Jerry Pratt, former CTO of Figure, has partnered with Nic Radford, former president and CEO of Nauticus Robotics, Inc., to found Persona AI, a new humanoid robotics firm in early development. Despite a saturated market, Persona aims to attract investors and 10 to 20 founder-level employees with a vision strongly resting on current AI breakthroughs in perception algorithms, onboard electronics, and machine learning. The startup distinguishes itself by choosing headquarters in Houston and Pensacola rather than traditional robotics hubs.
Great lecture from Sasha Rush, a professor at Cornell Tech and researcher at Hugging Face, about using diffusion as an objective for language modeling.
Kyutai, a French open research lab, has trained a pure audio LLM with minimal latency. It has managed to create a really impressive demo that will be open sourced in the coming months.
Researchers have developed a method to reconstruct 3D scenes from a single blurry image using neural radiance fields (NeRF) and event streams. This innovative approach models camera motion and synthesizes brightness changes, enabling high-quality, view-consistent images from blurry inputs without needing pre-computed camera poses.
MInference is a new method that accelerates the processing of long prompts in large language models. It uses sparse computation techniques to overcome the significant delays caused by traditional methods.
This AI-focused playbook outlines strategies for enterprise adoption of AI, highlighting the shift from human-performed services to software-driven workflows termed "Service-as-a-software." It delves into the implications for business models, such as performance-based pricing, and emphasizes the importance of workflow capture and AI accuracy for successful implementation. The guide also discusses threats like lateral attacks and stresses the need for AI to demonstrate clear performance, not just promise, in enterprise settings.
Enterprises shifting IT budgets toward AI proof-of-concept projects challenge customer support software companies like Salesforce and Zendesk. The growing integration of tools like ChatGPT in customer support has increased customer acquisition costs for traditional software providers, extending payback periods. The future investment in customer support software depends on these companies' innovation and the broader macroeconomic environment.
YouTube has updated its policies to allow individuals to request the takedown of AI-generated or synthetic content that mimics their voice or likeness.
Google's AI Overviews (AIO) feature now appears in only 8.71% of SERPs for 100K keywords, a significant decrease from 64% pre-rollout, with most AIOs linking to top 10 organic results. The length of AIO content and number of links have both increased post-rollout, indicating Google's emphasis on detailed answers and authoritative sources. SEO strategies must adapt to remain relevant in this evolving search landscape, where user queries with more words, lower search volumes, and lower CPC are more likely to trigger AI-generated responses.
Meta has trained a state-of-the-art 3D object generation model along with a PBR texture generation system. It uses the company's internal 2D image generation model as a tool for synthetic data generation.
This work allows for interactive text editing and controlling model generation at inference time. It allows for logical constraints in generation and achieves strong performance with small models.
Researchers have extended Pairwise Difference Learning (PDL), initially a regression technique, to the task of classification. Instead of predicting outcomes directly, PDL predicts differences between pairs of instances.
LLaRA is a framework that uses Large Language Models (LLMs) to improve robot action policy through conversation-style instruction-response pairs. By integrating visual inputs, these Vision Language Models (VLMs) process state information and generate optimal policy decisions.
Octomind abandoned the LangChain framework for AI test automation in favor of simpler, modular building blocks, which improved its codebase and team productivity. It encountered inflexibility with LangChain's high-level abstractions, complicating development and maintenance. By switching strategies, Octomind now enjoys a leaner architecture and faster iteration for its AI agent tasks.
Benjamin Bratton, the director of the Antikythera program at the Berggruen Institute and a professor at the University of California, San Diego, discusses the global reaction to artificial intelligence as a “Copernican Trauma," equating it with past shifts that have redefined humanity's self-perception. Bratton proposes five stages of "AI grief" — denial, anger, bargaining, depression, acceptance — to frame societal responses to AI's evolution, from skepticism to integration into our understanding of intelligence. He argues that the integration of AI reflects a broader biological and technological evolutionary process, rather than a distinctively human narrative.
Mustafa Suleyman, CEO of Microsoft AI, praised OpenAI's CEO Sam Altman and discussed the collaborative yet independent relationship between Microsoft and OpenAI in a recent interview at the Aspen Ideas Festival. Suleyman emphasized the need for a dialogue around AI safety and regulation rather than perceiving the development of AI as a race. He also expressed interest in AI cooperation with China and the potential role of AI as a teaching aid in classrooms.
Tech companies like Google and Meta are updating their privacy policies to use public and possibly private user data to train AI systems as they navigate a complex landscape of privacy laws and user consent. Backlashes have occurred as users and creators are fearful about the use of their content in training AI that could potentially replace them. The tensions highlight emerging challenges in AI development, data privacy rights, and the balance between innovation and ethics in the tech industry.
Figma AI is a new suite of AI-powered tools for Figma that helps designers with tasks such as visual search, asset search, text editing, image editing, prototyping, layer renaming, and design generation. These features are powered by third-party AI models and are free to use during the beta period.
This work treats web text as coming from a persona and then can condition on that persona to dramatically improve downstream task performance. The researchers find a jump of 20% points on MATH.
Researchers have developed a new process to improve Web2Code instruction tuning. It involves creating new webpage image-code pairs, refining existing webpage code generation data, creating new text question-answer pairs, and refining existing webpage understanding data.
Text-to-speech has improved by leaps and bounds in recent years. Can language data take it even further? Join experts from Oxford Languages [12] and WellSaid Labs for a live discussion [13] of what's next — AI voice generation, pronunciation data, text-to-speech linguistics challenges and more. Register (free) to submit questions to the
This repository introduces a new type of Transformer with a dramatically reduced KV cache size. It hasn't been proven at scale, but it can - in theory - match regular Transformer performance.
Unsupervised SAM (UnSAM) is a new model for whole-image segmentation that eliminates the need for human annotations. Using a divide-and-conquer approach, UnSAM identifies hierarchical structures in visual scenes by combining top-down and bottom-up clustering methods.
The future potential of LLMs reaching AGI through scaling alone is unlikely. While scaling has shown improvements in model capabilities, it primarily enhances perplexity rather than emergent abilities. Acquiring high-quality training data is becoming increasingly challenging.
Researchers have introduced a new anomaly detection problem that targets 'odd-looking' objects relative to others in a scene. Unlike traditional methods, anomalies here are scene-specific and identified through multiple viewpoints.
Discussion on the nature of AI bias has intensified, with critics arguing that the way models like Playground AI sometimes alter a user's ethnicity in images demonstrates biases in ML. However, some users deny this is a bug or relevant bias, citing examples where Asian features are overrepresented. The debate extends to broader implications of AI bias across industries. It is a complex issue with no simple resolution.