In recent developments in the AI sector, Meta's LLaMA 3 and Microsoft's Phi-3 have been making significant strides. LLaMA 3, which was released a week ago, has seen enhancements such as an extended context window from 8K to 128K and increased inference speeds of over 800 tokens per second. Microsoft's Phi-3 is highlighted as a small yet highly capable AI model, with potential applications in 10 wild examples. Apple is also advancing in AI, focusing on developing lightweight AI models that can operate on its devices, potentially integrating these into the upcoming iOS 18. Additionally, Apple has released several small AI language models and is in discussions with OpenAI to incorporate their technology into new iPhone features.
Llama 3 is the latest Meta open-source LLM to a arrive on the scene, so Eivind Kjosbakken decided to explore its power and offer a clear workflow for anyone who'd like to run the model locally. https://t.co/SMSK2Y6tze
ChatGPT Could Power the iPhone's AI Chatbot: Report https://t.co/0Yk8mPKrxV https://t.co/oWRdP1ULzT
Apple’s new flagship smartphone could be filled with AI features https://t.co/eTnegTICI2
iPhone 16 Pro Max To Launch With Generative AI Features: Apple Intensifies Talks With OpenAI https://t.co/5hwQjvTrmF
Apple is weighing a big decision impacting the next iPhone release: OpenAI or Google's Gemini https://t.co/6pytwsXflL
Apple's upcoming iOS 18 might revolutionize #AI interaction with on-device processing, promising more personal and secure AI features.🍏🧠 #ArtificialIntelligence #Tech #AINews https://t.co/DkLxHZyZNt
Apple is working on smaller local models, but they also want a large model to run under iOS 18 on the iPhone. They have had discussions with Anthropic, Google, and OpenAI. They have now reentered negotiations with OpenAI. Massive deal for whoever lands it. WWDC is on June 10th. https://t.co/GQl3UsUU5E
Apple shipping 2019-era AI capability in 2024 Appropriately captures the company's apparent lack of AI urgency as the revolution in language-based UI portends a new hardware-software era https://t.co/EMB6mOXAlF https://t.co/OvDkC7cUXE
Apple’s iOS 18 features they will highlight at WWDC are based on an in-house model. The talks with OpenAI and Google are for a chatbot/search component. https://t.co/preidMTAPR
NEWS: Apple in talks with OpenAI to use its technology to power some new features coming to the iPhone this year, according to Bloomberg But… why not Grok? @elonmusk https://t.co/SEGowI3iqL
Apple Releases 8 Small AI Language Models To Compete Wit Microsoft’s Phi-3 ► https://t.co/XSPyveKMPR https://t.co/XSPyveKMPR
Apple Offers Peek at Its AI Language Model as iOS 18 Looms - CNET https://t.co/Xq15EzFAtJ
Big week for open-source AI. In just a few days, we've seen two major models launch. Meta Llama 3 and Microsoft Phi-3, each claiming to outperform the other. So I conducted complex coding tests: 🧵 https://t.co/0MU7uCPC1t
Colossal-Inference now supports Llama 3 inference acceleration. They report a ~20% enhancement in training efficiency for Llama 3 8B and 70B and outperforming alternative inference solutions such as vLLM. This is why open-source AI matters. There are all kinds of innovations… https://t.co/BDlR2iGKo2 https://t.co/2r9WCBKQBP
Colossal-Inference now supports Llama 3 inference. They report a ~20% enhancement in training efficiency for Llama 3 8B and 70B and outperforming alternative inference solutions such as vLLM. This is why open-source AI matters. There are all kinds of innovations happening… https://t.co/eTH2AIXXii https://t.co/2r9WCBKQBP
Apple has submitted eight new AI models to the Hugging Face hub, an online platform for open-source AI. While similar to other models like OpenAI's GPT-4, the difference is in the size of these models. Apple is clearly building models designed to function on our devices. https://t.co/JHVvnxdwCB
LLaMA3 and Phi3 have made the splash this week in LLM Arena. But how strong is their visual understanding ability? ⚡We release LLaMA3-Vision and Phi3-Vision models that beat their larger size LLM competitors. Github: https://t.co/o4AVEn0AYF HF: https://t.co/AujQeYLMlG https://t.co/NJqNT9K8he
Top stories in AI today: -Chinese AI model bests GPT-4 Turbo -Sanctuary releases next-gen Phoenix -Access Llama 3 on your Phone -Synthesia unveils ‘Expressive Avatars’ -6 new AI tools & 4 new AI jobs Read more: https://t.co/82bSB4PgMk https://t.co/HMt0uEcAbg
Llama 3 extended to almost 100,000-token context! ✅ By Combining PoSE and continuing pre-training on Llama 3 8B base for 300M tokens, the community (@winglian) managed to extend the context from 8k to 64k. 🚀 Applying rope scaling afterward led to a supported context window of… https://t.co/3eICXS95eA
#FPTech: iPhone’s AI: Apple launches OpenELM, open-source model that runs locally on device, without cloud https://t.co/bVzNAJpA8X
This model extends LLama-3 8B's context length from 8k to > 160K, developed by @Gradient_AI_ , sponsored by compute from @CrusoeEnergy . Demonstrates that SOTA LLMs can learn to operate on long context with minimal training (< 200M tokens) by appropriately adjusting RoPE theta.… https://t.co/2e8NnarhfS
How to Run Llama 3 Locally? Let's Have a Look! #DL #AI #ML #DeepLearning #ArtificialIntelligence #MachineLearning #ComputerVision #AutonomousVehicles #NeuroMorphic #Robotics https://t.co/iLHp4Pojf0
Ahead of WWDC, Apple has released its open-source efficient language models, hinting at its ongoing progress with AI. #Apple #WWDC #AI #LLM https://t.co/ukKOEz1KhI
Run an AI Town locally, powered by llama3 🎉 No cloud signups needed. Make your own world, and then talk to it :) Runs the open-source @convex_dev backend locally. Use @ollama locally or @togethercompute for cloud LLM. @realaitown https://t.co/7YmK6wBVIS
The OS LLama3 is moving fast. Llama3 8B-instruct with 160K context window, done with progressive training on augmented generations of increasing context lengths of SlimPajama https://t.co/M8jIGOr62I
Quantized matmuls in the latest MLX are up to 40% faster thanks to @angeloskath and @DiganiJagrit QLoRA fine-tuning Llama 3 70B on a single M2 Ultra stats: - Batch size 4 with 16 LoRA layers - 95 toks/sec - Peak mem 41GB - Avg power 120 W https://t.co/dLuMAGrpzd
Apple’s most likely strategy: developing artificial intelligence software that’s small and light-weight enough to run on its devices. https://t.co/l1xrk432GA AI Agenda by @KalleyHuang
It's been a week since LLaMA 3 dropped. In that time, we've: - extended context from 8K -> 128K - trained multiple ridiculously performant fine-tunes - got inference working at 800+ tokens/second If Meta keeps releasing OSS models, closed providers won't be able to compete.
Llama 3 surprised everyone less than a week ago, but Microsoft just dropped Phi-3 and it's incredibly capable small AI model. We may soon see 7B models that can beat GPT-4. People are already coming up with incredible use cases. 10 wild examples: