Chat w/ AI Silently; Microsoft Embraces Anthropic; OpenAI Says 🤷🏼♂️
🤫 Alterego’s “Near-Telepathic” Wearable Reads Silent Speech
Boston startup Alterego (an MIT Media Lab spinout) unveiled a “near-telepathic” headset that tracks micro-movements in your jaw and throat to decode silent speech at the speed of thought.
In the demo, founder Arnav Kapur codes, texts, runs visual queries, and even “speaks” directly to other wearers using a system the team calls Silent Sense. The company says it works in noisy environments, handles multilingual speech, and can read both mouthed and motionless intent.
The project dates to 2018 research at MIT; the for-profit spinoff is new, and there’s no public availability date yet. Expect to see this and tech like it to close the gap to allowing people to access knowledge sans brain retention.
Why it’s spicy: This looks like BCI-style control without surgery - a potential new interface for AI agents and for accessibility. The big questions now are latency, accuracy, privacy, and whether “thought-to-text” crosses social lines the way AirPods once did (remember “are you talking to me?”).
Read more: The Rundown AI
📊 Microsoft Eyes Anthropic’s Claude for Office 365 (Alongside OpenAI)
The Information reports Microsoft is in talks to add Anthropic’s Claude to Office 365, alongside OpenAI’s models - a notable diversification move. Internally, Microsoft reportedly found Claude Sonnet 4 stronger for spreadsheets and presentations than GPT-5.
Anthropic just rolled out new file creation features (see below), and Microsoft would likely tap Claude via AWS, paying a cloud rival even while it has home-field access to OpenAI.
Why it matters: Publicly, Microsoft and OpenAI are buddy-buddy; privately, Redmond is hedging - shipping its own models, now potentially partnering with a competitor in productivity. Customers win when stacks are multi-model and task-routed; the interesting part is procurement and data governance when your office suite blends OpenAI + Anthropic under one roof.
Read more: The Rundown AI
📈 Claude Can Now Create Docs, Slides, Sheets, PDFs
Anthropic shipped file creation in Claude: in a private compute sandbox, Claude can write code, generate spreadsheets, compose Word-style docs, build slide decks, and export PDFs - plus convert formats (e.g., a PDF → slides).
It’s rolling out to Max, Team and Enterprise now (Pro “in the coming weeks”). Anthropic urges “start simple” and mind data-access risks when enabling web actions.
This pushes AI from “assistant in the chat” to “co-author in the file system” - and explains why Microsoft is exploring Claude for O365 workloads that favor structure, formulas, and layouts.
Read more: The Rundown AI, AI Secret
🧪 OpenAI’s “Cure” for Hallucinations: Reward “I Don’t Know”
A joint OpenAI + Georgia Tech paper argues hallucinations aren’t bugs so much as statistical pressure from benchmarks that penalize abstention. (Say that three times fast.)
The proposed fix: re-grade models so “I don’t know” gets credit when appropriate, with confidence calibration baked into evals. That would ripple into MMLU/GPQA/SWE-bench culture and force product teams to budget for calibration pipelines, not just bigger models.
The hard part isn’t technical - it’s incentives. The whole ecosystem loves leaderboard wins; “IDK” honesty dings scores. But if adopted, we could see dramatic reductions in overconfident wrongness in enterprise apps.
Read more: AI Secret
🍎 Apple “Ghosts” AI at iPhone 17 Event
Yesterday’s iPhone 17 keynote emphasized battery, silicon, watch sensors, and background neural engines, with muted “Apple Intelligence” chatter. Meanwhile, competitors (Google, Samsung) keep pushing agentic assistants to the front of house.
If you’re betting your product roadmap on Apple as an AI platform this cycle, consider a parallel track - Apple seems to be hedging while it tightens the on-device stack.
Read more: AI Secret
🎨 Tutorial: Create “Miniature Product” Photos With Nano Banana
Rundown’s (University) tutorial walks through using Nano Banana to generate luxury “mini” product shots - think Cartier watches or a Birkin - pinched between fingers, with studio lighting, shallow depth of field and soft shadows.
You can work through Gemini’s Create Image, Freepik or Adobe Firefly, then drop the assets straight into your e-commerce flow.
Pro tip from the guide: add a scale disclaimer so customers don’t think the product is toy-sized.
Read more: The Rundown AI
⚡ Quick Hits
- Veo 3: Costs cut ~50%, adds HD and vertical formats - friendlier for creators.
- Anthropic confirms a recent response-quality dip, denies intentional degradation.
- Baidu ERNIE X1.1: reasoning upgrade with fewer hallucinations; Alibaba Qwen3-ASR ships robust multilingual ASR.
- Eli Lilly TuneLab: AI drug-discovery platform trained on $1B of proprietary data.
- Google Veo 3 / 3 Fast: adds vertical video, 1080p, ~50% price cut.
Read more: The Rundown AI, AI Secret, The Neuron
🛠 Tools & Launches (Expanded)
🧠 Unframe
Enterprise platform to scale AI securely, pitched at bespoke setups and governance. (Listed as a Trending AI Tool today.)
📓 NotebookLM
Adds flashcards, quizzes, redesigned reports and interactive study flows. Education teams get a tidy on-ramp to AI-assisted study materials.
🎨 Ideogram Styles
New presets for consistent aesthetics and stylized typography - useful for brand kits and social creative.
🖼️ HunyuanImage 2.1
Tencent open-sources an upgraded text-to-image model with better realism, prompt following, and text rendering.
🧑🏻💻 Biela.dev
Turns vague ideas → precise instructions → working apps; a generator that forces definition before code.
💼 Macky
An AI business consultant that returns actionable insights to complex prompts - positioned for SMB strategy.
📖 Fableoo
Guided, choose-your-own-adventure storytelling for kids; creativity sandbox with AI scaffolding.
📄 OdysseyGPT
Converts static documents → interactive knowledge bases so you can query your PDFs and notes.
📱 Dreamflow
Combines code editing + visual design + AI for cross-platform app development.
📊 Online Gantt
Generates and edits Gantt charts from prompts or meeting notes; ditch drag-and-drop for LLM ops.
😜 Crazy Face AI
Converts stiff selfies → expressive portraits with controllable edits; social-friendly fun. (Note: Implies free, but after grabbing your email hits you with a charge.)
🎬 Animon
Transforms images into anime-style art and short animations - filters plus frame generation.
🚀 FlowPost
Automates brand-tuned social posts across channels; scheduling + generation in one.
🌟 VoiceType
Voice-first writing tool (“9× faster” claim) for drafting and editing by speech.
🎓 Resea AI
An academic research agent for end-to-end workflows (lit review → analysis).
🤖 Momen
No-code builder for fully customizable, production-ready web apps.
💡 Community Corner
Language-Learning Workflow - Teacher Chiara M. uses YouTube clips + transcripts, then ElevenLabs for clearer audio at matched speed; NotebookLM turns class notes into audio overviews in either native or target language. It’s a lightweight stack that makes listening drills less frustrating and more bingeable.
Read more: The Rundown AI
Today’s Sources:
AI Secret
The Rundown AI
The Neuron