- AIdeations
- Posts
- AI Scent Teleportation, Spy Radar, and ChatGPT’s New Search Tool
AI Scent Teleportation, Spy Radar, and ChatGPT’s New Search Tool
From scent-sharing tech to AI-driven design and spy-proofing your phone, here’s the latest in groundbreaking AI innovations.

Osmo’s Scent Teleportation: The Wild Tech Bringing Smell to the Digital World
Quick Byte:
Imagine capturing the scent of a freshly sliced coconut in one room and recreating it on the other side of the world. That’s what Osmo’s Scent Teleportation is working on. It sounds like sci-fi, but they’ve already pulled it off in the lab. Osmo is taking smell, breaking it down with AI, and sending it across the globe—think “digital scent” in the same way we share photos or songs.
Key Takeaways:
How It Works: Osmo’s tech sniffs out scent molecules using GCMS (that’s Gas Chromatography-Mass Spectrometry if you’re curious), uploads the data to their Principal Odor Map (a.k.a. their AI’s “nose”), and then reassembles the scent with a Formulation Robot. Just pick a smell, analyze it, and boom—it’s teleported.
Why It Matters: This isn’t just a fun science experiment. With scent teleportation, we’re looking at a future where you could send a smell like you send a meme. Imagine retail, events, or even just smelling a mountain breeze from home.
The AI Powerhouse: Osmo’s built the largest scent database out there, training their AI to nail down even the subtlest molecules that give scents depth. Every lab test improves the accuracy, taking them closer to fully automated, error-free scent teleportation.
Practical Tips for Business Owners:
Get Ready for Immersive Marketing: When scent becomes shareable, businesses can create deeper sensory experiences. Scent teleportation could take immersive customer engagement to a new level—just imagine customers smelling your brand.
Futureproof Your Brand Experience: Retailers, travel brands, or event companies could all tap into this technology as it matures. Scent could become part of your brand experience as easily as a logo or a jingle.
Stay on Top of Sensory Tech: The sensory revolution is coming, and scent is only part of it. Knowing the latest in multi-sensory tech could give you the edge when these experiences become mainstream.
Bigger Picture:
Osmo’s Scent Teleportation is tech from the future, designed to bring the most primal of senses into the digital age. Sure, we’ve had sensory marketing with visuals and sounds, but scent? That’s a new level. Osmo has hit a major milestone, moving from lab demos to public trials, where you’ll soon be able to pick a flower and send the scent across the room.

Recraft V3: The AI Design Model That Actually Thinks in Design Language

Quick Byte:
Recraft’s latest AI, V3, is out here setting records. It’s not just generating images—it’s thinking in design terms. With control over text, style, and image positioning, V3 offers an AI-driven toolbox for designers who want control over every pixel. Whether you need vector art or just an image that nails your prompt, V3 has a solution.
Key Takeaways:
New Image Quality Standard: Recraft V3 tops the charts, outperforming Midjourney and OpenAI, as confirmed by Hugging Face’s leaderboard. It leads in text clarity, anatomical accuracy, and overall aesthetics—finally tackling issues like awkward hands and spatial oddities.
Precision Design Control: V3 isn’t just for creating; it’s for designing. Users can specify text size, position, and style to build consistent branding. From tiny pictograms to high-def vector art, it’s all customizable, making it a graphic designer’s dream.
The Full Designer’s Suite: Recraft V3 offers a suite of tools, including inpainting, outpainting, AI erasers, upscaling, and vectorization. Plus, the new API means businesses can integrate V3’s graphic prowess directly into their workflows.
Practical Tips for Business Owners:
Explore AI-Driven Branding: V3’s control over style and positioning lets brands create assets with consistency, ideal for businesses looking to maintain a strong visual identity.
Boost Design Flexibility: Integrate Recraft’s API to give your design team control over every element—from color schemes to positioning. It’s a way to streamline design work while keeping quality high.
Upscale Your Content: If your business involves producing a high volume of graphics, Recraft V3 can handle the quality control, providing seamless upscaling, image refinement, and on-brand style management.
Bigger Picture:
Recraft V3 represents a significant leap forward in AI design capabilities. More than just a text-to-image model, V3 is built for designers by enabling detailed control over each element of an image. While AI has often been about “good enough” image generation, Recraft’s latest version redefines that narrative—giving creatives the precision they need without sacrificing quality. V3’s features, from vector image generation to advanced inpainting, go beyond automation, offering tools that are genuinely useful to professionals in the design space.

Spies Are Now Using Radar to Eavesdrop on Your Phone’s Vibrations

Quick Byte:
Imagine a world where spies don’t need to hack into your phone to hear your conversations—they just need to watch it. Researchers have discovered that they can use radar to pick up tiny vibrations from a smartphone’s speaker, allowing an AI model to transcribe conversations from across the room. Yep, eavesdropping just got a lot more high-tech.
Key Takeaways:
Vibrations Are the New Data Leak: A standard millimeter-wave radar sensor can detect vibrations as small as 7 micrometers in a smartphone speaker. Even in noisy environments, it can isolate the conversation from all the background sounds, homing in on the phone’s tiny movements alone.
AI Translation: Once the radar picks up the vibrations, AI steps in to transcribe the conversation. While the audio is scratchy, an AI model managed to get a word accuracy rate of over 50% and a character accuracy rate of 67%—all from about 4 feet away.
Technical Feats and Limits: The real challenge? Filtering out all the other micro-movements, like breathing and heartbeat vibrations. By applying some clever statistical error correction, the researchers were able to separate these from the conversation, though it still requires a stable setting.
Bigger Picture:
This research is more than a spy-novel plot; it’s a glimpse into the future of surveillance. By using off-the-shelf millimeter-wave radar sensors, researchers demonstrated a cheap, accessible way to turn a phone’s vibrations into a transcript of your conversation. And while this technology isn’t flawless yet (the environment needs to be relatively stable), it’s only a matter of time before more precise tools emerge.
Suryoday Basak’s team at Penn State essentially showed that physical motion—even down to the level of micrometers—is enough for AI to make out the words of a conversation. Basak declined to comment on whether any government agencies were showing interest, but given the tech’s potential, you can bet it’s on their radar (pun intended). This tech isn’t just about eavesdropping; it’s reshaping the boundaries of privacy itself. In a world where AI can pick up whispers through a wall, privacy will need a serious upgrade.

ChatGPT’s New AI Search Tool is Here: Game-Changer or Just Another Feature?

Quick Byte:
OpenAI’s latest move in the search game isn’t your average feature update. We’re talking about a next-gen AI search tool integrated straight into ChatGPT. But can it hold its own against the Google Goliath? Here’s what we know so far and why it matters for anyone tired of traditional search engines.
Key Takeaways:
Real-Time Info, But It’s Gated: The new AI-powered search is available to ChatGPT Plus users ($20/month) and some team accounts—free users will have to wait until early next year.
Big Deals, Big Data: OpenAI has partnered with major publishers like The Atlantic, Vox Media, and Condé Nast, allowing ChatGPT to pull high-quality data into your search results without skirting copyright issues.
Less “Hallucinating,” More Fact-Checking: The tool cites sources more reliably than past versions, but it’s not without flaws. Just like Google’s AI features, it can still mix up facts if you ask it something dicey or out of left field.
It’s Not Just About Keywords Anymore: ChatGPT’s search shines when you give it natural language questions, and it can now pull up stock charts, real-time updates, and product recommendations in a single response.
First Impressions:
In the lead-up to launch, OpenAI promised an experience that could rival the search giants. The new ChatGPT tool looks and feels a lot like Perplexity but offers more depth in areas like interactive search and detailed citations. OpenAI even threw in new features like audio conversation and editing help, but when it comes to core search results, there’s room for improvement. Expect to see some trial and error while they work out the kinks.
When it comes to controversial queries, the AI search tool has its blind spots. It’s still tripping up on outdated or misleading data, though OpenAI says they’re tackling these inconsistencies head-on. That said, this new version’s transparency around sources makes it a much better tool for quick research than previous iterations.
The Bigger Picture:
AI search is about to get crowded. Google, Meta, and now OpenAI are all fighting for space in a billion-dollar market that’s no longer limited to the tech-savvy few. OpenAI’s new search tool could very well shift how we think about “search,” especially if it nails integrations with online shopping and more complex queries. So, if user habits start shifting in the next few years, OpenAI might just find itself a slice of the search pie—complete with shopping recommendations, research data, and yes, sports scores.
For now, the ChatGPT search tool isn’t going to dethrone Google, but it’s a solid start. Think of it as a hybrid between research assistant and personal shopper—a sign of where search is heading, with OpenAI betting that people want more than just a list of links.


Build An App and Deploy It In Minutes


Authors:
Jason Wei, Nguyen Karina, Hyung Won Chung, Yunxin Joy Jiao, Spencer Papay, Amelia Glaese, John Schulman, William Fedus
Affiliated with: OpenAI
Summary:
SimpleQA is a benchmark designed to measure the factual accuracy of language models by asking short, fact-seeking questions that have a single, indisputable answer. By focusing on concise questions, SimpleQA provides a clear-cut evaluation of whether models "know what they know" or if they produce incorrect responses, a common issue known as hallucinations. It’s challenging enough to test cutting-edge models like GPT-4, as well as manageable to grade efficiently, making it a valuable tool for assessing factuality in AI.
Why This Research Matters:
AI hallucinations, where models generate false or unsupported statements, are a major obstacle for trustworthy AI deployment. SimpleQA addresses this by setting a factuality standard with questions that demand accuracy, and it helps researchers understand if models can reliably answer without guessing. Given the increasing reliance on AI for information, a benchmark like SimpleQA helps guide development toward models that can be trusted across applications, from customer support to educational tools.
Use Cases:
AI Model Development: Researchers and developers can use SimpleQA to refine models, identifying areas where they are likely to produce inaccurate answers.
Customer Service: AI-driven support systems can benefit from factual benchmarks, ensuring answers to customer queries are reliable and minimize incorrect responses.
Education: For educational platforms utilizing AI, factuality benchmarks like SimpleQA ensure that students receive accurate, verifiable information.
Healthcare and Legal Assistance: In fields where accuracy is paramount, models validated against SimpleQA can reduce misinformation risks.
Immediate Impact:
SimpleQA provides a rigorous yet manageable way to assess and compare factual accuracy across models. With an easy-to-grade format, it allows researchers to quickly evaluate factuality, offering insights for fine-tuning model reliability. This is especially useful for models that handle specific, knowledge-based tasks requiring precision.
Future Impact:
As AI models grow more complex, SimpleQA's approach to factuality benchmarking will likely become foundational. It sets a high standard for factual precision that can be expanded to other types of AI tasks, like long-form text generation, enhancing model transparency. Ultimately, this could lead to greater public trust in AI, as models trained against SimpleQA and similar benchmarks will be better equipped to deliver accurate, dependable information across various industries.


I've tried every AI Coding agent and IDE,
As someone who's been coding for 20 years,
I think coding as we know it is cooked 🧵 :— John Rush (@johnrushx)
6:00 AM • Oct 30, 2024