AI2's new open-source LLM may reset the definition of open AI'
AI2 has released a new large language model (OLMo 7B) and made all software components and training data available on GitHub and Hugging Face.
The goal is to give the AI research community full visibility into the model, enabling them to advance natural language processing and improve existing models.
This move aims to address the challenge of attributing specific outputs by an LLM to training data, allowing researchers to understand and evaluate model behavior. [ more ]
Google Brings Gemini Nano to Chrome to Enable On-Device Generative AI
Google announced plans to bring on-device large language models, like Gemini Nano, to Chrome for better privacy, reduced latency, offline access, and a hybrid computation approach. [ more ]
The role of a prompt engineer has attracted significant interest due to the potential for high salaries without a traditional tech background.
Prompt engineer roles are not simply typing questions into a prompt window, but involve designing intricate sequences of prompts to guide powerful language models. [ more ]
Build and Deploy Multiple Large Language Models in Kubernetes with LangChain
Deploying large language model architectures requires a mix of specialized, generic, and externally sourced models to meet various departmental needs. [ more ]
Scientists increasingly using AI to write research papers
Generative AI is potentially writing a significant portion of scientific literature based on linguistic and statistical analyses of research papers. [ more ]
AI Model Backed by Asia's Richest Person to Launch in March
India's BharatGPT group, in collaboration with engineering schools and Reliance, is set to launch ChatGPT-style service named Hanooman for various sectors.
Startups in India are developing open-sourced AI models tailored for Indian needs, in contrast to Silicon Valley's large language models. [ more ]
Meta's AI chief: LLMs will never reach human-level intelligence
AGI predictions vary widely, with some industry leaders suggesting it could arrive within five years while others, like Yann LeCun, argue that human-level AI is a more feasible goal.
Current AI systems lack key cognitive capabilities essential for human-like intelligence, such as reasoning, planning, memory, and understanding the physical world. [ more ]
Using the PowerInfer method, language models can be made more efficient by offloading some neurons to GPU and the rest to CPU.
PowerInfer offers significant efficiency improvements over previous methods by utilizing a power law distribution of neuron activation in language models. [ more ]
Meta AI has introduced an interactive guide called 'Prompt Engineering with Llama 2' to elevate the skills of developers, researchers, and enthusiasts in the domain of large language models
The guide provides hands-on experience in prompt engineering, which involves crafting inputs to guide language models to produce desired outputs [ more ]
Researchers Introduce Proxy-Tuning: An Efficient Alternative to Finetuning Large Language Models
Researchers have introduced a method called proxy-tuning to streamline the adaptation of large pretrained LMs efficiently.
Proxy-tuning is a lightweight, decoding-time algorithm that involves tuning a smaller language model and applying the predictive differences to shift the predictions toward the desired goal. [ more ]
Kong's new open source AI Gateway makes building multi-LLM apps easier | TechCrunch
Kong is launching an open source AI Gateway as an extension of its existing API gateway to integrate applications with large language models.
The AI Gateway includes features for prompt engineering, credential management, and more to make building on AI more productive for developers. [ more ]
Learn About LLMs With These ODSC East 2024 Sessions
Large Language Models (LLMs) are transforming the world and the field of data science at an unprecedented pace.
The ODSC East conference offers training sessions and workshops focused on LLMs, including topics like NLP with GPT-4 and enabling complex reasoning with LLMs. [ more ]
Stability AI Releases 1.6 Billion Parameter Language Model Stable LM 2
Stability AI has released pre-trained model weights for the Stable LM 2 language model, a 1.6B parameter model trained on 2 trillion tokens of text data from seven languages.
The model is available in two versions: the base model and an instruction-tuned version called Stable LM 2 Zephyr. [ more ]
Google's multimodal large language models, Gemini, now power the conversational experience within Google Ads, making it easier for advertisers to build and scale Search ad campaigns.
The conversational experience in Google Ads uses a chat-based tool that generates relevant ad content, including assets and keywords, based on a website URL. It also suggests images using generative AI.
Beta access to the conversational experience is currently available to English language advertisers in the US and UK, with global access opening up in the next few weeks and plans to expand to additional languages in the future. [ more ]
Meta's AI chief doesn't think AI super intelligence is coming anytime soon, and is skeptical on quantum computing
Yann LeCun believes current AI systems are decades away from reaching sentience and common sense capabilities.
LeCun believes the technology industry's current focus on language models and text data will not be enough to create advanced human-like AI systems. [ more ]
Why it's important to remember that AI isn't human
ChatGPT remains a topic of debate among experts, with opinions ranging from it being a potential threat to civilization to it being a sophisticated auto-complete tool.
The emergence of language models like ChatGPT raises questions about the link between language and the mind, and whether a new form of mind has been created.
Interacting with chatbots can be misleading due to the ambiguity in language, requiring us to rely on our intention-guessing mechanism for effective communication. [ more ]
Silo AI releases checkpoint on mission to democratise LLMs
Large language models work more effectively in English, creating language bias and limiting access to knowledge and innovation in other languages.
Silo AI has released the multilingual open European LLM Poro 34B, which has shown best-in-class performance for low-resource languages like Finnish. [ more ]