TLDR AI 2024-05-21

Building an AI product is hard. Engineers who understand AI are expensive and difficult to find. And there's no way of telling who's legit and who's not.

That's why companies around the world trust AE Studio. They'll help you craft and implement the optimal Al solution for your business with a team of world class Al experts.

Their development, design, and data science teams work closely with founders and executives to create custom software and Al solutions that get the job done for a fraction of the cost of searching and hiring your own team.

Book a free consultation session today. Get in touch now

Inflection AI has unveiled a new leadership led by co-founder Reid Hoffman. Hoffman claims that the startup has 18 months of runway left and that it will double down on building emotional intelligence within AI.

OpenAI paused the "Sky" voice mode for ChatGPT-4o after backlash for allegedly mimicking Scarlett Johansson's voice from the 2013 film Her. The actress is also pursuing legal action against the AI firm.

Google DeepMind introduced the Frontier Safety Framework to address risks posed by future advanced AI models. This framework identifies critical capability levels (CCLs) for potentially harmful AI capabilities, evaluates models against these CCLs, and applies mitigation strategies when thresholds are reached.

LeMeViT is a new approach to reduce computational costs in Vision Transformers by using learnable meta tokens. These tokens efficiently capture key information, improving inference speed.

Generating artistic 2D images is a fun and creative use of AI. This work brings that capability to 3D with Gaussian Splatting guided by textual input.

Understanding where things are in a 3D scene is challenging. By using language-guided 3D understanding, you can pinpoint semantic labels for objects in 3D space.

An open-source pair programmer.

The Gcore engineering team decided to see what happens if they ran an image classification model on their CDN infrastructure. This white paper has the full details of Gcore's experiment — tuning a WASM-at-the-edge environment for AI inference workloads. Bonus: there are pictures of cats. Get the white paper here.

It is easy to download language model weights and run them with a variety of tools. However, it is harder to know exactly what is going on in the code itself. This great tutorial shows every step of reconstructing Llama 3 and running the trained weights.

PSG-4D is a new representation that helps AI understand the 4D world of space and time.

Researchers have discovered a new security threat for Vision Transformers, a popular AI model. The attack, named SWARM, uses a "switch token" to secretly activate backdoor behavior in a model, making it highly stealthy and dangerous for users.

Meta's Llama 3-70B language model has surpassed rivals in the English Chatbot Arena, excelling at open-ended creative tasks but struggling with more technical prompts. Analysis reveals that Llama 3's win rate decreases as prompts become more challenging and that it is particularly strong with friendly, conversational responses. While Llama 3's approachability may contribute to its success, its overall competitive edge needs further study.

Perplexity AI's CEO Aravind Srinivas is unfazed by Google's AI-driven summaries, suggesting they could harm Google's ad-reliant business model. In this post, he critiques Google's frequent rebranding in response to AI competition while highlighting his company's user-focused approach to ads.

Stability AI has engaged in talks with potential buyers amid a cash crunch.

The SoundHound Chat AI assistant will leverage Perplexity to provide accurate, up-to-date responses to web-based queries that static LLMs cannot currently answer.

Google's AI chatbot Gemini remains unable to generate images of people, months after pausing the feature due to inaccuracy concerns and racial stereotyping.