3297625
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).
🤖 Anthropic launches Claude Code Security and upgrades Desktop
Anthropic released Claude Code Security, an agent that scans full repositories and finds complex context-level vulnerabilities.
It uncovered 500+ issues in production open source projects, some decades old.
The agent reasons across the codebase instead of relying on fixed SAST rules.
Claude Code Desktop now runs Server Previews, fixes console errors, reviews before push, monitors PRs, and supports auto-merge.
💧 Sponsored by @rainbetcom - rainbet.com
📊 @tech
⬇️ The quiet creator of WinRAR
Eugene Roshal built WinRAR and largely stayed out of the spotlight.
RAR stands for Roshal Archive.
He avoided Silicon Valley, skipped media attention, and kept his personal life private.
WinRAR became one of the most installed software tools in the world. The 40-day trial became a running joke since it never truly locked users out.
Estimated net worth ranges between $100M and $250M.
While others chased headlines, he built a compression algorithm that powered millions of PCs.
💻 @programmer
🗺Market Map shows how crowded the AI agent stack has become
A new market map lays out today’s AI agent landscape. Around LLMs, entire layers have formed.
Engines, environments, agent frameworks, memory, code review. The stack keeps expanding beyond the model itself.
💧 Sponsored by @rainbetcom - rainbet.com
📊 @tech
🎵 Google launches Lyria 3 with music generation from text, images, and video
Google introduced Lyria 3, a new music generation model. It creates tracks up to 30 seconds long and is already available in the Gemini app and on desktop, with a wider rollout coming soon.
Lyria 3 is multimodal. Users can generate music from a text prompt, or upload a photo or video and get a track that matches the mood and context of that content.
Google says every composition includes an invisible watermark. Distribution gets easier. Attribution becomes part of the product.
💧 Sponsored by @rainbetcom - rainbet.com
📊 @tech
🤖 Chinese robots go from simple dances to flips in 1 year
A new video shows how far Chinese humanoid robots have progressed over the past year.
In early 2025, most demos focused on basic dance routines and controlled movements.
Now, the same class of robots is performing jump flips and more dynamic maneuvers.
💧 Sponsored by @rainbetcom - rainbet.com
📊 @tech
💻 CEO of Shopify is shipping more code than ever
🔸 2024: 94 commits
🔸 2025: 833 commits
🔸 2026: 957 commits (in first 45 days of the year)
Claude is turning CEOs back to builder mode.
💧 Sponsored by @rainbetcom - rainbet.com
📊 @tech
🤖 Kimi integrates OpenClaw to let users spin up custom agents
OpenClaw is now integrated into Kimi. Users can create their own agents in a few clicks and run them inside Kimi’s web interface.
Agents can execute actions across connected tools and build long task chains.
🔸 Through ClawHub, users can combine more than 5,000 skills
🔸 Each agent gets up to 40GB of cloud storage for files
🔸 Search can be configured to run
continuously and fetch fresh information
🔸 External OpenClaw setups can also be linked and controlled from Kimi
💧 Sponsored by @rainbetcom - rainbet.com
📊 @tech
OpenClaw creator Peter Steinberg is the literal “Ok Claude, make a $1B company, make no mistakes” gigachad meme.
💧 Sponsored by @rainbetcom - rainbet.com
📊 @tech
They are now 3D printing a 12 meter boat in one piece with robots, no mold and no extra cost.
What once needed a full shipyard can now be done with a giant printer.
💧 Sponsored by @rainbetcom - rainbet.com
📊 @tech
🦞 OpenAI acquires OpenClaw and hires its creator
OpenAI has absorbed the OpenClaw project and brought its creator, Peter Steinberg, into the company. He will lead development of the next generation of AI agents. Sam Altman called him a genius and said his ideas will soon shape OpenAI products.
OpenClaw itself will remain open source. OpenAI says it will continue to support and develop the project.
The repository has been one of the fastest-growing in GitHub history by stars. At its peak, it gained 34,000 stars in two days. In about 60 days, it went from roughly 9,000 to nearly 200,000 stars. For comparison, Kubernetes took about three years to reach 100,000.
Anthropic reportedly had an opportunity to move earlier but focused on trademark concerns around the original name ClawdBot.
This time, OpenAI moved first and secured the team behind the momentum.
💧 Sponsored by @rainbetcom - rainbet.com
📊 @tech
Silicon valley joked about AI Agents contacting each other.
💧 Sponsored by @rainbetcom - rainbet.com
📊 @tech
Most people think AI started with ChatGPT, but Steve Jobs called it in 1983.
He explains why the next Aristotle will live forever through a machine.
💧 Sponsored by @rainbetcom - rainbet.com
📊 @tech
🚀 China lands Long March-10A first stage on water for the first time
China has completed a water landing of the first stage of Long March-10A, its super-heavy reusable rocket. The vehicle stands 67 meters tall with a 5-meter diameter.
Long March-10A is designed for lunar missions. China plans to use it to deliver taikonauts to the Moon by 2030.
Reusable heavy lift is now part of the lunar timeline.
💧 Sponsored by @rainbetcom - rainbet.com
📊 @tech
🎓 Udemy integrates its 290,000 courses into ChatGPT
Udemy announced a partnership with OpenAI. Its catalog of more than 290,000 courses is now accessible inside ChatGPT through a course discovery feature.
Users can browse and interact with Udemy’s learning content directly in chat instead of leaving the interface.
The move comes as Udemy prepares to merge with Coursera in a deal valued at about $2.5B. The transaction, announced in December 2025, is expected to close in the second half of 2026 and would combine the two largest online learning platforms under the Coursera brand.
AI search is becoming the new front door for education platforms.
💧 Sponsored by @rainbetcom - rainbet.com
📊 @tech
🎮 10-minute animated fight made with Seedance 2 for $60
A creator produced a 10-minute animated fighting scene using Seedance 2, a relatively new Chinese model from ByteDance. Total spend: $60.
Seedance 2 is already filling timelines with short clips. Some look impressive. Some look chaotic. The cost barrier keeps dropping.
High-end game and film production used to mean serious budgets. Now a laptop and a small budget can get you minutes of animated action.
💧 Sponsored by @rainbetcom - rainbet.com
📊 @tech
🔍 Chrome adds built-in Split View for two tabs in one window
Google rolled out Split View in Chrome. Users can now place two tabs side by side inside a single window, without extensions.
The feature is aimed at people who switch between tabs constantly. Teachers use it to review assignments. Developers keep documentation visible while coding. Some take notes while watching YouTube.
It simplifies basic multitasking inside the browser.
💧 Sponsored by @rainbetcom - rainbet.com
📊 @tech
💡 Microsoft builds glass data storage with 10,000-year lifespan
Microsoft developed a glass-based data storage system using laser-written 3D pixels inside transparent glass. The company says it is the first to offer a full end-to-end system for writing, storing, and reading this format.
Data is written with a femtosecond laser that alters the glass structure through ultra-short pulses. The changes do not damage the material and can only be detected optically. Reading requires a microscope, while convolutional neural networks help reduce errors caused by noisy light signals.
Tests suggest the medium can preserve data for up to 10,000 years, compared to 40–50 years for typical storage without maintenance. Density reaches about 4.8 TB on a ~12 cm² disk.
Storage consumes virtually no energy once written.
💧 Sponsored by @rainbetcom - rainbet.com
📊 @tech
🚕 Waymo robotaxis stop running in heavy rain and snow
Waymo robotaxis rely on LiDAR for perception. In heavy rain, snow, or dust, the sensor struggles due to reflection and scattering.
As a result, the vehicles are taken offline during strong precipitation.
Autonomous driving still has weather limits.
💧 Sponsored by @rainbetcom - rainbet.com
📊 @tech
🖌 Claude Code can now generate full designs directly in Figma
Claude Code is now integrated with Figma, allowing it to generate complete, editable designs inside Figma.
The reverse flow already existed through Figma MCP, turning designs into code. This update completes the loop. Code written in the IDE can be translated into structured Figma layers, adjusted visually, then pushed back into code.
Development no longer has to follow a strict idea → design → code sequence. A product can start in code, move into design for review, then return to code with changes applied.
It turns coding agents into design-aware collaborators.
💧 Sponsored by @rainbetcom - rainbet.com
📊 @tech
🍏Apple prepares 3 AI wearables built around visual Siri
Mark Gurman reports that Apple is accelerating work on three wearable devices centered on Siri with visual context.
🔸 The first is smart glasses, code-named N50, positioned as a rival to Ray-Ban Meta. They would include two cameras, one for photos and video, another for computer vision similar to Vision Pro. There is no display. Interaction runs through speakers, microphones, and Siri. Production is targeted for December, with a potential 2027 release.
🔸 The second is a pendant device, roughly AirTag-sized, worn as a clip or necklace. It includes a camera and microphone and relies on the iPhone for processing. Unlike Humane AI Pin, it would function as an iPhone accessory rather than a standalone product. The project remains early stage and could launch in 2027.
🔸 The third is AirPods with a low-resolution camera designed to provide visual context for AI. Rumors have circulated since 2024, with a possible release this year.
At a recent internal meeting, Tim Cook said Apple is working on new product categories enabled by AI.
Apple’s next hardware cycle may revolve around giving Siri eyes.
💧 Sponsored by @rainbetcom - rainbet.com
📊 @tech
Programmers were asked to make the worst volume control for a contest
💻 @programmer
🤖 The man who taught AI to see
If you have ever scanned a bank card with your phone camera or watched an app extract data from a receipt, you used technology shaped by Yann LeCun.
In the 1980s, computer vision was stuck. Systems tried to process entire images at once, consumed heavy compute, and produced weak results. LeCun took a different path inspired by biology. He designed neural networks that break images into small patterns like edges and curves, then assemble meaning from those pieces.
In 1989, he introduced a handwritten digit recognition system. By the 1990s, his algorithms were processing roughly 10% of bank checks in the US. It was one of the first cases where neural networks delivered measurable profit to large companies.
💧 Sponsored by @rainbetcom - rainbet.com
📊 @tech
🍏 Apple to unveil $699 MacBook with A18 Pro in March
Mark Gurman reports that Apple plans to announce a lower-cost MacBook at its March event. The device is expected to start at $699.
Instead of an M-series chip, it will use the A18 Pro from the iPhone 16 Pro. While positioned below Apple Silicon M chips, the A18 Pro is said to approach original M1-level performance. The display is rumored at 12.9 inches. Apple has also developed a cheaper and faster aluminum manufacturing process for the chassis.
Color testing reportedly includes light yellow, light green, blue, pink, silver, and dark gray. Current MacBook Air models start at $999.
A sub-$700 Mac would reset the entry point to the Mac lineup.
💧 Sponsored by @rainbetcom - rainbet.com
📊 @tech
🥋 AGIBOT shows kung fu robot A3 with jump kicks and flips
Shanghai-based AGIBOT presented its robot A3 performing jump kicks, spinning strikes at different angles, and backflips. The company says the robot can consistently hit its target during demonstrations.
Despite the martial arts moves, the intended use case is far calmer. AGIBOT plans to deploy A3 for guided tours and escort services.
💧 Sponsored by @rainbetcom - rainbet.com
📊 @tech
A human tried to troll a Tesla robot, then the robot trolled him back.
💧 Sponsored by @rainbetcom - rainbet.com
📊 @tech
“Don’t worry, there will still be great jobs even when AI automates everything”
The jobs:
💧 Sponsored by @rainbetcom - rainbet.com
📊 @tech
Full stack developer in 2026 be like
💻 @programmer
⚠️ Anthropic safety researcher resigns, says “the world is in danger” from fast AI progress
🖱 A senior AI safety researcher at Anthropic, Mrinank Sharma, stepped down, warning that rapid AI development puts the world at risk. He said safety teams face ongoing pressure to sideline core concerns, including bioterrorism and other catastrophic scenarios.
🖱 Anthropic was founded to build safer AI systems, yet even its CEO Dario Amodei has publicly cautioned that progress may be moving too quickly and called for slowing deployment.
🖱 Similar tensions have surfaced elsewhere. Senior safety researchers, including members of the former Superalignment team at OpenAI, have left, arguing that commercial priorities are overtaking efforts to reduce risks from systems that could surpass human intelligence.
🖱 Former OpenAI researcher Zoe Hitzig resigned after ads were introduced in ChatGPT. She warned that advertising inside highly personal conversations could enable subtle manipulation, especially as users share sensitive details about health, finances, and relationships.
Safety debates are no longer internal. They are playing out in public as AI systems scale.
✔️ Subscribe on @venture
🧠 DeepMind builds an AI that scores 91.9% on elite math proof test
Google DeepMind introduced Aletheia, a math-focused AI agent that scored 91.9% on IMO-ProofBench Advanced, one of the toughest public benchmarks for Olympiad-style proofs.
Aletheia runs on Gemini Deep Think and uses a loop of generating proofs, checking them, then refining mistakes. DeepMind says it outperforms even newer Gemini Deep Think Advanced setups while using less compute.
Beyond benchmarks, the agent has formally solved four problems from the Erdős list, with one likely not previously closed in literature. It also wrote a full math paper with correct results and assisted researchers on real academic work.
Proof-based reasoning is becoming measurable, scalable, and cheaper at the same time.
💧 Sponsored by @rainbetcom - rainbet.com
📊 @tech
🧠 MIT finds heavy ChatGPT use weakens recall and brain activity
MIT studied how ChatGPT affects users during writing tasks. 83.3% of participants using the tool could not recall a single sentence they had written minutes earlier. Participants who wrote without AI recalled their work easily.
Neural connectivity scores for ChatGPT users dropped from 79 to 42, a 47% decline. This was the lowest level among all groups. Even after stopping AI use, these participants showed persistently lower activity in later sessions.
AI-written essays were technically correct but often described as robotic and shallow. ChatGPT made users 60% faster while reducing mental effort by 32%.
The strongest results came from people who wrote on their own first and used AI later.
💧 Sponsored by @rainbetcom - rainbet.com
📊 @tech