Search the archive

COGVLM-17B OPEN VISION LANGUAGE MODEL

With 10B vision parameters and 7B language parameters, this multimodal model excels at many standard benchmarks and performs well on human evaluation.

LANGROID MULTI-AGENT PROGRAMMING FRAMEWORK

Inspired by the Actor framework, this lightweight Python library makes it easy to create LLM-powered agents.

INSTANT 3D SEARCH

OVIR-3D innovates 3D object retrieval from text prompts using 2D image fusion, bypassing the need for 3D data training and facilitating instant, real-time search capabilities ideal for robotics.

ALMOST AN AGENT: WHAT GPTS CAN DO

OpenAI's latest GPTs edge closer to the vision of autonomous AI agents. They are capable of executing tasks like crafting academic papers with minimal human input, despite current constraints like occasional inaccuracies. This leap forward in AI autonomy promises significant advancements but also underscores the urgency for vigilant development to mitigate potential vulnerabilities as AI begins to navigate more complex and independent roles.

ARE LANGUAGE MODELS GOOD AT MAKING PREDICTIONS?

In a study using 5000 questions from Manifold Markets, GPT-4 was found to be consistently overconfident in its predictions.

OPENAI IS A LOT MORE VULNERABLE THAN YOU THINK

On the heels of OpenAI releasing functionality that decimated half GPT-wrapper companies, here is a contrarian view of the blind spots and weaknesses of OpenAI as this market continues to evolve.

AUTOMATIC SPORTS NARRATION (JUPYTER NOTEBOOK)

This notebook uses many new technologies from OpenAI to convincingly narrate a sporting event.

AI NEGOTIATES CONTRACT FOR THE FIRST TIME EVER

British AI firm Luminance has developed an AI system that can autonomously negotiate contracts, aiming to streamline the workload of lawyers by handling routine negotiations.

ORBA (PRODUCT)

Convert your web visitors with personalized AI conversations.

OPENAI DEV DAY RECAP

GPT-4 Turbo at a fraction of the cost with 128k context, custom GPTs, and vision in the API.

XAI RELEASES PROMPTIDE

xAI has released PromptIDE, an integrated development environment for prompt engineering and interpretability research.

30T OPEN DATASET RELEASED

Red Pajama v2 has been released. It is 2.5x larger than the dataset used to train GPT-4. There is no code data, but it does have multilingual data from 5 languages and covers an impressive breadth. Importantly, much of the deduplication work and some heuristic filtering has already been done.

SHARPER IMAGES WITH GAUSSIAN MIXTURE SOLVERS FOR DIFFUSION MODELS

Gaussian Mixture Solvers enhance diffusion models, resolving the speed-quality trade-off in image synthesis and setting new standards for image detail and clarity.

ENHANCING ANOMALY DETECTION WITH DEPTH-AWARE TECH

The DADA system integrates 3D data with RGB images for superior surface anomaly detection, outperforming traditional methods in speed and accuracy without extensive training datasets.

A UNIFIED APPROACH FOR LANGUAGE-DRIVEN RECOMMENDER SYSTEMS

CLLM4Rec synergizes large language models with ID-based systems, elevating recommendation precision through advanced pretraining and novel language embeddings.

OPENAI RELEASES CONSISTENCY DECODER FOR STABLE DIFFUSION

Consistency decoder for the VAE dramatically improves performance for generated images for latent diffusion models. They released one for Stable Diffusion. Open source!

TOPICGPT: DISCOVERING TOPICS IN TEXT

TopicGPT revolutionizes topic modeling with LLMs, producing more relevant, interpretable topics and allowing for user-guided topic refinement.

DETECTING DEEP FAKES WITHOUT SEEING ANY

FACTOR applies fact-checking techniques to digital media, offering a novel, effective tool for detecting unprecedented deepfake attacks without prior training.

WILL GENERATIVE AI TRANSFORM BUSINESS?

Generative AI's increased sophistication has led to a rise in business applications, with Insider Intelligence predicting that 25% of US internet users will use it monthly by 2023. The technology's capacity to communicate in natural language has increased accessibility, extending its use beyond tech-savvy individuals. Although still in the early stages, businesses like Yext and numerous startups are already leveraging generative AI for various tasks such as marketing content generation, customer service, product design, and data analysis.

A MINOR CHATGPT UPDATE IS A WARNING TO FOUNDERS: BIG TECH CAN BLOW UP YOUR STARTUP AT ANY TIME

We’ve already seen it with investors not willing to invest in companies that are just wrapping GPT but now it is even more clear. With the ability to query uploaded PDFs, a whole swath of startups just got wiped off the map.

PIXEL PERFECT: HOW AI UNLOCKS CREATIVITY

This post breaks down how AI, creators, and scaled distribution tactics like virality and community-led growth are generating huge momentum across these vectors.

CHATGPT CONTINUES TO BE ONE OF THE FASTEST-GROWING SERVICES EVER

Sam Altman announced significant usage milestones for ChatGPT at OpenAI’s developer conference, with the platform reaching 100 million weekly users and engaging over two million developers, including a majority of Fortune 500 companies.

XBOX PARTNERS WITH INWORLD AI TO BUILD AI TOOLS FOR GAME DEVELOPMENT

Microsoft's Xbox is collaborating with Inworld AI to provide AI-powered tools for game development, focusing on narrative and character creation to facilitate and enrich the game design process.

SALESFORGE (PRODUCT)

Personalize every single email at any scale and in any language.

ELON MUSK'S XAI BEATS CHATGPT ON LLM BENCHMARKS

xAI trained an autoregressive language model at 34B params. It is quite performant and will power an AI system built into the X platform.

WHAT TO EXPECT AT OPENAI DEV DAY

OpenAI Dev Day is today and it is rumored that we can expect enhancements to OpenAI’s developer tools, new pricing plans for ChatGPT, and a peek at Gizmo V8, a new and improved version of the ChatGPT iOS app.

EXPANDING VRP FOR ENHANCED GENERATIVE AI SECURITY

HIGHLIGHTS FROM NEURIPS 2023

Paper digest generated summaries of many papers from NeurIPS 2023. Interestingly, much of this work is almost a year old at this point and has already been readily adopted by the community.

HELIXNET IS 3 MODELS COMBINED INTO ONE

If you fine-tune three task-specific models from the Mistral base, one for generation, one for critique, and a final one for regeneration then the entire system shows dramatically improved generation performance. Synthetic data is used to tune these models.

SKETCH WHAT YOU WANT YOUR ROBOT TO DO

Sketching a rough outline of what you want a robot to accomplish is a novel form of communication and turns out to work surprisingly well for standard pick and place tasks.

GISKARD: OPEN-SOURCE TESTING FRAMEWORK FOR LLMS & ML MODELS (GITHUB REPO SPONSOR)

Giskard is a Python library [11] that automatically detects vulnerabilities of AI models, from tabular models to LLM, including: performance biases, data leakage, spurious correlation, hallucination, toxicity, security issues and many more. Get your model to production quickly and safely.

Install Giskard (via GitHub) [11] or try it in Colab

DEEPSPEED RELEASES INFERENCE SOFTWARE

This is Deepspeed's new software for competing with vLLM and text-generation interfaces for the fast serving of language models. It comes with many state-of-the-art acceleration features and preliminary results suggest it's up to 2.4x faster.

X "SIMILAR POSTS" POWERED BY RUST-BASED VECTOR DATABASE

Qdrant uses a lightning-fast vector database to store embeddings.

VIDEO2MUSIC CREATES TAILORED MUSIC FOR VIDEOS

Video2Music’s AI framework generates emotionally congruent music for videos, leveraging a unique dataset and transformer model validated by user studies.

AI DATA PIPELINE ATTACKS

Poisoning the data well and other data pipeline attacks are a huge challenge for the cyber security community and often a blindspot for most AI organizations. This post outlines (with code) what the attacks are to help with future mitigation.

COULD CRUISE BE THE THERANOS OF AI?

Cruise, GM's driverless car company, frequently relies on remote operators for its vehicles, challenging the company's claims of autonomy. The revelation raises concerns about the true capabilities of self-driving cars and prompts calls for transparency and investigation into the safety and autonomy of such vehicles.

EXTENSIONS PROMPT INJECTION AND DATA EXFILTRATION

A great deep dive into some novel vulnerabilities that come from generative AI plugins. In this case, the attack resembles SQL injection but came via an infected Google Doc. While Google has fixed the issue, the security researcher was able to exfiltrate other users’ prompts.

Start with free credits!

GOOGLE'S RESEARCH SCHOLAR PROGRAM IS OPEN FOR APPLICATIONS

Google's program allows early career postdoc-level researchers to join a team with resources and continue their research agenda.

CHATGPT PLUS CAN NOW ANALYZE FILES

The latest update on ChatGPT Plus introduces fresh beta functionalities, including the ability to work with PDFs. Users can upload and interact with documents alongside receiving multimodal assistance.

LAYER (PRODUCT)

Layer cuts your research time by 40% with AI-powered summarization, information extraction, and more.

META’S AI RESEARCH HEAD WANTS OPEN SOURCE LICENSING TO CHANGE

Meta's Llama 2, touted as open-source, faces scrutiny due to its limited license not meeting all Open Source Initiative requirements. Despite this, Meta's VP for AI research, Joelle Pineau, emphasizes the balance between information-sharing and business costs, aiming for collaborative innovation akin to its previous open-source project PyTorch.

STABILITY’S AI LATEST TOOL USES AI TO GENERATE 3D MODELS

Stability AI has unveiled Stable 3D, an AI-driven app aimed at simplifying 3D model creation for modeling and game development.

UK INVESTS $273M IN AI SUPERCOMPUTER

The UK government is investing $273 million in an AI supercomputer dubbed Isambard-AI, which will be constructed by the University of Bristol, showcasing its aspiration to excel in AI technology.

LONG(ER) CONTEXT MISTRAL MODEL

Amazon enters the open source LLM game with a RoPE fine-tuned Mistral 7B that scales up to 32k tokens in context.

LONG(EST) CONTEXT MISTRAL MODEL

Nous research has used their Yarn method to fine-tune Mistral 7B to operate on contexts up to 128k. It shows minimal degradation on short sequence lengths and vastly improved performance on long documents.

TUNING LANGUAGE MODELS FOR EMPATHY

Researchers have developed a way to improve empathy in large language models used for psychological counseling by creating a dataset for empathetic conversation training.

OPEN SOURCE TOOLKIT FOR RAG, FINE-TUNING, AND MODEL SERVING

A minimal and open source Python toolkit for building LLM applications.