- AI Breakfast
- Posts
- New Windows AI Feature Records Everything You Do On Your PC
New Windows AI Feature Records Everything You Do On Your PC
AI Breakfast
May 22, 2024
Good morning. It’s Wednesday, May 22nd.
Did you know: Laszlo Hanyecz, a programmer and early Bitcoin enthusiast, made history by paying 10,000 Bitcoins for two Papa John's pizzas. At the time, this was about $41. As of today, 10,000 Bitcoin is worth roughly $700M.
In today’s email:
Microsoft To Record Everything You Do
AI Surpasses Humans on Theory-of-Mind
Stories you may have missed
10 New AI Tools
Latest AI Research Papers
You read. We listen. Let us know what you think by replying to this email.
In partnership with
Transform How You Work and Organize Information
Flot is your all-in-one AI assistant, enhancing your writing, speeding up your reading, and helping you remember information across websites and apps.（Available on Windows and macOS）
What Flot Brings:
Seamless integration with any website and app
Comprehensive writing and reading support:
Improve writing
Rewrite content
Grammar check
Translate
Summarize
Adjust tone
Customized Prompts
Save, search and recall information with one click
Subscribe to Oncely LTD & Flot AI Newsletters by clicking.
Today’s trending AI news stories
New Windows AI feature records everything you’ve done on your PC
Microsoft's newly released Recall feature for Copilot+ PCs captures snapshots of user activities, encrypting and storing them locally for easy retrieval. Unlike a cloud-based solution, Recall captures encrypted snapshots of your screen activity directly on your device's Qualcomm Snapdragon X Elite chip (with NPU) – a response for those wary of the cloud.
While this feature promises enhanced productivity, it raises significant privacy concerns. Despite Microsoft's assurances of encryption and local storage, the potential for misuse remains, particularly if someone gains access to your account. Recall requires a hefty 256GB of storage and is currently in beta, with Microsoft seeking user feedback to polish this potentially game-changing, or privacy-invasive (depending on your perspective) feature before full release. Read more.
AI Outperforms Humans in Theory-of-Mind Tests
Recent research reveals that large language models like GPT-4 exhibit remarkable proficiency in theory of mind tasks, surpassing expectations. Theory of mind, crucial for understanding others' mental states, is adeptly mimicked by LLMs such as ChatGPT.
This "theory of mind," crucial for understanding others' intentions, was thought to be beyond the grasp of machines. The study found LLMs like ChatGPT performing well on tasks designed to test this ability, including interpreting irony and deciphering strange stories. GPT-4 even outperformed humans in these areas.
However, some remain skeptical, questioning whether LLMs truly understand human thought or are simply mimicking patterns. Concerns linger about potential biases and anthropomorphization of LLMs. The study doesn't claim LLMs possess theory of mind, but it highlights their ability to mimic human-like reasoning with uncanny accuracy. Read more.
OpenAI is ‘in conversations’ with Scarlett Johansson over the ChatGPT voice that sounds just like her: A recent dustup between OpenAI and actress Scarlett Johansson highlights the challenges of voice selection in chatbots. Johansson raised concerns about the uncanny resemblance between her voice and "Sky," a voice option within OpenAI's ChatGPT. While Johansson maintains she declined an offer to voice ChatGPT 4.0, OpenAI denies intentionally replicating her voice and apologizes for any confusion. Sky, drawing comparisons to Johansson's character in "Her," was subsequently removed by OpenAI. Read more.
US FDA clears Neuralink's brain chip implant in second patient, WSJ reports: This follows a hiccup during the first procedure, where tiny wires proved a bit too eager for a retreat – a challenge previously encountered in animal testing. Neuralink is addressing this by upping the ante on wire placement depth. Read More.
Microsoft makes Phi-3 generally available, previews its Phi-3-vision multimodal small language model: Microsoft Research has announced the general availability of its Phi-3 family of lightweight language models. This release encompasses three models: Phi-3-mini (3.8 billion parameters), Phi-3-small (7 billion parameters), and Phi-3-medium (14 billion parameters). All models are designed for cost-effective deployment while maintaining strong reasoning capabilities. The launch includes Phi-3-vision, a 4.2 billion parameter multimodal variant capable of visual reasoning tasks such as analyzing charts and graphs. Read more.
Etcetera: Stories you may have missed
10 new AI-powered tools from around the web
Smartli simplifies e-commerce content creation with AI. Generate captivating copy, blogs, and ads effortlessly. Tailor content to enhance your brand across multiple platforms.
Becca generates LinkedIn posts that match your tone using AI. It analyzes existing posts and niches, sending engaging content via WhatsApp.
Brand24 is an AI-powered social listening tool, monitors and analyzes online brand mentions, providing insights on reputation, sentiment, and market trends for businesses.
MemoryLane is a reminiscence therapy tool that utilizes AI to transcribe and curate life stories shared in English or Chinese, producing personalized, professionally printed books.
Stable Artisan is a Discord bot on the Stability AI Platform API that utilizes Stable Diffusion 3 for generating and editing high-quality images from natural language prompts.
Octoverse provides AI agents that enhance user interactions in apps by understanding and completing tasks efficiently, outperforming GPT-4o in various use cases.
Neolocus is an AI interior designer that allows users to design and furnish rooms by uploading their space, facilitating easy and personalized home styling.
Native AI integrates various business apps into a single interface, enhancing productivity through seamless interaction and automation, reducing the need for context switching.
Universal-1 by AssemblyAI is a multilingual speech AI model trained on 12.5 million hours of audio, offering top-tier speech-to-text accuracy and reduced errors.
summate.io is an AI-powered tool that condenses newsletter content into personalized summary digests, saving time by summarizing long articles and videos.
arXiv is a free online library where researchers share pre-publication papers.
What? No comic?
ChatGPT is experiencing outages today.
Check back on Friday :)
Thank you for reading today’s edition.
Your feedback is valuable. Respond to this email and tell us how you think we could add more value to this newsletter.
Interested in reaching smart readers like you? To become an AI Breakfast sponsor, apply here.