Full-text search across article titles and summaries. Refine with sector, year, content type, and reading-time filters.
Researchers have demonstrated that OpenAI's GPT-4 model can autonomously exploit security vulnerabilities detailed in CVE advisories with an 87% success rate, far outperforming other models and tools like vulnerability scanners.
50 vision/language datasets combined into a single format to allow for improved training of models.
🆓 Get FREE access [4] to the third Semantic Layer Summit
— a one-day virtual event hosted by AtScale.
Sponsored
LAST CHANCE TO JOIN: On April 24th, top enterprise leaders and
Meta has released an 8B and 70B model with dramatically improved performance, particularly in reasoning, context length, and code. It is still training a 400B parameter model, which will match Opus in performance. These models are easily the most powerful available open models.
Mathematicians and Google's DeepMind researchers have utilized AI to find large collections of objects that lack specific patterns, assisting in understanding potential catastrophic failures like internet severing due to server outages. Their approach employs large language models to iteratively generate and refine set-free collections, facilitating the study of worst-case scenarios. This research reflects the combined power of AI and human ingenuity in tackling complex problems.
The launch of OpenAI's DALL-E 2 in April 2022 marked a groundbreaking and tumultuous period in AI history, as a tight-knit group of artists and tech enthusiasts explored the intersection between language and visual arts using the technology. However, the amazement and exhilaration soon gave way to concerns about the ethics of training AI models on copyrighted creative work without permission or compensation, leading to a polarizing debate that continues to reverberate in the AI space as OpenAI moves on to DALL-E 3 and other AI image synthesis models emerge.
Researchers have developed a new method called Federated Proxy Fine-Tuning (FedPFT) that improves the adaptation of foundation models for specific tasks while preserving data privacy.
This paper introduces a new approach to enhancing In-Context Learning (ICL) in large language models like Llama-2 and GPT-J. Its authors present a new optimization method that refines what they call 'state vectors' — compressed representations of the model's knowledge.
Sponsored
Whether you're building customer-facing or internal apps, the high cost of paid genAI tools [11] can be a roadblock. Open-source LLMs can be dramatically cheaper; they're also better for compliance and local training on proprietary data. This blog post by AgileEngine [11] reviews the leading open-source LLMs, highlighting their
Component modeling breaks down a model's prediction process into its basic elements, like convolution filters and attention heads, to understand their specific contributions to the final output.
AI Gateway is an interface between apps and hosted large language models. It streamlines API requests to LLM providers using a unified API. AI Gateway is fast, with a tiny footprint, and it can load balance across multiple models, providers, and keys. It has fallbacks to ensure app resiliency and supports plug-in middleware as needed.
Moondream is a tiny vision language model trained on top of Phi-2. It has enormously powerful performance for its size, although it still struggles with some hallucination. Moondream is small enough to run on a phone and on edge devices. It enables a myriad of commercial vision capabilities.
The emergence of sophisticated AIs is challenging fundamental notions of what it means to be human and pushing us to explore how we embody true understanding and agency across a spectrum of intelligent beings. To navigate this new landscape, we must develop principled frameworks for scaling our moral concern to the essential qualities of being, recognize the similarities and differences among various forms of intelligence, and cultivate mutually beneficial relationships between radically different entities.
This guide explores using open-source embedding models to enhance AI projects. It covers criteria for model selection and methods for effective deployment. The guide utilizes Sentence Transformers, an open-source library, for practical examples.
NVIDIA's dominance in the AI space continues to be secured not just by hardware, but by its CUDA software ecosystem and proprietary interconnects. Alternatives like AMD's ROCM struggle to match CUDA's ease of use and performance optimization, ensuring NVIDIA's GPUs remain the preferred choice for AI workloads. Investments in the CUDA ecosystem and community education solidify NVIDIA's stronghold in AI compute.
Sponsored
Create tons of fun, shareable AI videos with 1 selfie! Choose from a wide range of templates or create your own masterpiece with text to
Stability AI has laid off about 20 workers.
A private, personalized, and protected AI coding assistant.
The Humane AI Pin has many faults, but the biggest letdown may be its translation feature.
Sponsored
The #1 concern around GenAI today is accuracy - how can enterprises scale a technology they can't trust? Narratize is cracking the code of GenAI accuracy for every innovative brand. Narratize is built to deliver accuracy, always:
There have been reports of the open source pioneer Mistral seeking several hundred million dollars of funding to train more models.
Stability AI has made its latest text-to-image AI model, Stable Diffusion 3, available to some developers via API and its new content creation platform called Stable Assistant Beta. The model is still in preview and not yet available to the general public.
Sanctuary AI will deliver its humanoid robot, Phoenix, to a Magna auto manufacturing facility, following Magna's 2021 investment in the company. Phoenix is notable for its walking capability and dexterous hands aimed at enhancing manufacturing efficiency. The specifics of the pilot, including the number of robots and deployment details, remain undisclosed.
The new base Mistral report was released today. Additionally, their own instruction tuned model was released under a similar permissive license. They report a strong MMLU and HumanEval performance along with incredible multilingual English, French, Italian, German, and Spanish performance, function calling, and 64k native context length.
The latest completely open 7B language model had an upgrade with a much-improved reasoning score on MMLU. Its development team found that changing the data mixture had a substantial effect on performance. They have provided the exact statistics of new data sources and percentages that led to this improvement.
This paper describes Stability AI's diffusion transformer model for audio synthesis.
Sponsored
The third annual Semantic Layer Summit [14] is less than a week away! This free, one-day virtual event features some of the top voices in enterprise analytics, BI, and genAI — including leading technical leaders at Mode, Netflix, T-Mobile, and AtScale.
A deep dive interview from industry experts on the challenges and solutions of getting AI models into production and where MLOps differs from traditional engineering. They discuss why so few ML projects make it to production and how to focus as an organization to actually launch.
The Effort library allows for real-time adjustment of the number of calculations performed during inference of an LLM model, resulting in significant speed improvements while retaining most of the quality. Despite some implementation overhead, initial results suggest that the Effort library has the potential to significantly improve LLM inference speed while maintaining quality. The author encourages others to test the 0.0.1B version and provide feedback to further improve the library.
This study revisits using diffusion models for image-to-text generation. It introduces the LaDiC architecture, which enhances diffusion models' performance in image captioning tasks.
Wayve's Lingo-2 is a VLM train in simulation for autonomous driving tasks. It performs driving actions based on video input. Lingo-2 includes descriptions of reasoning for the driving actions it takes based on the scene.
Researchers are developing alternative architectures to address the limitations of Transformers in long-context learning, generation, and inference speed, showing competitive quality at smaller scales but uncertain scalability. The rapid progress in this subfield suggests that the Pareto frontier will continue to expand, enabling longer context modeling and greater inference throughput, ultimately increasing the number of use cases for AI.
Building a successful AI company requires navigating complex short, mid, and long-term dynamics and maintaining elite speed and execution, owning more of the stack, gathering unique data, and leveraging synthetic data generation. As the AI industry matures, companies must adapt to shifting talent dynamics and understand the machine they are building and the axes of competition they are predicated on to create lasting moats and stand out amidst the noise.
Verifying training when sharding across devices can be challenging - this handy snippet verifies that things have been done and updated properly.
Luminal is a deep learning library that uses composable compilers to achieve high performance.
Google Maps will use AI to summarize directions to EV chargers as well as provide reliability and wait times.
Sponsored
If you want to build AI applications with access to the web, you need to check out the Brave Search API [4]. With a single API call, you'll get relevant search results from an billions of webpages [4] — high-quality, real-time data you can build on.
Apple's upcoming AI features in iOS 18 are rumored to focus on privacy, with the initial set of enhancements functioning entirely on-device without the need for an internet connection or cloud-based processing, thanks to the company's in-house large language model known internally as "Ajax."
The Stanford Institute for Human-Centered AI has released its seventh annual AI Index report. This year's report covers the rise of multimodal foundation models, major cash investments into generative AI, new performance benchmarks, shifting global opinions, and new major regulations.
Google researchers have introduced Infini-attention, a technique that enables LLMs to work with text of infinite length while keeping memory and compute requirements constant.
Most modern AI is built around the idea of compressing a training dataset into a model. The better the compression, the better the model. This paper shows that relation rigorously and posits that scale benchmark scores correlate strongly to a model's ability to compress novel text.
TransformerFAM provides a feedback mechanism that allows Transformers to attend to their own latent representations. This can, in theory, introduce recurrence into the model for processing extremely long inputs in context.
Another long context paper - this time, a new architecture that uses two novel weight updating schemes. It outperforms Llama 2 on the same number of training tokens 2T. It also scales to infinite context length at inference time.