A.I. Tracker 🤖

OpenAI introduces the o1 reasoning AI model to its API for tier 5 developers. They also release new versions of GPT-4o models and add capabilities to the Realtime API, including WebRTC integration and preference fine-tuning.

OpenAI is hosting "12 Days of OpenAI" reveals including ChatGPT Pro, reasoning model updates, Sora release, Canvas, Apple Intelligence integration, and real-time video capabilities. Watch live on YouTube or follow updates on their blog.

Samsung teases new software feature on Galaxy S25 series called Now Bar, offering personalized data and AI-like functionality. Focus on personalization and privacy, set to launch in One UI 7 in early 2025.

Google launched "Whisk," an AI tool that combines uploaded images into one AI-generated image without text input. Whisk is a creative tool for inspiration and rapid visual exploration. It builds on Google's Gemini AI and DeepMind's Imagen 3. OpenAI also released a text-to-video generator called Sora, showing competition in consumer AI products.

Google updates its terms to allow customers to use its generative AI for automated decisions in high-risk areas like healthcare, as long as a human is involved. Regulators are concerned about bias in AI decision-making, prompting stricter rules and oversight measures.

The U.K. government is consulting on implementing an opt-out copyright regime for AI training, requiring rights holders to take action if they don't want their intellectual property used. The goal is to balance the interests of the creative and AI industries, prioritizing transparency and remuneration for rights holders.

NVIDIA has unveiled the Jetson Orin Nano Super Developer Kit, a compact generative AI supercomputer with improved performance. It can reach 68 TOPS, offers software updates for its predecessor, and is available for just $249.

Moonvalley and Asteria have developed the Marey model, a video generation AI trained on ethically sourced data for Hollywood productions. The model offers creative control features and aims to provide a clean alternative to models trained on scraped data. VIP+ Special Report on the rise of generative AI in the entertainment industry.

UK Ministry of Defence AI tool hosted by Amazon for recruitment poses risk of identifying defense personnel due to stored data. Government assures low risk with safeguards in place for public sector AI tools, aiming to boost transparency and trust.

Microsoft's AI-powered "Recall" tool, designed to take snapshots of screen activity, is still capturing sensitive data despite a new "filter sensitive information" setting. Even after a revamp, the tool has been found to capture credit card numbers and personal data, raising privacy concerns. Microsoft initially reversed the tool's launch due to backlash, but it is now being tested with Insiders. Critics argue that the privacy measures are still lacking.

Dexcom announced AI feature for Stelo CGM, giving personalized insights on glucose levels. Uses Google Cloud AI platform for personalized reports and potential real-time feedback, aiming to improve user experience and engagement.

UK government proposes opt-out scheme for creatives allowing tech firms to use copyrighted material for AI training. Creative professionals can reserve rights or demand license fees. Consultation seeks views on protecting celebrities' likeness from AI replication.

The UK is considering regulations to control copyrighted content used by tech companies for training AI models. This comes after legal disputes and concerns over rights infringement by companies like OpenAI and Google. The government is planning consultations to clarify the use of intellectual property and ensure fair compensation for creators.

Google introduces Whisk, an AI image generation tool based on Imagen 3 model, allowing users to create images using other images as prompts. It offers a fast-paced exploration experience, including editing options and a My Library feature to save creations. Whisk also provides text prompts for more control.

Data centers in the US have tripled carbon emissions since 2018, with AI's growing energy demands exacerbating the issue. AI companies are turning to nuclear energy, but face challenges. Anduril showcases AI-powered military tech for future conflicts. Bluesky deals with impersonators, tech elite donate to Trump, AI-generated movies premiere, Meta challenges OpenAI's nonprofit status, and Silicon Valley's impact on democracy is examined in two books.

Cohere, a prominent AI startup, valued at $5.5 billion, is working with Palantir to deploy AI models for enterprise customers. Details revealed at a Palantir conference, showcasing potential use in Arabic language inference and strict data storage constraints. Palantir, known for defense work, may be using Cohere's AI for commercial clients as well.

Chinese AI models are gaining popularity and surpassing US models. Open-source LLMs like Qwen and DeepSeek-R1 are widely used globally. Despite restrictions, China invests in AI infrastructure independent of Nvidia chips.

Skechers faced backlash for using generative AI in a Vogue ad. The art was criticized for being low quality and not featuring the advertised shoes, leading to a negative consumer response.

Google's new AI tool, Whisk, generates images based on a user's prompt, capturing the essence rather than details. It is good for brainstorming but not for precise edits. The tool uses Gemini language model and Imagen 3 image generator, and is currently only available in the US.

YouTube is allowing creators to opt-in for third-party companies to use their videos to train AI models. The default setting is off, but creators can choose from a list of companies or allow all. This follows reports of big companies scraping YouTube content for AI training.