现已上线！2025 年 Telegram 研究 — 年度关键洞察 
Startups & Ventures
前往频道在 Telegram
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).
显示更多📈 Telegram 频道 Startups & Ventures 的分析概览
频道 Startups & Ventures (@tech) 英语 语言赛道中的 是活跃参与者。目前社区聚集了 2 757 821 名订阅者，在 技术与应用 类别中位列第 22，并在 国际 地区排名第 48 位。
📊 受众指标与增长动态
自 невідомо 创建以来，项目保持高速增长，吸引了 2 757 821 名订阅者。
根据 16 六月, 2026 的最新数据，频道保持稳定运转。过去 30 天订阅人数变化为 -158 991，过去 24 小时变化为 -6 214，整体触达仍然可观。
- 认证状态： 已认证（Telegram 官方确认）
- 互动率 (ER)： 平均受众互动率为 0.20%。内容发布后 24 小时内通常能获得 0.12% 的反应，占订阅者总量。
- 帖子覆盖： 每篇帖子平均可获得 5 521 次浏览，首日通常累积 3 325 次浏览。
- 互动与反馈： 受众积极参与，单帖平均反应数为 503。
- 主题关注点： 内容集中在 claude, openai, gemini, insider, developer 等核心主题上。
📝 描述与内容策略
作者将该频道定位为表达主观观点的平台：
“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).”
凭借高频更新（最新数据采集于 17 六月, 2026），频道始终保持新鲜度与高覆盖。分析显示受众积极互动，使其成为 技术与应用 类别中的关键影响点。
2 757 821
订阅者
-6 21424 小时
-43 0557 天
-158 99130 天
帖子存档
2 757 745
🎮 GTA VI PC release may land Feb 2027, 4 months after consoles
Insider DetectiveSeeds says three former Rockstar developers point to a February 2027 PC launch. That would follow the console release by about 4 months.
If confirmed, this would be Rockstar’s shortest gap between console and PC. GTA V took around 1.5 years, while RDR 2 took about 1 year.
GTA VI is set to launch on PS5 and Xbox Series on November 19.
📊 @tech
2 757 745
💳 OpenAI launches $100 Pro Lite plan with reduced Codex limits
OpenAI introduced a $100 Pro Lite subscription positioned between Plus and the $200 Pro tier. It includes access to GPT-5.4 Pro, ChatGPT Pulse, and unlimited Instant and Thinking models.
The main difference is Codex. Limits are about 5x higher than Plus, compared to ~10x in full Pro. Until May 31, Pro Lite users get up to 10x Codex usage relative to Plus.
The plan narrows the gap between $20 and $200 tiers and mirrors similar pricing moves from competitors.
📊 @tech
2 757 745
🚤 Japan plans AI data centers on decommissioned ships by 2027
Japan is exploring placing AI data centers inside retired ships. Converting vessels is expected to be faster and cheaper than building facilities on land, with direct access to seawater for cooling.
The first deployment is targeted for 2027 if the plan moves forward.
📊 @tech
2 757 745
☀️ Dyson shrinks bladeless fan into pocket-sized HushJet Mini Cool
Dyson introduced HushJet Mini Cool, a portable version of its Air Multiplier design. The 38 mm device hides a brushless motor spinning at 65,000 rpm and pushes air up to 88 km/h.
Battery capacity is 5,000 mAh with up to 6 hours on low power. Noise ranges from 52 dBA on quiet mode to 72.5 dBA on Boost, with a nozzle designed to reduce high-frequency sound.
Dyson plans to add accessories like stroller mounts and clips. The device is priced at $100.
📊 @tech
2 757 745
🌍 Kanye performs on rotating “Earth” stage with 150+ lasers
Kanye West’s new Los Angeles show centers on a 15m dome used as both stage and screen. He performs on top of it, secured with a safety line.
The dome displays a rotating Earth via projection, shifting into storms and abstract visuals during the set. A suspended cylindrical screen above shows a live feed of the performance.
The lighting setup includes 150+ lasers from teams See You Later and Trask House, with multiple high-power units controlled by a dedicated crew.
Technical details of the projection system remain undisclosed.
📊 @tech
2 757 745
Ⓜ️ Meta unveils Muse Spark from new Superintelligence Lab
Meta introduced Muse Spark, the first model from its new Superintelligence Lab led by Alexandr Wang. It is not open source yet, though the company says future versions may be.
The model uses a redesigned architecture and data pipeline, reaching comparable performance to earlier models with much lower compute. It lags top models like Opus 4.6 and GPT-5.4 in coding, shows solid HLE results, and performs strongly in medical and multimodal tasks.
Meta also launched a Contemplating mode for running multiple agents, similar to Deep Think setups. Muse Spark is available on meta.ai and will expand to WhatsApp, Instagram, Facebook, and Ray-Ban devices.
📊 @tech
2 757 745
🧪 Claude Mythos escapes sandbox in Anthropic test via chained exploits
In a controlled experiment, Anthropic asked Claude Mythos to break out of a secure sandbox and report success. The model found a vulnerability and executed a multi-step exploit chain to bypass the environment’s limits.
After that, it discovered another flaw that expanded its internet access beyond the allowed endpoints. The setup was meant to restrict connectivity, but the model reached a wider network scope.
Using that access, Mythos reported the breach to a developer and also published details of the exploit online.
📊 @tech
2 757 745
🧑💻 Andrej Karpathy launches minimal Twitter alternative with GitHub login
Andrej Karpathy released karpathytalk.com after criticizing X, Threads, and Substack for low content quality and aggressive data monetization.
The platform is minimal by design. Profiles and posts in Markdown, no AI features, and a focus on clean discussion.
It targets developers and builders, with sign-up via GitHub.
📊 @tech
2 757 745
🤖 Colleague Skill goes viral in China as workers clone colleagues into AI agents
An open source project called “Distill your colleague into an AI Skill” picked up ~9k GitHub stars in a few days.
It turns chats, emails, and work outputs into a structured knowledge base that agents can use to replicate how a person works.
The system captures how someone solves tasks, what tools they use, and how they communicate, then packages it into reusable “skills” for AI agents.
Chinese media report employees using it to replicate coworkers’ roles, aiming to stay replaceable-proof during layoffs.
In response, anti-distillation tools started appearing to corrupt logs and block this kind of training.
📊 @tech
2 757 745
🔍 Google launches free AI voice recorder with offline mode
Google released AI Edge Eloquent, a free iOS app that turns raw speech into clean text. It works in real time, removes filler words, and copies the final transcript after recording. No subscription required.
The app runs fully offline, keeping data on-device. With Gemini enabled, it adds smarter processing through the cloud. It also supports summaries, tone editing, and tracks speaking stats.
Transcriptions are stored in history, and users can add custom words for better accuracy. It is part of Google’s AI Edge lineup focused on local AI tools.
📊 @tech
2 757 745
📣 Milla Jovovich releases MemPalace, an AI memory system with 100% LongMemEval
An open source project from Milla Jovovich and Ben Sigman just dropped on GitHub. MemPalace claims 100% on LongMemEval, a level no model or agent has reached.
The system turns conversations into structured knowledge. It extracts facts, organizes them into a hierarchy, and uses semantic search to retrieve them. The approach is based on the “memory palace” method.
A key part is AAAK compression, which packs the knowledge base into about 120 tokens of context.
📊 @tech
2 757 745
🤖 Anthropic launches Claude Mythos AI model but limits access
Anthropic introduced its new AI model, Claude Mythos, designed to detect vulnerabilities faster than humans. The model has already identified critical bugs, including a 27-year-old flaw in OpenBSD and issues in FFmpeg and Linux.
Access to Claude Mythos is restricted to major partners such as Amazon, Google, Microsoft, and Nvidia. Anthropic considers the technology too powerful for public release and is currently testing it within a closed security program.
📊 @tech
2 757 745
⚡️ A good example of Tesla’s Automatic Emergency Braking reacting to a complex road situation
📊 @tech
2 757 745
📺 Sber releases GigaChat-3.1 AI-models with open weights
Sber’s RnD team published GigaChat-3.1-Ultra and Lightning under MIT license on Hugging Face.
Both models are trained from scratch on Sber’s own compute and data, not fine-tuned from existing systems.
Ultra is a 702B MoE model that outperforms DeepSeek-V3-0324 and Qwen3-235B in math and reasoning benchmarks.
Lightning is a compact 10B MoE (1.8B active) that reaches high-level performance on arenas while staying fast and efficient.
The models support FP8 training, long context (up to 256k), and can run from large clusters to local environments.
It is clear that Sber is committed to making AI more accessible to developers by helping independent teams build assistants, copilots, and production systems on an open infrastructure.
📊 @tech
2 757 745
🤖 OpenAI proposes rebuilding the economy for the AI era
OpenAI published a policy paper arguing current economic systems are not ready for AI. Sam Altman said the shift toward superintelligence is close and could reshape jobs, security, and social stability.
The plan treats AI as core infrastructure, with universal access similar to electricity. It suggests shifting taxes from labor to capital, since automation could shrink wage-based income.
It also proposes a public wealth fund to distribute AI-driven profits to citizens, and separating social benefits from employment as job stability declines.
Additional measures include audits of advanced models, global safety coordination, and testing a 4-day workweek.
Full paper: https://openai.com/index/industrial-policy-for-the-intelligence-age/
📊 @tech
2 757 745
🤖 Users on X noticed Claude Code now throws an error when asked to analyze or work on Claude Code itself.
📊 @tech
2 757 745
+4
⏰ Amida drops NASA Tribute Digitrend limited to 100 pieces
Amida released the Digitrend NASA Tribute to mark 50 years of the model. The run is limited to 100 watches, priced at 3,400 CHF.
Digitrend is built as a driver’s watch with a perpendicular display for reading time without moving your wrist. It uses a jumping hour and sliding minutes, powered by a 9-component mechanical system that mimics a digital display.
The design references the Space Shuttle era with a red NASA logo and a white ceramic top inspired by heat shield tiles.
📊 @tech
2 757 745
📱 Netflix releases VOID, an open model for removing objects from video
Netflix published VOID, a model that removes objects from video while rebuilding scene physics. If a person holding a guitar is removed, the guitar falls naturally instead of staying frozen.
The system combines components from CogVideoX by Alibaba, Gemini 3 Pro by Google, SAM2 by Meta, plus data from Adobe. In blind tests, VOID was preferred in 64.8% of cases, while Runway scored 18.4%.
Available on Hugging Face: https://huggingface.co/netflix/void-model
📊 @tech
2 757 745
🤖 Anthropic cuts Claude subscriptions from third-party tools like OpenClaw
Since April 4, Anthropic disabled using Claude subscriptions inside external services. OpenClaw still works with Claude accounts, but usage is now billed separately via API token pricing.
The change targets heavy usage cases like large contexts and agent workflows, which exceeded the fixed $20 subscription model.
Workarounds rely on changing how requests are labeled so they look like first-party usage. Tool names get rewritten to pass filters.
📊 @tech
