4273206
A hub for startup news, trends, and insights, covering the global startup ecosystem for founders, investors, and innovators. Community: @startupdis Buy Ads: @strategy (this is our only account).
🧬 Anthrogen unveils Odyssey, the most powerful AI for protein design
San Francisco–based Anthrogen has introduced Odyssey, a large language model built specifically for bioengineering and protein editing. The company calls it the most advanced and efficient system of its kind.
🔸 Odyssey can interpret amino acid sequences, 3D structures, and protein functions, then design new molecules that meet set goals such as “binds to this receptor” or “scalable for production.”
🔸 It uses discrete diffusion, a training method inspired by evolution that introduces mutations and keeps only the best-performing results.
🔸 Anthrogen reports that Odyssey achieves superior protein design outcomes using 10× less data than competing models.
From drug discovery to synthetic biology, Odyssey could redefine how we engineer life itself.
🧼 Japan unveils the “washing machine for humans”
A Japanese startup has created a capsule that completely replaces the shower and bath, users simply lie inside while the machine automatically washes their entire body and hair.
🔸 The chamber fills with warm water up to the neck and uses jets, bubbles, and ultrasonic waves for cleaning.
🔸 A built-in screen lets users watch videos or relax during the wash cycle.
🔸 Hotels and capsule hostels in Japan are already pre-ordering the device for guest rooms.
Japan’s obsession with automation has finally reached peak cleanliness.
🤖 Jack & Jill launches AI bots for job interviews and candidate matching
Jack & Jill is redesigning the hiring process with conversational AI that interviews both candidates and employers, replacing resumes and forms with natural dialogue. The startup has raised $20M from Creandum to expand from London to the U.S.
🔸 Candidates talk to “Jack,” a bot that runs a 20-minute interview, builds their profile, and finds fitting roles.
🔸 Employers use “Jill,” which defines ideal candidate profiles and filters matches.
🔸 Over 50,000 users are already active in London, with early traction among mid-sized employers.
By betting on conversation over credentials, Jack & Jill wants to cut hiring noise and fake applications and move one step closer to bot-to-bot recruiting.
💬 Andrej Karpathy builds a full LLM for just $100
Andrej Karpathy, former OpenAI researcher and Tesla AI lead, has launched nanochat, an open project showing how to train a functioning language model from scratch for around $100 using cloud resources.
🔸 Developed under the LLM101n course by Eureka Labs, nanochat walks through the entire lifecycle of model creation, from tokenization to inference.
🔸 The repo includes a tokenizer written in Rust, a FineWeb-EDU dataset paging system, training and fine-tuning scripts, plus evaluation tools.
🔸 A built-in web interface lets users chat with their trained model directly, making it a practical learning and prototyping tool.
By shrinking the cost and complexity of LLM development, Karpathy is effectively open-sourcing AI education, turning large-scale model training into a weekend project.
✅ The most vulnerable professions to AI
Microsoft analyzed user conversations with Copilot and compiled a top 40 list of professions whose representatives may lose their jobs to AI.
📊 Powered by Crypto Insider
🎬 Google unveils Veo 3.1, the most cinematic AI video model yet
Google has released Veo 3.1, its next-generation AI video model, and while it doesn’t yet hit the rumored 60-second or full-1080p mark, it’s still a massive leap forward in realism and creative control. Forget blinking artifacts and robotic motion, Veo 3.1 finally makes AI video feel alive.
🎥 What’s new in Veo 3.1
🔸 Insert Object / Remove Object — add or erase elements directly within the scene.
🔸 Ingredients to Video — blend multiple references into one cohesive clip for full creative control.
🔸 Flow Extend — expand scenes seamlessly to create long, continuous shots.
🔸 First & Last Frame — define your opening and closing frames; Veo fills in the rest like a real cinematographer.
🔸 Enhanced sound, textures, and lighting, creating depth and emotional tone in every frame.
🔸 Audio support now integrated into Flow, Vertex AI, Gemini app, and Gemini API.
Veo 3.1 doesn’t just render, it interprets tempo, mood, and cinematic rhythm. For the first time, AI video feels less like a simulation and more like a director’s cut.
💋 ChatGPT to introduce adult mode with erotic conversations
OpenAI plans to launch an “adult interaction” mode for ChatGPT this December, allowing erotic and flirtatious conversations for verified adults. The move marks a major shift in how AI chat platforms handle intimacy and content moderation.
🔸 The update will include age verification, requiring users to confirm they are over 18, and in some cases, to upload ID documents. Underage users will be blocked from adult interactions.
🔸 OpenAI says the broader update will also make ChatGPT “more human-like, expressive, and emotionally responsive”, hinting at a friendlier personality model.
🔸 The change is seen as a response to the explosive popularity of character.ai, Replika, and other “AI companion” platforms that thrive on emotional and romantic engagement.
🔸 Critics, however, warn that document-based age verification could become a dangerous precedent, today for erotica, tomorrow for healthcare, and later for politics.
If OpenAI truly opens the gates to intimacy, it won’t just challenge competitors, it’ll redefine what “personal AI” means in a world where companionship, data, and desire start to merge.
⚡️ Nvidia launches $3,999 DGX Spark, a personal AI supercomputer for your desk
Nvidia has unveiled the DGX Spark, a compact AI workstation that delivers server-level performance in a desktop form. Starting at $3,999, it goes on sale this week, aiming to make petaFLOP-class compute accessible to researchers, developers, and small AI labs.
🔸 Powered by Nvidia’s new GB10 Grace-Blackwell Superchip, combining an ARM CPU and Blackwell GPU for unified AI workloads.
🔸 Comes with 128 GB unified memory and up to 4 TB NVMe SSD storage, offering nearly 1 petaFLOP of AI performance.
🔸 Can locally train or fine-tune models with up to 200 billion parameters, bridging the gap between desktop and data center.
🔸 OEMs like Dell, HP, and ASUS are expected to build Spark-based systems using the same architecture.
By compressing server-grade AI power into a desktop, Nvidia isn’t just selling hardware, it’s seeding an ecosystem for local frontier-model experimentation and on-prem AI autonomy.
💰 New VC funds for early-stage founders, global roundup
Let’s kick off the week with a strong batch of freshly launched venture funds, covering AI, climate, biotech, and quantum. If you’re raising or scouting partners, this is your early-stage investor radar.
🔸 Alt Capital (USA) — $275M second fund focused on AI enterprise software; early-stage, U.S. and global.
🔸 Tandem Ventures (USA) — $50M first fund (+$100M SPVs) backing B2B SaaS and vertical AI startups.
🔸 T.Rx Capital (USA) — $77.5M first fund investing in biotech and life sciences.
🔸 VoLo Earth Ventures (USA) — $135M climate tech fund targeting energy, mobility, and industry.
🔸 Archetype (USA) — >$100M third fund for crypto and decentralized infrastructure plays.
🔸 Touring Capital (USA) — $330M debut fund backing AI-driven software at the early-growth stage.
🔸 Toyota Invention Partners & Woven Capital (Japan) — $670M + $800M funds focused on AI, automation, and climate tech.
🔸 Concept Ventures (UK) — €75M second fund, now Europe’s largest pre-seed vehicle.
🔸 Better Tomorrow Ventures (USA) — $140M third fund specializing in fintech.
🔸 55North (Denmark) — €134M (first close) quantum fund targeting computing, sensing, and communications.
🔸 Wave Function Ventures (USA) — $15.1M debut deep tech fund in nuclear, robotics, and aerospace.
🔸 Sugar Free Capital (USA) — $32M fund dedicated to AI-native startups.
🔸 Crystal Venture Partners (USA) — $33M fund for AI in insurance and regulated industries.
🔸 Ascenta (USA) — $325M fund building biotech platform companies.
The early-stage market is alive and well, from quantum to climate, capital is moving fast toward deep and defensible tech.
🛡️ Fake Detail launches privacy-first identity generator
Fake Detail has introduced a privacy-focused tool that lets users create disposable digital identities, temporary data sets designed to protect personal information during sign-ups, tests, or online interactions.
🔸 Generates temporary emails, usernames, and phone numbers from selected operators for secure, one-time use.
🔸 Creates pseudonymous names and addresses, helping users mask real data in non-critical contexts.
🔸 Builds mock social media profiles (YouTube, X, TikTok) for testing or design purposes, without linking to real users.
🔸 Simulates sample chats and interactions to support app testing and UX prototyping.
The goal isn’t deception, it’s data minimalism: giving users control over what information they share online. Fake Detail turns privacy protection into a seamless, everyday habit.
⚠️ Top reasons why startups shut down.
📊 Powered by Crypto Insider
💡 Lumi, AI that builds full websites from a single prompt
Meet Lumi, a new AI tool that can create complete websites in seconds no coding required. Just describe your idea in chat, upload design references if you have them, and Lumi automatically generates the entire product.
🔸 Builds the full stack: interface, backend, database, login form, and analytics.
🔸 Supports real-time editing, you can tweak design or functionality directly in chat.
🔸 Works for landing pages, dashboards, or even SaaS prototypes.
🔸 Ideal for founders, marketers, or designers who want to go from concept to demo instantly.
Lumi turns the idea of “no-code” into “no-setup” shifting web development from writing code to describing intent.
🧱 KIRI Engine releases 3DGS to Mesh 2.0, turn any object into a 3D model
KIRI Engine has launched 3DGS to Mesh 2.0, a breakthrough AI model that lets anyone scan real-world objects and instantly convert them into editable 3D models. The system uses a refined pipeline that merges mesh generation, normal prediction, and reflection filtering for ultra-realistic results.
🔸 Generates high-fidelity 3D meshes with accurate lighting and textures straight from a phone camera.
🔸 Combines 3D Gaussian Splatting (3DGS) with optimized mesh reconstruction for smoother surfaces.
🔸 Models can be exported and edited in most 3D software, ideal for game devs, artists, and product designers.
🔸 The app is free to use, making high-quality 3D capture accessible to everyone, for both Android and IOS users.
AI photogrammetry just leveled up, 3D creation is becoming as simple as taking a photo.
⌨️ A viral TikTok just taught the internet how to properly delete text
A short TikTok clip went viral after revealing that pressing Ctrl + Backspace deletes entire words instead of single letters, something millions of users apparently never knew.
🔸 The video hit 30 million views in days, sparking disbelief across social media.
🔸 Even Microsoft’s official account commented, admitting they “didn’t know” about the shortcut.
🔸 The trick works across most text editors, browsers, and messaging apps, saving seconds that now feel like years of lost time.
The internet just collectively discovered productivity’s biggest cheat code and yes, the world will never be the same.
🎧 David AI secures $50M to expand its global audio data lab
San Francisco–based David AI, a YC-backed audio data research company, has raised $50M in Series B funding led by Meritech and NVIDIA, with participation from Alt Capital, First Round, Amplify Partners, and Y Combinator. The round brings total funding to $80M.
Founded in 2024 by ex-Scale AI engineers Tomer Cohen and Ben Wiley, David AI calls itself the world’s first dedicated audio data research lab. The company builds large, diverse datasets that capture global human speech — across languages, emotions, and real-world environments — to train next-generation voice and audio AI models.
🔸 The new capital will fund research expansion, team growth, and deeper collaboration with global AI and hardware firms.
🔸 David AI’s datasets already power voice models used by top AI labs and several “Mag 7” tech giants.
🔸 The company recently surpassed an eight-figure annual revenue run rate.
🔸 Competitors include Defined.ai, PublicAI, Human Native AI, Ydata, and Syntheticus.
David AI positions itself as the data backbone for the coming wave of audio-first AI - powering the interfaces that will connect humans and machines in the physical world.
🚀 Starship deploys next-gen Starlink satellites during 11th test flight
SpaceX’s 11th Starship flight wasn’t just a technical milestone, it also marked the first successful deployment of mock-ups of the third-generation Starlink satellites, confirming the spacecraft’s role as a heavy-lift system for the network’s future.
🔸 The two-ton Starlink V3 satellites are designed to deliver gigabit internet globally via over 1 Tbps downlink and 200 Gbps uplink capacity, roughly 10× faster than current models.
🔸 Their size and power make them impossible to launch on existing rockets, underscoring why Starship’s payload capacity is critical to the project.
🔸 Operational launches are scheduled to begin next year, using Starship’s commercial missions.
Building the internet for Earth may turn out to be as ambitious as reaching Mars.
To be honest, they didn't even make a mistake.
📊 Powered by Crypto Insider
🧠 Perplexity adds interactive language learning mode
Perplexity has launched a new language learning mode, turning its AI chatbot into an interactive tutor that helps users build vocabulary and understand grammar in real time.
🔸 The feature generates flashcards with new words, pronunciation tips, and example sentences directly in chat.
🔸 Users can also ask grammar questions or request explanations within ongoing conversations.
🔸 The mode is already live on iOS and the web, expanding Perplexity’s use beyond search and research.
Perplexity is positioning itself as both a knowledge engine and a personal language coach.
🪟 Microsoft turns Copilot into a full AI agent for Windows 11
Microsoft’s latest Windows 11 update transforms Copilot from a chat assistant into an autonomous, voice-controlled AI agent capable of managing your PC and interacting with apps and the web. Activated by saying “Hey, Copilot,” it can now take real actions on your computer.
🔸 Copilot can see what’s on the screen, respond to voice commands, and perform complex tasks like booking tables, ordering food, or organizing local files.
🔸 All actions require explicit user permission, and the autonomous mode is disabled by default for privacy and safety reasons.
🔸 A new Gaming Copilot is also built in, a live assistant that helps players with strategy tips, in-game information, and performance tweaks.
🔸 The feature rolls out first to Windows Insider testers before reaching all Windows 11 users.
With Copilot’s evolution into a true desktop agent, Microsoft is positioning Windows as the first mainstream operating system with a native, AI-driven co-pilot for daily computing.
⚡ Anthropic launches Claude Haiku 4.5, faster, cheaper, and smarter
Anthropic has released Claude Haiku 4.5, the newest version of its lightweight model, optimized for speed, efficiency, and cost. It now delivers coding and reasoning quality comparable to Claude Sonnet 4, but runs twice as fast and at one-third the price.
🔸 Haiku 4.5 excels at agentic tasks, code generation, and lightweight reasoning, making it ideal for production pipelines and real-time apps.
🔸 Pricing remains aggressively low: $1 per million input tokens and $5 per million output tokens.
🔸 The model is available today through the Claude API and integrated into Claude Code for developers.
With Haiku 4.5, Anthropic is pushing a clear message, performance parity doesn’t need to come at premium pricing.
🌌 A gravitational wave detector that fits on a table
Physicists from the Universities of Birmingham and Sussex have designed a tabletop gravitational wave detector, potentially filling the “frequency gap” that existing observatories can’t reach. Unlike kilometer-scale systems like LIGO, this compact instrument, based on optical atomic clock technology, can detect waves in the elusive millihertz range (10⁻⁵–1 Hz).
🔸 The detector uses two optical resonators set at right angles, each only 10–50 cm long, but light bounces inside thousands of times, building an effective path of tens of kilometers.
🔸 When a gravitational wave passes, it subtly warps spacetime, changing the optical path length, not the mirrors themselves, which shifts the laser’s frequency by as little as 1 part in 10¹⁶.
🔸 That’s equivalent to detecting a 15-micrometer change over the Earth–Sun distance. By comparing the two resonators, the device can measure both the amplitude and polarization of the passing wave.
🔸 Its small size makes it nearly immune to seismic noise, the main limitation for large interferometers.
This miniature observatory could open a new observational window, catching signals from intermediate-mass black holes, double white dwarfs, and even echoes from the early Universe, all from the top of a lab bench.
🧠 Gemini 3.0 Pro now builds full web simulations of Windows and macOS
Google’s upcoming Gemini 3.0 Pro is already demonstrating wild creative capability, users report it can generate fully functional web-based operating systems that mimic Windows and macOS, all from a single prompt and inside one HTML file.
🔸 One user asked Gemini to “create a web OS like Windows OS” with a text editor, terminal (with Python), game, file manager, paint app, video editor, and other bundled tools, Gemini delivered the entire code in one go.
🔸 The result: a complete interactive simulation of Windows and macOS that runs directly in the browser via HTML, CSS, and JavaScript.
🔸 Demos are live — Windows version and macOS version — showcasing detailed icons, apps, and animations.
🔸 The rumored public release of Gemini 3.0 is set for October, with Pro-level models expected to debut in Google’s AI Studio soon after.
If this demo reflects real model capabilities, Gemini may soon blur the line between coding assistant and autonomous software creator.
🕊️ Icarus aims to conquer the stratosphere with autonomous “birds”
From Y Combinator’s latest batch comes Icarus, a rare non-AI startup developing autonomous drones designed to cruise the stratosphere. The idea: build long-endurance, solar-powered “birds” that can film, relay communications, and operate above weather systems for extended periods.
🔸 Founder Henry Kwan shared a 27-slide pitch deck, notably light on numbers but heavy on conviction, citing his 10,000+ hours in aerospace R&D as proof of execution capacity.
🔸 The concept echoes past attempts by Google (Project Loon) and Facebook’s Aquila, both of which failed to make stratospheric flight commercially viable.
🔸 Icarus claims advances in lightweight materials, autonomy, and solar efficiency now make persistent flight feasible, though technical details remain undisclosed.
🔸 The company’s vision blends aerial connectivity, high-altitude imaging, and atmospheric data, potentially creating a new layer between satellites and drones.
Meanwhile, Russia’s Stratolink, founded by Nicholas Oxman, is pursuing a similar goal, a small solar aircraft capable of flying up to 40 days autonomously. The firm announced a ₽1.7B raise and a ₽3.4B valuation, though its prototype and facilities remain largely unverified.
Stratospheric autonomy sits at the intersection of aerospace and infrastructure and may become the next race between frontier engineers and gravity itself.
🚐 Einride, the driverless future of electric logistics
Swedish startup Einride is building what might be the most complete vision of autonomous, electric transport, trucks without cabins, powered by AI, and coordinated by a digital logistics platform. Their philosophy: “You can’t go electric without going digital.”
🔸 Fully electric, cabless trucks, futuristic design increases cargo space and cuts weight.
🔸 Range up to 650 km on a single charge, with remote human intervention possible in emergencies.
🔸 Proprietary Saga platform manages routes, charging, and optimization across fleets.
🔸 Network expanding across Europe, the U.S., and Benelux, with clients including PepsiCo, Lidl, Carlsberg, and AB InBev.
🔸 Sales doubled in 2024, fleet size now exceeds 150 vehicles.
Einride isn’t just electrifying trucks, it’s re-engineering the logistics stack from wheels to cloud. The only question left: are we ready to go driverless before we go fully electric?
🤖 AI won’t replace you. But someone who knows how to use it will.
Most people open ChatGPT, type something random, and hope for magic.
They waste hours because they don’t know how to ask.
👉 This book fixes that. 👈
It gives you 100 proven prompts that make ChatGPT useful from day one — for work, business, learning, creativity, and daily life.
Here’s what you’ll find inside:
🔸 7 sections — Productivity, Creativity, Business, Learning, Finance, Lifestyle, and tools for Pro-Users.
🔸 100 proven prompts with examples and pro tips
🔸 Step-by-step methods for building better prompts
🔸 Real tools to save time, generate ideas, and automate routine
Plan your week. Write faster. Find better ideas. Automate boring stuff.
All with simple copy-paste prompts that actually work.
For everyone tired of AI that sounds clever but helps none.Читать полностью…
🛸 The 2025 State of AI Report: from static models to autonomous agents
A new State of AI report has been released, not entirely “fresh,” since by the time it’s compiled, the field has already moved on. But it offers a powerful retrospective on how much has changed in just one year.
🔸 Mathematical breakthroughs: several models have now become gold medalists in real math olympiads, a clear, binary benchmark of progress.
🔸 Continuous learning trend: the shift from static pretraining to models that learn continuously is gaining momentum.
🔸 Creative autonomy: AlphaZero proved that AI trained without human input can invent new strategies later studied by humans, dismantling the claim that “AI can’t create anything new.”
🔸 China’s open-model dominance: Chinese labs are now leading the West in open-source AI development.
🔸 Agents as collaborators: the paradigm is evolving from humans using tools to humans partnering with autonomous AI agents.
🔸 AI-first advantage: companies built around AI from inception are outperforming traditional firms in both revenue and growth.
Among the bold forecasts for the next 12 months:
• One major retailer will see 5% of offline purchases made by AI agents.
• A generative real-time video game will break Twitch viewership records.
• A film co-created with AI will resonate with audiences, and spark backlash from creators.
The report spans 300+ pages, full of insights, contradictions, and bold predictions. But one thing is clear: AI is shifting from a tool to an autonomous economic force.
⚙️ n8n raises $180M to become the orchestration layer for AI automation
n8n has closed a $180 million round at a $2.5 billion valuation, cementing its position as one of the leading infrastructure players in the emerging agent economy. Originally launched as an open-source alternative to Zapier, n8n has evolved into a hybrid AI orchestration platform, giving companies fine-grained control over how their automation agents think and act.
🔸 Founded in 2019 by Jan Oberhauser, n8n started as a visual workflow builder and has grown into a default tool across AI developer communities.
🔸 The platform offers configurable autonomy, allowing users to define when logic is manual, semi-automated, or AI-driven.
🔸 Supports 1,000+ integrations with SaaS and enterprise systems, bridging human workflows with AI agents.
🔸 Combines no-code simplicity with code-level extensibility, enabling both developers and business teams to collaborate in real time.
🔸 Backed by Accel, Salesforce Ventures, Felicis, and Sequoia, and powered by a strong open-source ecosystem.
As OpenAI pushes its GPT Agents ecosystem, n8n stands out as the neutral, open orchestration layer, a system that works with AI models rather than being locked into any single one.
n8n is betting that the future of automation will not be defined by closed ecosystems, but by flexible orchestration platforms that keep enterprises in control of their AI workflows.
🔔 Google launches AI fitting room for virtual shoe try-ons
Google has rolled out a new AI-powered shoe try-on feature, letting users see how sneakers look on their own feet before buying. Just tap “Try on” on a product card and upload a full-length photo, Google’s model then generates a realistic preview.
🔸 Available for now in the U.S., Australia, Canada, and Japan, with plans to expand to more regions.
🔸 The tool uses generative AI and pose-estimation to match shoes to your body type and lighting.
🔸 Initially supports major brands on Google Shopping, with apparel integration expected next.
🔸 Builds on Google’s earlier AI try-on for clothing, expanding into full-body visualization for retail.
E-commerce is entering its AR moment, and Google wants to own the virtual dressing room.
🤖 Figure unveils “Figure 03”, its first humanoid ready for mass production
Figure AI has introduced Figure 03, a next-generation humanoid robot designed as a versatile assistant for both homes and businesses. It can handle everyday tasks like doing laundry or greeting guests at a hotel reception, marking Figure’s shift from prototypes to production-ready robotics.
🔸 The robot features a textile suit and soft protective inserts to make human interaction safer and more natural.
🔸 It adds new sensors, cameras in its hands, an upgraded battery, and wireless charging, all powered by Figure’s in-house Helix neural network.
🔸 Figure 03 is optimized for large-scale manufacturing, with the first line capable of producing up to 12,000 units per year.
🔸 Pricing and launch details remain undisclosed, but the design suggests a move toward commercial deployment.
With Figure 03, the company is signaling that humanoid robotics is moving out of the lab, and into real-world service.
🔘Founders map the next AI winners at AI Challenges
The AI Challenges online conference gathered visionary founders to outline the unsolved problems that will decide which AI startups survive the next decade, from data access and inference costs to human alignment and product feedback loops. Here’s a concise, investor-facing digest.
🔸 “Data Freedom” confidential, siloed data is the biggest blocker; secure, auditable data pipes will unlock healthcare, legal, and enterprise verticals.
🔸 Inference & LM-training costs, fine-tuning and per-query inference are inefficient and expensive; parameter-efficient adaptation and hybrid edge/cloud designs are prime infra bets.
🔸 The AI Dealmaker: an agent for VCs that automates sourcing, diligence, and portfolio monitoring; retention hinges on measurable time-saved.
🔸 Turning LLM feedback into product outcomes: teams need SDKs and control planes to convert user signals into reliable model and UX improvements.
🔸 AI-native economy: autonomous agents will create new marketplaces (agent payroll, billing, reputation) and reshape how work is exchanged.
🔸 The Future of Human-AI Relationship: empathy, trust, and explainability are now product features, not optional extras.
🔸 Pitch session: five strong AI-native startups presented (themes: agent platforms, data-vaults, feedback control planes, vertical AI employees, cost-efficient inference).
🔸 Online afterparty hosted by venture investor DJ Mak.
A clear signal for investors: bet on speed + trust + economics products that prove time-saved, keep customer data in control, and make inference affordable will act like profitable, accountable “AI employees.”