTlgqO0s
Grok-1 314B 8 experts top2 MoE - 314B parameters mixture of experts (MoE). 8 experts top 2 selection. - The MoE implementation is different from Mixtral 8x7b. Mixtral 8x7b applies softmax over top2 experts, while Grok-1 applies top2 over the softmax of all 8 experts. - written… https://t.co/MlNTlgqO0s
Given Grok-1 is 314B MoE (~80B active params) , Im genuinely surprised why it's scored terribly 😑 https://t.co/tRh55WZA9e
Grok, OpenAI and Musk entered the chat !!! https://t.co/r2oAQ30xJi
Congrats @xai and @elonmusk on the open source release 🫡 Grok-1 cheat sheet 📝 - MoE architecture for 314B total params - 8 experts with 2 active - Apache 2.0 license - Trained on JAX and Rust - Finished training October 2023 - Base model, no task-specific fine-tune - No… https://t.co/ioKun85fm6
Grok is 2x the size of GPT-3.5? And still this dumb?
Grok-1 code is released. It's 8 experts, 2 selected at a time. Trend of large vocab (131k) continues. Attention output multiplier is interesting. Overall, it's much large model than I'd thought so bit surprised about lag in perf than other models. https://t.co/o6ZAUGLXiP https://t.co/EMTfp0vNeA
Cooling down a bit, Grok is 8x33B model with roughly Mixtral 8x7B's performance. Still huge respect for releasing it. To the best of my knowledge this is the largest decoder only model ever released. (The 1.6T params switch transformer on Huggingface is an encoder-decoder)
Elon Musk kept his word and released Grok-1🤯 Grok-1 is a 314B big Mixture-of-Experts (MoE) transformer. 🧐 What we know so far: 🧠 Base model, not fine-tuned ⚖️ Apache 2.0 license 🧮 314B MoE with 25% active on a token 📊 According to the initial announcement; 73% on MMLU,… https://t.co/Ep6i8uoKDY
can someone who knows more about model pretraining than i explain how grok is worse than mixtral at 7 times the size
can someone who knows more about model pretraining than i explain how grok is worse than mixtral at 1/7th the size
Grok-1 is MOE. Have big chance of finetune of it being bugged as Mixtral and the benchmarks dont improve so much.
Grok seems about as powerful as LLaMa 2, less powerful than GPT3.5, so probably not a big deal that the weights are open now. https://t.co/RS0C1tUjP9
HOLY SH*T @grok IS 314 BILLION PARAMETERS Mixture of 8 Experts, not RLHFd/moralized THIS IS HUGE https://t.co/kjsvEWp5O8 https://t.co/8CGAtNpbwT
Apple researchers have hit on a new multi-modal method of quickly training large language models (LLMs) that can enable more flexible and powerful machine-learning and "AI" type systems. https://t.co/jPCwURQLFq
AI and Robotics is moving fast. So, I share the most important research every week. Here's everything you need to know and how to make sense out of it:
New AI bombs are dropping daily. Massive Developments in AI from this week: - Google just dropped SIMA - Devin, the First AI Software Engineer - RunwayML adds Lip Sync - OpenAI Sora data revelation - ChatGPT got body. Here are 5 things you don't want to miss
The weekly AI overview is here to keep you informed on the latest news and top stories in AI. #AI #News #Overview #Week11 #AISpace (1/11) https://t.co/NKcgvRvaU4
“openai/grok” No this is not Grok from https://t.co/KthS0SlWM4. It is the humor of some folks at OpenAI. I say more soon. Link: https://t.co/74g6F0QZs3
This week's #AI updates: ➡️#AIAct sets global standards ➡️China’s railways get smarter with AI ➡️#ChatGPT bot now with robotic arm ➡️AI advances in surgery, an boosts crime data analysis #ShareForSuccess #SwissCognitive https://t.co/FQzrWRSELu
The openai/grok repository is the code for a paper on "Grokking" by: Alethea Power, Yuri Burda, Harri Edwards, Igor Babuschkin (OpenAI) Vedant Misra (Google) Submitted From: Alethea Power on Thursday 6 Jan 2022 18:43:37 UTC https://t.co/vuGX4nw5re
Apple announces a new series of multimodal LLMs https://t.co/uEzCLH3GxO
Apple paper on developing an LLM: "MM1: Methods, Analysis & Insights from Multimodal LLM Pre-training" https://t.co/XyYtnQk3rL
Apple's strides in multimodal AI research signal a turning point in the tech giant's investment strategy #AI #Apple #artificialintelligence #Investment #investments #llm #machinelearning #MM1 #model #multimodal #research #Techgiants https://t.co/w3NLw1BxkQ https://t.co/W3PANNQ2nR
“Apple researchers achieve breakthroughs in multimodal AI as company ramps up investments” Article: https://t.co/tEP39Ay4m0
The world of AI x Robotics is progressing fast. Just this month, we got massive announcements from Google, OpenAI, Midjourney, Nvidia, PIKA, Figure 1, and more. 1/ Nvidia just released Chat with RTX https://t.co/9SMy6wWjVG
➡️ Apple introduces MM1, a powerful multimodal AI system that competes with ChatGPT and sets new industry standards. Discover more here! https://t.co/SQlMnSDheV
AI Weekly Rundown (March 9 to March 15) Major AI announcements from Meta, OpenAI, DeepMind, Apple, Cohere, and more https://t.co/5lCTfgPxhi
Apple continues to make moves in AI --unveils Multimodal Large Language Models (MLLMs) #Apple #MLLMs https://t.co/0yej6TGQxJ
Apple continues to make moves in AI --announces Multimodal Large Language Models (MLLMs) #Apple #MLLMs https://t.co/0yej6TGQxJ
AI is moving at an incredible pace. 🤯 Massive announcements from this week: - Devin - OpenAI Sora - Google Deepmind - Figure Here's what you need to know:
➡️ Apple researchers make significant strides in multimodal AI, unlocking new possibilities for innovation. Don't miss out on the breakthroughs! https://t.co/bNcyefwBss
Apple is in the LLm game! Benchmarks seem pretty comparable to Gemini and a bit behind GPT4? Seems promising! Cmon plz do something cool with Siri https://t.co/lxlV5X9t6A https://t.co/b8GvzwTuUz