𝗥𝘂𝗻𝘄𝗮𝘆 𝗿𝗲𝗹𝗲𝗮𝘀𝗲𝗱 𝗠𝘂𝗹𝘁𝗶 𝗠𝗼𝘁𝗶𝗼𝗻 𝗕𝗿𝘂𝘀𝗵 𝗳𝗼𝗿 𝗚𝗲𝗻-𝟮 🖌️💥
It allows you to apply different animations to multiple elements in static images.
#ai
#ml
#generativeai
NVIDIA unveiled worlds most powerful AI chip 🦾
➲ Blackwell reduces cost up to 25% vs. H100
➲ It’s designed for trillion-parameter models
➲ 2000 Blackwell chips would take 90 days to train GPT-4 (1.8T parameters)
#artificialintelligence
#nvidia
#gtc
You can now visualise your RAG Data 〽️🧬
The animation displays UMAP embeddings of document snippets, color-coded by relevance to the question 'Who built the Nürburgring?'
This two-dimensional mapping of RAG data can help debug and enhance performance.
#ai
#generativeai
#rag
This paper explores 32 techniques to mitigate hallucinations in LLMs 🧠💭
It considers techniques such as Retrieval Augmented Generation (RAG) and Knowledge Retrieval, with a specific emphasis on dataset utilisation, tasks, and feedback mechanisms.
#ai
#rag
𝗡𝗲𝘄 𝘀𝗵𝗼𝗿𝘁 𝗰𝗼𝘂𝗿𝘀𝗲 𝗼𝗻 𝗟𝗟𝗠𝗢𝗽𝘀 📚💫
Taught by Google Cloud's Erwin Huizenga and AI Pioneer Andrew Ng, you'll learn to use automation to make building, tuning and deploying an LLM-based application less manual and more efficient.
#ai
#llmops
𝗕𝘆𝘁𝗲𝗗𝗮𝗻𝗰𝗲, 𝘁𝗵𝗲 𝗰𝗼𝗺𝗽𝗮𝗻𝘆 𝗯𝗲𝗵𝗶𝗻𝗱 𝗧𝗶𝗸𝗧𝗼𝗸, 𝗿𝗲𝘃𝗲𝗶𝗹𝗲𝗱 𝗠𝗮𝗴𝗶𝗰𝗩𝗶𝗱𝗲𝗼-𝗩𝟮 💥📹
Its new text-to-video generation model outperforms market leaders, including Pika 1.0 and SVD-XT, according to human evaluations.
#ai
#artificialintelligence
𝗜𝘀 𝘁𝗵𝗶𝘀 𝘁𝗵𝗲 𝗲𝗮𝘀𝗶𝗲𝘀𝘁 𝘄𝗮𝘆 𝘁𝗼 𝗳𝗶𝗻𝗲-𝘁𝘂𝗻𝗲 𝗮𝗻 𝗟𝗟𝗠? 💃
With gpt-llm-trainer, input your task description, and the system will generate a dataset, parse it, and fine-tune a LLaMA 2 or GPT-3.5 model.
It’s open-source. Link to repository in comments.
FDA clears first AI-powered handheld device that detects all major skin cancers 🧬💫
In an FDA study, DermaSensor accurately identified 96% of skin cancers and demonstrated a 97% success rate in correctly ruling out cancer.
#ai
#artificialintelligence
#healthtech
𝗔𝗻𝗱𝗿𝗲𝗷 𝗞𝗮𝗿𝗽𝗮𝘁𝗵𝘆 𝘁𝗮𝗹𝗸𝘀 𝗟𝗟𝗠 𝗢𝗽𝗲𝗿𝗮𝘁𝗶𝗻𝗴 𝗦𝘆𝘀𝘁𝗲𝗺 + 𝗺𝗼𝗿𝗲 🎙️
↓
Highlights:
➲ Major players like OpenAI are working towards an "LLM Operating System", similar to a traditional operating system, where LLMs serve as the "CPU" linked to various
Stanford researchers unveiled a robot that performs your tasks through imitation learning🧹🍳
ALOHA replicates human actions with precision after observing the performance of a specific task on a few occasions.
#ai
#machinelearning
𝗧𝗵𝗶𝘀 𝗶𝘀 𝗶𝗻𝘀𝗮𝗻𝗲! 𝗧𝗵𝗶𝘀 𝗔𝗜 𝘃𝗼𝗶𝗰𝗲 𝗰𝗵𝗮𝘁 𝘀𝗼𝘂𝗻𝗱𝘀 𝘄𝗮𝘆 𝘁𝗼𝗼 𝗵𝘂𝗺𝗮𝗻 😳
And it’s also 2x faster than any other setup of comparable quality 👇
Gazelle is a pioneering public LLM that:
- Listens directly to audio without needing to convert it to…
𝗕𝗿𝗲𝗮𝗸𝘁𝗵𝗿𝗼𝘂𝗴𝗵 𝗔𝗜 𝗶𝗺𝗮𝗴𝗲 𝗴𝗲𝗻𝗲𝗿𝗮𝘁𝗶𝗼𝗻 𝗶𝗻𝘁𝗿𝗼𝗱𝘂𝗰𝗲𝗱 𝗯𝘆 𝗠𝗜𝗧 ⚡
30 times faster than Stable Diffusion
Distribution Matching Distillation (DMD) is built different:
➲ Condenses knowledge of a multi-step model (like Stable Diffusion) into a
𝗚𝗼𝗼𝗴𝗹𝗲'𝘀 𝗚𝗲𝗺𝗶𝗻𝗶 𝗨𝗹𝘁𝗿𝗮 𝗲𝘅𝗽𝗲𝗰𝘁𝗲𝗱 𝘁𝗼 𝗯𝗲 𝗿𝗲𝗹𝗲𝗮𝘀𝗲𝗱 𝗹𝗮𝘁𝗲𝗿 𝘁𝗵𝗶𝘀 𝘄𝗲𝗲𝗸 💎☄️
It appears to have been confirmed through the Google Bard change log.
Link to source in comments.
Credit: Dylan Roussel on X
#ai
#generativeai
#google
𝗖𝗵𝗶𝗻𝗮 𝗮𝗶𝗿𝘀 𝗳𝗶𝗿𝘀𝘁-𝗲𝘃𝗲𝗿 𝗔𝗜 𝗮𝗻𝗶𝗺𝗮𝘁𝗲𝗱 𝘀𝗲𝗿𝗶𝗲𝘀 🧧
The text-to-video model series by China Media Group and Shanghai AI Lab features ancient tales inspired by classic Chinese poetry.
Via powerai
#artificialintelligence
#generativeai
#china
𝗢𝗽𝗲𝗻𝗔𝗜'𝘀 𝘂𝗽𝗱𝗮𝘁𝗲𝗱 𝗚𝗣𝗧-𝟰 𝗧𝘂𝗿𝗯𝗼 𝗿𝗲𝗰𝗹𝗮𝗶𝗺𝘀 𝘀𝘂𝗽𝗿𝗲𝗺𝗮𝗰𝘆 📈
Plus, more insights in Generative AI today from Cohere, Amazon, Meta and more.
Here's everything you need to know in Generative AI right now:
Google DeepMind introduced Genie, an AI that can generate interactive video games from text or image prompts 🧞
The model is trained on 200,000 hours of unsupervised public internet gaming footage.
#artificialintelligence
#generativeai
#google
𝗚𝗲𝗻𝗲𝗿𝗮𝘁𝗶𝘃𝗲 𝗔𝗜 𝗶𝘀 𝗴𝗲𝘁𝘁𝗶𝗻𝗴 𝘄𝗶𝗹𝗱: The First LLM Agent Operating System
Additionally, breakthroughs and developments from AMD, Cohere, Stanford, YouTube, Gretel, and more.
The most impactful Generative AI insights today:
The LLM API Hosts Leaderboard 💯
It compares the inference performance of model-host API combinations across various metrics including quality, throughput, latency, price, context window, and more.
Source: ArtificialAnalysis/X
#ai
#artificialintelligence
#inference
𝗜𝘀 𝗔𝗿𝘁𝗶𝗳𝗶𝗰𝗶𝗮𝗹 𝗜𝗻𝘁𝗲𝗹𝗹𝗶𝗴𝗲𝗻𝗰𝗲 𝘁𝗵𝗲 𝗳𝘂𝘁𝘂𝗿𝗲 𝗼𝗳 𝗼𝗽𝗲𝗿𝗮𝘁𝗶𝗻𝗴 𝘀𝘆𝘀𝘁𝗲𝗺𝘀? 💻🧠
Open Interpreter lets you control your computer via Large Language Models.
Video source: Open Interpreter
#ai
#artificialintelligence
#llm
Tencent researchers released a paper on recent advances in MultiModal Large Language Models 💎⚡
The paper analysed 26 MM-LLMs, comparing across benchmarks and detailing ‘key recipes’ for enhancing the models.
Link to paper:
#ai
#generativeai
Alibaba's EMO transforms any image/audio into an animated singing video 💫
Utilising frame encoding, diffusion, and expression mapping, it ensures facial expressions align seamlessly with the voice and mood of the audio.
Video credit: Freaky AI
#artificialintelligence
#emo
𝗢𝗽𝗲𝗻𝗔𝗜 𝗷𝘂𝘀𝘁 𝗮𝗻𝗻𝗼𝘂𝗻𝗰𝗲𝗱 𝗚𝗣𝗧-𝟰𝗼
A new model providing enhanced GPT-4-level capabilities to everyone, free of charge.
"im-also-a-good-gpt2-chatbot" in the LMSYS Arena last week was GPT-4o all along.
#generativeai
#artificialintelligence
#openai
𝗔𝗜 𝘂𝗻𝗶𝗰𝗼𝗿𝗻 𝗚𝗿𝗼𝗾 𝗙𝗼𝘂𝗻𝗱𝗲𝗿 & 𝗖𝗘𝗢:
“It’s completely broken that we keep thinking of each technology displacing work” - Johnathan Ross
He then gives a concrete example:
The Jevons Paradox.
"In the 1860’s.. every time steam engines got more efficient,…
𝗔 𝘃𝗶𝘀𝘂𝗮𝗹𝗶𝘀𝗮𝘁𝗶𝗼𝗻 𝗼𝗳 𝗗𝗕𝗦𝗖𝗔𝗡 𝗰𝗹𝘂𝘀𝘁𝗲𝗿𝗶𝗻𝗴 📿☄️
DBSCAN, a clustering algorithm, groups data points based on density, making it useful for identifying clusters of any shape.
#ai
#machinelearning
#data
Hugging Face launches open source AI Assistant maker to rival OpenAI’s custom GPTs 🤗💫
It's free to use and share, in contrast to requiring a $20 subscription.
#ai
#artificialintelligence
#huggingface
𝗣𝗿𝗼𝗺𝗽𝘁 𝗲𝗻𝗴𝗶𝗻𝗲𝗲𝗿𝗶𝗻𝗴 𝗽𝗮𝘁𝘁𝗲𝗿𝗻𝘀 📈⚡
An extensive area of knowledge and research crucial for developing applications with LLMs.
Image source:
#ai
#generativeai
#promptengineering
𝗡𝗩𝗜𝗗𝗜𝗔'𝘀 𝗯𝗶𝗹𝗹𝗶𝗼𝗻𝗮𝗶𝗿𝗲 𝗔𝗜 𝗳𝗼𝘂𝗻𝗱𝗲𝗿 & 𝗖𝗘𝗢 𝗝𝗲𝗻𝘀𝗲𝗻 𝗛𝘂𝗮𝗻𝗴:
“If you need expert advice on anything related to IT, software, business processes... this is the best team in the world”
He also added:
“Ladies and gentlemen: Deloitte.…
This paper explores both RAG and Fine-Tuning, including the accuracy improvements achieved by each approach 👩💻📈
This study revealed a cumulative accuracy increase of over 6 p.p. through fine-tuning the model, which further enhanced by 5 p.p. with RAG.
#ai
#rag
#finetuning
𝗢𝗽𝗲𝗻𝗔𝗜 𝗲𝘅𝗽𝗮𝗻𝗱𝘀 𝗳𝗶𝗻𝗲-𝘁𝘂𝗻𝗶𝗻𝗴 𝗔𝗣𝗜 𝗮𝗻𝗱 𝗰𝘂𝘀𝘁𝗼𝗺 𝗺𝗼𝗱𝗲𝗹𝘀 👌
It means greater control, specialised models and differentiated outcomes
New Fine-Tuning API Features:
➲ Epoch-based checkpoint creation for more effective training
➲ Comparative…
Researcher predict AI could beat humans at all possible tasks by 2047 🧠💫
The survey was conducted on 2,778 researchers who have published in leading artificial intelligence (AI) venues.
sSource: Research paper "Thousands of AI Authors on the future of AI"
#ai
#generativeai
⚡ 𝗡𝗘𝗪 𝗜𝗡 𝗚𝗘𝗡𝗘𝗥𝗔𝗧𝗜𝗩𝗘 𝗔𝗜 𝗧𝗢𝗗𝗔𝗬
Apple acquires new startup
Plus, more insights in Generative AI today from MIT, Harvard, OpenAI, Llama 3 and so much more!
Here's everything you need to know in Generative AI right now:
𝗜𝘀 𝘁𝗵𝗶𝘀 𝗵𝗼𝘄 𝘆𝗼𝘂 𝘂𝗻𝗱𝗲𝗿𝘀𝘁𝗮𝗻𝗱 𝗔𝗿𝘁𝗶𝗳𝗶𝗰𝗶𝗮𝗹 𝗜𝗻𝘁𝗲𝗹𝗹𝗶𝗴𝗲𝗻𝗰𝗲? 💃 💫
In most Artificial Intelligence subfields, a universally accepted definition or clear boundary is often elusive.
Image source: Jeff Winter
#ai
#generativeai
𝗠𝗼𝗹𝗲𝗰𝘂𝗹𝗲𝗦𝗧𝗠 𝗶𝘀 𝘁𝗵𝗲 𝗳𝗶𝗿𝘀𝘁 𝗺𝘂𝗹𝘁𝗶𝗺𝗼𝗱𝗮𝗹 𝘁𝗲𝘅𝘁-𝗰𝗵𝗲𝗺𝗶𝗰𝗮𝗹 𝘀𝘁𝗿𝘂𝗰𝘁𝘂𝗿𝗲 𝗺𝗼𝗱𝗲𝗹 🧬
Users can input text prompts to generate molecules tailored to specific properties, helping to advance drug design.
Link to paper and code in comments.
The new MetaVoice-1B is a 1.2 billion parameter based model trained on 100,000 hours of speech for text-to-speech 🗣️💫
It supports cross-lingual voice cloning with fine-tuning, zero-shot cloning for British & American voices, and more.
#ai
#artificialintelligence
#metavoice
𝗪𝗵𝗲𝗻 𝘄𝗶𝗹𝗹 𝘄𝗲 𝘀𝗲𝗲 𝗮 𝟭 𝗽𝗲𝗿𝘀𝗼𝗻 𝗯𝗶𝗹𝗹𝗶𝗼𝗻-𝗱𝗼𝗹𝗹𝗮𝗿 𝗰𝗼𝗺𝗽𝗮𝗻𝘆? 💸🧠
OpenAI CEO Sam Altman shares that AI will make this a reality.
#ai
#artificialintelligence
#open
Deep Reinforcement Learning (DRL) holds immense power, especially in scenarios that require decision-making and sequential interactions.
Experience how
#AI
systems that can learn, adapt, and make informed choices.
@interestingengineering
#machinelearning
#teels
𝗛𝗶𝗴𝗴𝘀𝗳𝗶𝗲𝗹𝗱 𝗔𝗜 𝗿𝗲𝗹𝗲𝗮𝘀𝗲𝘀 𝗺𝗼𝗱𝗲𝗹 𝗳𝗼𝗿 𝗽𝗲𝗿𝘀𝗼𝗻𝗮𝗹𝗶𝘀𝗲𝗱 𝗰𝗵𝗮𝗿𝗮𝗰𝘁𝗲𝗿 𝘃𝗶𝗱𝗲𝗼𝘀 𝗼𝗻 𝘀𝗼𝗰𝗶𝗮𝗹 𝗺𝗲𝗱𝗶𝗮 💃
You can take a selfie and get a character with lifelike motion using its iOS app.
#generativeai
#artificialintelligence
…
𝗬𝗼𝘂 𝗰𝗮𝗻 𝗻𝗼𝘄 𝗴𝗲𝗻𝗲𝗿𝗮𝘁𝗲 𝗽𝗿𝗼𝗱𝘂𝗰𝘁𝗶𝗼𝗻-𝗿𝗲𝗮𝗱𝘆 𝗽𝗿𝗼𝗺𝗽𝘁𝘀 𝗶𝗻 𝘁𝗵𝗲 𝗔𝗻𝘁𝗵𝗿𝗼𝗽𝗶𝗰 𝗖𝗼𝗻𝘀𝗼𝗹𝗲..
Describe your objective, and Claude will use prompt engineering techniques, like chain-of-thought reasoning, to craft prompts.
#generativeai
𝗡𝘃𝗶𝗱𝗶𝗮 𝗖𝗘𝗢 𝗝𝗲𝗻𝘀𝗲𝗻 𝗛𝘂𝗮𝗻𝗴 𝗲𝗻𝘃𝗶𝘀𝗶𝗼𝗻𝘀 𝗮 𝗳𝘂𝘁𝘂𝗿𝗲 𝘄𝗵𝗲𝗿𝗲 𝗵𝘂𝗺𝗮𝗻𝘀 𝗱𝗼𝗻'𝘁 𝗻𝗲𝗲𝗱 𝘁𝗼 𝗸𝗻𝗼𝘄 𝗵𝗼𝘄 𝘁𝗼 𝗰𝗼𝗱𝗲 👩💻
He also argues that children should stop learning how to code.
What are your thoughts?
#generativeai
#nvidia
You can now download Google OpenStreetMap, the largest building dataset for any country with one line of code 🗺️〽️
It's part of Leafmap, a Python package that's both free and open-source.
#ai
#python
#google
Is Cognition AI's Devin the most capable AI software agent?👩💻
Devin achieved an unassisted score of 13.86% on the real-world software engineering benchmark (SWE-bench), surpassing previous state-of-the-art performance ~2%.
#artificialintelligence
#generativeai
#devin
𝗚𝗲𝗻𝗲𝗿𝗮𝘁𝗶𝘃𝗲 𝗔𝗜 𝗷𝘂𝘀𝘁 𝗸𝗲𝗲𝗽𝘀 𝗼𝗻 𝗴𝗶𝘃𝗶𝗻𝗴 💃
Google DeepMind’s CEO said it’d be difficult to catch up to Sora.
Additionally, breakthroughs and developments from Meta, Elon, Taiwan Semiconductor, Spotify, and more.
The most impactful Generative AI insights…
𝗧𝗵𝗶𝘀 𝗽𝗮𝗽𝗲𝗿 𝗲𝘅𝗽𝗹𝗼𝗿𝗲𝘀 𝟰𝟰𝟰 𝗟𝗟𝗠 𝗱𝗮𝘁𝗮𝘀𝗲𝘁𝘀 𝗮𝗰𝗿𝗼𝘀𝘀 𝟯𝟮 𝗱𝗼𝗺𝗮𝗶𝗻𝘀 🪢
Several datasets are colour categorised in the timeline below.
🔶 = Pre-training
🌝 = Instruction fine-tuning
🍀 = Preference
💞 = Evaluation
Link to paper in comments.
Shopify's new AI image editor for rapid creation of professional product photos 👠💫
Media Editor lets sellers change item backgrounds, styles, colours , etc. with prompts.
#ai
#artificialintelligence
#generativeai
𝗛𝗼𝗹𝗹𝘆𝘄𝗼𝗼𝗱 𝗮𝗴𝗲𝗻𝗰𝘆 𝗔𝗜 𝗰𝗹𝗼𝗻𝗶𝗻𝗴 𝗮𝗰𝘁𝗼𝗿𝘀 🔮
Plus, more insights in Generative AI today from JPMorgan, OpenAI, LMSYS, Meta and so much more
Here's everything you need to know in Generative AI right now:
Is Groq the world's fastest chatbot? 🐊
Unlike traditional GPUs, which are parallel processors with hundreds of cores designed for graphics rendering, LPUs are architected to deliver deterministic performance for AI computations.
#artificialintelligence
#generativeai
#groq
𝗦𝘁𝗮𝗯𝗶𝗹𝗶𝘁𝘆 𝗔𝗜 𝗿𝗲𝗹𝗲𝗮𝘀𝗲𝘀 𝗦𝘁𝗮𝗯𝗹𝗲 𝗔𝘂𝗱𝗶𝗼 𝟮.𝟬 💯
↓
Highlights:
➲ It creates 3-minute, high-quality (44.1 kHz stereo) full coherent music tracks from a single prompt
➲ You can even upload audio samples and transform them into various sounds through…
𝗬𝗼𝘂 𝗰𝗮𝗻 𝘁𝗿𝗮𝗶𝗻 𝗚𝗣𝗧-𝟮 (𝟭𝟮𝟰𝗠) 𝗶𝗻 𝟵𝟬 𝗺𝗶𝗻𝘂𝘁𝗲𝘀 𝗳𝗼𝗿 $𝟮𝟬..
Andrej Karpathy’s new tutorial breaks it down in llm.c using an 8xA100 GPU.
The 124M model is the smallest model in the GPT-2 series released by OpenAI in 2019.
The 350M version requires 14
The First LLM Agent Operating System
AIOS (LLM Agent Operating System) is a new agent orchestration framework that embeds large language models into operating systems, creating an OS with a "brain" that can "understand"
𝗧𝗿𝗶𝘀𝘁𝗮𝗻 𝗛𝗮𝗿𝗿𝗶𝘀 𝘄𝗮𝗿𝗻𝘀 𝗮𝗯𝗼𝘂𝘁 𝗔𝗜’𝘀 𝗿𝗲𝗰𝘂𝗿𝘀𝗶𝘃𝗲 𝘀𝗲𝗹𝗳-𝗶𝗺𝗽𝗿𝗼𝘃𝗲𝗺𝗲𝗻𝘁 ☄️
↓
What are your thoughts?
Credit: Tristan Harris, Co-Founder of the Center for Humane Technology / Summit
#generativeai
#artificialintelligence
New research indicates that LLMs demonstrate higher performance when prompted to act as gender-neutral or male rather than female 🧬📊
The University of Michigan study underscores the importance of rectifying biases that may be present in the training data.
#ai
#generativeai
This visualisation shows fractal patterns in neural network hyperparameters☄️💎
➲ Blue colors = hyperparameters where training converges
➲ Red colors = hyperparameters where training diverges
Credit: Jascha Sohl-Dickstein
#ai
#artificialintelligence
#neuralnetworks
𝘅.𝗔𝗜 𝘂𝗻𝘃𝗲𝗶𝗹𝗲𝗱 𝗚𝗿𝗼𝗸-𝟭.𝟱 𝗩𝗶𝘀𝗶𝗼𝗻 👀
Plus, more insights in Generative AI today from OpenAI, Netflix, Elon Musk, Mojo and more.
Here's everything you need to know in Generative AI right now:
𝗥𝗲𝘀𝗲𝗮𝗿𝗰𝗵𝗲𝗿𝘀 𝗳𝗿𝗼𝗺 𝗧𝗲𝗻𝗰𝗲𝗻𝘁 𝗿𝗲𝗹𝗲𝗮𝘀𝗲 𝗜𝗻𝘀𝘁𝗮𝗻𝘁𝗠𝗲𝘀𝗵 🐘
Plus, more insights in Generative AI today from Microsoft, LangChain, Sam Altman, and more
Here's everything you need to know in Generative AI right now: