OPENAI MAKES GPT-4 GENERALLY AVAILABLE
Get out from under GPT-3.5 and start using GPT-4, which is generally available starting today.
2023.07.07
Search the archive
Full-text search across article titles and summaries. Refine with sector, year, content type, and reading-time filters.
Results
16,678 matchesCHATGPT DROPS ABOUT 10% IN TRAFFIC AS THE NOVELTY WEARS OFF
Worldwide desktop and mobile web traffic to the ChatGPT website dropped 9.7% from May to June, according to preliminary estimates. In the U.S., the month-over-month decline was 10.3%. Worldwide unique visitors to ChatGPT’s website dropped 5.7%. The amount of time visitors spent on the website was also down 8.5%. ChatGPT still attracts more worldwide visitors than Bing, Microsoft’s search engine.
2023.07.07
DEVELOPMENT
LlamaIndex has launched version 0.7.0 with the aim of improving the library's modularity and customizability. The new version includes standalone LLM abstractions for OpenAI, HuggingFace, and PaLM, allows users to use the response synthesis module independently, and offers enhanced metadata management for Document/Node objects. The update was designed to enhance LLM application development and offer developers more control over their workflows.
2023.07.07
TEXT-TO-IMAGE EDITING THROUGH FEATURE CORRESPONDENCE
This repo introduces DragonDiffusion, a novel method that enhances the editing precision of text-to-image models. Utilizing feature correspondence within diffusion models, DragonDiffusion allows various detailed image alterations, such as moving and resizing objects, replacing appearances, and dragging content, all while maintaining the image's original consistency and without the need for additional model fine-tuning.
2023.07.07
MODELS GENERATING TRAINING DATA: HUGE WIN OR FAKE WIN?
There has been buzz the past few days around a number of works suggesting that models generating training data is a new unlock for performance. On the other hand, the community has pushed back, saying that the data processing inequality suggests this is impossible. What if the truth is somewhere in between and generating training data acts as a sort of filter for sampling from the distribution you actually care about?
2023.07.07
ENHANCING PROBLEM-SOLVING ABILITIES OF LLMS
This technical report details the creation of FLACUNA, a language model that improves problem-solving capabilities by fine-tuning VICUNA, another large language model, using a custom instruction dataset called FLANMINI.
2023.07.07
IMAGE SEGMENTATION WITH HIERARCHICAL, OPEN-VOCABULARY LEARNING
This paper presents HIPIE, a novel approach to image segmentation that accounts for complex scenes by incorporating different semantic levels into its learning process. It outperforms existing models across a range of image comprehension tasks, offering a unified solution for segmentation tasks in diverse contexts.
2023.07.07
INTERNLM
InternLM has open-sourced a 7 billion parameter base chat model tailored for practical scenarios.
2023.07.07
TALKING ABOUT A “SCHISM” IS AHISTORICAL
This article argues that the framing of the debate between AI safety and alignment as a "schism" is ahistorical and misleading. The author argues that there has never been a unified AI safety community and that the two camps have different goals and values.
2023.07.07
ADOBE LIMITS ITS EMPLOYEES’ USE OF GENERATIVE AI
Adobe has restricted employees' use of external AI tools, urging them to use in-house offerings instead, according to an internal email from CIO Cindy Stoddard. This move mirrors similar policies at Amazon, Alphabet, Apple, and Samsung, indicating a broader trend in data protection and promotion of proprietary tech.
2023.07.07
A HUMAN-CENTRIC APPROACH TO ADOPTING AI
This article discusses the importance of taking a human-centric approach when implementing artificial intelligence (AI) systems. It emphasizes the need for ethical considerations, transparency, and collaboration between humans and AI to ensure that AI technologies are adopted responsibly and effectively.
2023.07.07
FULL-BODY AI SCANS COULD BE THE FUTURE OF PREVENTIVE MEDICINE
CNET explores the potential of full-body AI scans in revolutionizing preventive medicine. It highlights how AI-powered imaging technologies can detect early signs of diseases and provide personalized health insights, leading to more proactive and targeted healthcare interventions.
2023.07.07
OPENAI LAUNCHES SUPERALIGNMENT TEAM
How do you align a model much smarter than humans? OpenAI has been working on this question across teams for some time, but now they're bringing 20% of the entire org (compute-wise) to bear on this challenge. They're hiring extensively and many top scientists internally are working on this problem.
2023.07.06
PLAYGROUND AI RAISES $40M FOR IMAGE EDITING
The popular AI powered image editing suite raised money to continue pushing forward text driven creativity expressed via pixels.
2023.07.06
1 BILLION TOKENS IN CONTEXT
It seems like dilated attention now finally works! This is an exciting advancement if it holds at scale, because it means that we can essentially treat entire corpi of documents as context. All of the Python code on GitHub is only a few hundred billion tokens. LongNet is a linear attention variant, but they seem to have stabilized training, and it doesn’t sacrifice performance on shorter sequences.
2023.07.06
ENHANCED VISION-LANGUAGE ZERO-SHOT LEARNING
This study presents a technique that improves AI's understanding of image-text tasks by focusing on fine details and keywords, enhancing performance, especially when faced with unseen scenarios.
2023.07.06
KOSMOS-2 CHECKPOINTS RELEASED
One of the top multimodal models today was developed by the UniLM team at Microsoft. They have now released demo code, weight checkpoints, and more for you to play with and integrate into your systems.
2023.07.06
NEW VECTOR DATABASE, WRITTEN WITH METAL SHADERS
There's a lot of excitement around using vector databases to augment the memory of different language models. This vector database has a simple interface and is primarily used for ChatGPT plugins. It runs many times faster than competitors while maintaining simplicity. It doesn't have the bells and whistles of many other offerings, but it performs the task well for plugins.
2023.07.06
EFFICIENT NIGHTTIME UAV TRACKING
This research introduces SAM-DA, a new system that improves real-time tracking of unmanned aerial vehicles (UAVs) at night. By using the Segment Anything Model (SAM) to generate detailed training samples from limited nighttime images, it significantly enhances the tracking ability while using fewer resources.
2023.07.06
WEBGLM: A COST-EFFECTIVE ONLINE Q&A TOOL
WebGLM is a new online tool powered by artificial intelligence designed to answer questions by searching the web and providing the most relevant information. It improves upon previous models by offering increased accuracy and efficiency, and studies show that it performs as well as, if not better than, more complex systems.
2023.07.06
PLAYING WITH STREAMLIT AND LLMS
If you’re doing anything with generative AI and playing internally with your own LLMs, you should check out this post about Streamlit.
2023.07.06
CELEBRITIES USE AI TO TAKE CONTROL OF THEIR OWN IMAGES
Celebrities are partnering with brands to create AI-generated duplicates of themselves that can be used in marketing campaigns. This gives celebrities more control over their own images and allows brands to use the stars in ways they never could before.
2023.07.06
NANOT5
We’ve covered this in previous newsletters, but it just got an upgrade. Inspired by Andrej’s great NanoGPT repo, the NanoT5 repo can train a model to equivalent T5 paper performance 150x faster. It’s definitely worth checking out if you care about training models from scratch.
2023.07.06
THE MANY WAYS THAT DIGITAL MINDS CAN KNOW
Ryan Moulton discusses the various methods through which digital minds can acquire knowledge, such as direct programming, machine learning, simulation, and training.
2023.07.06
AI WEIGHTS ARE NOT OPEN “SOURCE”
The article explores the issue of AI model weights and their availability as open-source. It argues that while the source code of AI models may be open, the weights—which contain the actual learned knowledge—are typically not shared openly due to various reasons such as intellectual property concerns, privacy, and commercial interests.
2023.07.06
WHAT THE EUROPEAN AI ACT MEANS FOR YOU, AI DEVELOPER
The EU AI Act is the world's first legislation on artificial intelligence, similar to GDPR but for AI. It introduces definitions for 'foundation models' and 'general-purpose AI systems' and sets out dos and don'ts for AI practices. The Act also focuses on trustworthy AI development and includes obligations for high-risk AI systems. The legislation is currently going through final negotiations before enforcement.
2023.07.05
WHY TRANSFORMATIVE AI IS REALLY, REALLY HARD TO ACHIEVE
While humans have a good track record of innovation, the trend growth rate of GDP per capita has never exceeded three percent per year. There has yet to be a technology that breaks the three percent threshold. Only strong evidence should cause us to expect anything different. AI is likely transformative in the same way that the internet was, but it has many hurdles in the way to the accelerating growth rates predicted by some.
2023.07.05
OPENAI TURNS OFF CHATGPT BROWSING FEATURE
Despite complaints from ChatGPT Plus subscribers who benefited from the feature, OpenAI has temporarily disabled "Browse with Bing" on its ChatGPT service, citing concerns that the AI was inadvertently bypassing paywalls and privacy settings. OpenAI insists this move was made to respect content ownership rights, with a return of the browsing feature planned after necessary adjustments.
2023.07.05
MACHINE LEARNING FOR POLYNOMIAL EQUATIONS
Solving systems of polynomial equations sometimes results in singularities which are poorly behaved points. There are certain resolution strategies which allow for solutions at these points. Resolution is a challenging process, but if you frame it as a game and apply RL, you can achieve impressive performance.
2023.07.05
A FASTER, SMARTER WAY TO INPAINT PICTURES
Imagine having a tool that can fill in missing parts of a picture quickly and accurately - that's what the WavePaint model does! It uses something called a WaveMix-based architecture to mend images, working faster and needing fewer steps than current methods.
2023.07.05
PREDICTING SUICIDE RISK: A MULTI-TASK LEARNING APPROACH
This research introduces a new model to predict suicide risk in people with bipolar disorder. The model uses symptoms found in these posts to understand which are most influential in predicting future suicide risk, helping doctors to better understand these patients and provide timely help.
2023.07.05
S.A.T.U.R.D.A.Y
S.A.T.U.R.D.A.Y provides tools to build elegant vocal interfaces to modern LLMs.
2023.07.05
AIDER
Aider is a command-line chat tool that allows you to write and edit code with OpenAI's GPT models.
2023.07.05
TALES OF TAKEOVER IN CCF-WORLD
Scott Alexander explores the different scenarios of a post-AI takeoff world.
2023.07.05
AI, OZYMANDIAS
In the late 19th century, there was a widespread belief that human knowledge was nearing completion. This was due to rapid progress in science and technology, particularly in physics and biology. However, this confidence was misplaced, as new discoveries soon overturned many of the old assumptions. Freddie DeBoer argues that we are currently seeing a similar recurrence of hubris in AI. While AI has achieved impressive results in recent years, it is important to remember that it is still a very young field with much that we do not understand. We should be wary of making grandiose claims about AI's potential, as we may eventually be proven wrong in the same way that the late Victorians were.
2023.07.05
HYPERDIMENSIONAL COMPUTING REIMAGINES AI
A new approach to computation called hypervectors is being explored as a way to improve artificial intelligence. Hypervectors are large vectors that represent concepts. They can be used to reason about the world in a more abstract and efficient way than traditional methods.
2023.07.05
SOFTBANK IS PLAYING CATCHUP WITH AI
SoftBank is attempting to catch up to the current AI hype wave, despite already investing $140B into AI startups.
2023.07.05
LEARNING THEORY BOOK
One of the most prolific learning theorists is writing a book. It covers many of the important topics of research in modern ML learning theory. A preprint PDF is available in the link.
2023.07.05
A COMPREHENSIVE CODE-BASED FOR COMPUTER VISION
Scenic is a codebase with a focus on research around attention-based models for computer vision.
2023.07.05
INSIGHTS FROM GLOBAL CONVERSATIONS
OpenAI discusses the insights gained from their global conversations initiative, which aimed to gather public input on AI-related topics.
2023.07.04
RUNWAYML RAISES $141M SERIES C EXTENSION
This short announcement shows that RunwayML is still dedicated to building creative tools. Much of it will go to fund internal R&D.
2023.07.04
CLASSIFIER-FREE GUIDANCE FOR LANGUAGE MODELS
If you're looking for interesting trends in AI today, the idea of decoding tokens in new ways is an avenue of active research. Currently, each model gives the same amount of compute to each token, and the sampling method is somewhat simple. If we can improve sampling and guidance, we may be able to dramatically improve performance without retraining. This paper suggests that we can double our effective model size just by using classifier free guidance, which was first explored in the context of text to image diffusion models.
2023.07.04
A NEW FRAMEWORK TO DETECT ANY DEEPFAKES
This research presents a new system called Detect Any Deepfakes (DADF) for improved detection and precise location identification of manipulated facial images (deepfakes). Using the Segment Anything Model (SAM) and incorporating a multiscale adapter and Reconstruction Guided Attention (RGA), DADF improves deepfake detection performance.
2023.07.04
HUMAN BRAIN ACTIVITY
Researchers have proposed a new method using a diffusion model, specifically a latent diffusion model (LDM) named Stable Diffusion, for reconstructing high-resolution, high-fidelity images from human brain activity captured via fMRI. The method, which reduces computational cost while maintaining generative performance, does not require extra training or fine-tuning of complex deep learning models. The study also gives a neuroscientific interpretation of different LDM components, presenting a promising new approach to understand how our brain represents visual experiences.
2023.07.04
CREATING 3D SHAPES FROM 2D IMAGES AND TEXTS
This project introduces a method to generate 3D shapes from 2D images or texts, reducing inconsistencies that result from the dimensional differences. The method uses a two-model framework, the Shape-Image-Text-Aligned Variational Auto-Encoder (SITA-VAE) and Aligned Shape Latent Diffusion Model (ASLDM), to map the 3D shapes into a space that aligns with images and texts.
2023.07.04
A NEW METHOD FOR MORE REALISTIC 3D MOVEMENT ANALYSIS
The research introduces a new method that improves the understanding of 3D movement in relation to the environment, offering more realistic poses in contrast to existing techniques that produce implausible results.
2023.07.04
SEAMLESS VIDEO GENERATION WITH TEXT
This paper presents a new method for generating consistent, high-quality videos from text. It involves creating key frames using an adapted image model, and then spreading these styles across the video with special matching and blending techniques, resulting in videos that look great and flow smoothly.
2023.07.04
AI AND THE AUTOMATION OF WORK
An exploration into how AI will affect work, specifically whether it will automate away the need for humans.
2023.07.04
OPENAI’S CODE INTERPRETER IS ABOUT TO REMAKE FINANCE
This article details a breakthrough in artificial intelligence automation for accounting minutiae due to a ChatGPT plug-in called Code Interpreter and how it could potentially revolutionize the way finance knowledge work is completed.
2023.07.04
TECH WORKERS ARE SCRAMBLING TO BECOME AI EXPERTS
As AI continues to grow in importance, tech workers are scrambling to learn new skills in order to stay relevant in the job market. Many are turning to boot camps, online courses, and self-education in order to gain the AI skills that employers are increasingly seeking.
2023.07.04