Advertising on the Telegram channel «Yohannes_melese»

THIS WEEK AI NEWS

1. Fish Audio S2 open-source release — New TTS model (built on Qwen3-4B backbone) with insane fine-grained control: use natural-language inline tags like [laugh], [whisper], [super happy], [giggles], or [emphasis] anywhere in text for word/phrase-level emotion, prosody, and style. Supports multi-speaker in one pass, 80+ languages, multi-turn dialogue, sub-150ms latency streaming, production-ready inference (SGLang engine). Fully open weights/code on Hugging Face/GitHub — perfect for realistic voiceovers, podcasts, videos, memes, or fun cloning/experiments. Huge for creators wanting "absurdly controllable" voices without paid limits.

2. Perplexity Personal Computer agent — Turns a dedicated Mac mini into a local, always-on AI agent (Comet) with persistent file/app access, remote control for long-running tasks (research, planning, workflows). Waitlist rolling — privacy-focused "personal Jarvis" that runs unsupervised on your hardware.

3. Claude adds interactive charts/visualizations (Imagine beta) — Claude now generates/tweaks custom charts, diagrams, graphs, and visuals in-chat (on by default or prompt it). Great for data, reports, presentations — auto-decides when to build one, or request directly.

4. NVIDIA Nemotron 3 Super open model — 87GB open-weights model optimized for multi-agent reasoning, coding, long tasks. Downloadable on Hugging Face — strong for custom/local agent builds or high-performance tinkering.

5. Olmo Hybrid (Allen AI) — New 7B fully open model family with hybrid architecture (transformer + linear recurrent layers). Hits same MMLU accuracy as Olmo 3 but with ~2x data efficiency (half the training data/tokens). Free/local run — big for efficient small-model experiments or on-device use.

6. Microsoft Copilot Cowork — Agent for reading/analyzing/editing local files (Excel, docs, etc.) with multi-model selection. Competes in "AI coworker" space — handy for file-heavy workflows.

7. Google Gemini Workspace upgrades — Pulls from emails/chats/files/web to generate full docs/spreadsheets/presentations via natural prompts. In Workspace — turns Drive into active AI knowledge base for productivity.

8. OpenClaw ecosystem buzz (post-acquisition) — After OpenAI pickup, spinoffs and integrations exploding (e.g., Moltbook agent social network acquired by Meta). More autonomous agent tools chaining for real actions in apps/browsers.

9.
AlphaFold-powered personalized pet medicine story — Viral case: user + ChatGPT sequenced dog's DNA mutations, ran through Google's AlphaFold to design custom cancer vaccine → 50% tumor reduction in weeks. Highlights real-world gen AI for health/creatives (human trials coming).


10. ByteDance pauses video AI model launch — Suspended due to copyright disputes (The Information report) — impacts creative video gen tools in apps, potential delays for short clips/reels.

🚀 Dataset.ET Update

The community response has been amazing.

📊 Current Stats

👥 538+ users
🎧 1,500+ audio clips recorded

🤖 The Telegram bot is now smarter and can: (shout out to @dataset_et_bot)
• create challenges
• track progress
• answer questions
• post stats & leaderboards

🎁 New: starting today contributors receive small rewards for helping build the dataset. We don’t have funding yet, but any future support will go directly back to the community.

Please share Dataset.ET with your friends, family, and university groups. You don’t need tech skills — just your voice.

Let’s build AI that understands Ethiopian languages 💪

I want to remind everybody how important @dataset_et_bot is, not just for technical AI builders, but for everyday non-technical people too.

Even if you’re not in tech, what @chapidevtalks did gives you a way to play a role in the AI revolution. By sharing your voice ( literally) you help your language, culture, and country get represented in the AI race.

@b1n1yamBuilds

🚨 አዲስ የ AI ዜና፡ Fish Audio አዲሱን የ"S2" (Text-to-Speech) ሞዴል ይፋ አደረገ! 🎙️✨
Fish Audio በድምጽ ቴክኖሎጂ አለም ትልቅ ለውጥ የሚያመጣውን S2 (Fish Audio S2) የተሰኘ አዲስ ሞዴል በነፃ (Open-Source) አቅርቧል።

ይህ ሞዴል ጽሑፍን ወደ ድምጽ (TTS) በመቀየር እስካሁን ካየናቸው ሞዴሎች ሁሉ እጅግ የላቀ እና ስሜትን በትክክል መግለጽ የሚችል ነው!
ዋና ዋና ገጽታዎቹ (Key Features):
🎭 የላቀ የስሜት መግለጫ (Emotion Control)፡ በጽሑፋችሁ መሃል እንደ [ሳቅ], [ሹክሹክታ], [ቁጣ], ወይም [በደስታ] የመሳሰሉ ትዕዛዞችን በማስገባት ሞዴሉ ያንን ስሜት ተላብሶ እንዲያነብ ማድረግ ትችላላችሁ።


🗣️ ፈጣን የድምጽ ቅጂ (Voice Cloning)፡ የሰውን ድምጽ ከነሙሉ ስሜቱ እና አነጋገሩ ኮፒ ለማድረግ የ 10 ሰከንድ ጥራት ያለው የድምጽ ሪከርድ ብቻ ነው የሚፈልገው!
🌍 የብዙ ቋንቋዎች ድጋፍ፡ ከ 50 በላይ ቋንቋዎችን በከፍተኛ ጥራት ያነባል።

⚡ እጅግ ፈጣን (Low Latency)፡ ጽሑፉን እንደሰጣችሁት በቅጽበት (ከ150ms ባነሰ ጊዜ ውስጥ) ድምጹን ማመንጨት ይጀምራል፤ ይህም ለቀጥታ (Real-time) ውይይቶች ተመራጭ ያደርገዋል።


👥 የብዙ ሰዎች ውይይት (Multi-speaker)፡ በአንድ ጊዜ የሁለት እና ከዚያ በላይ ሰዎችን ድምጽ በማቀላቀል የውይይት (Dialogue) ቅርጽ ያላቸውን ኦዲዮዎች መስራት ይችላል።


ምን አዲስ ነገር አለው?
S2 ሞዴል በውስጡ 4 ቢሊዮን ፓራሜትሮች (4B parameters) ያሉት ሲሆን የ 'Dual-Autoregressive' ቴክኖሎጂን ይጠቀማል።

ሞዴሉን በቅርቡ ከተሰሩት እንደነ Google እና OpenAI ዝግ (Closed-source) ሞዴሎች ጋር ሲነጻጸር በድምጽ ጥራት የተሻለ ውጤት እንዳስመዘገበ ተነግሮለታል።

🔗 ሞዴሉ አሁን ላይ በ GitHub እና HuggingFace ላይ ለማንኛውም ዴቨሎፐር በነፃ ይገኛል። በተጨማሪም በቀጥታ በድህረ-ገጻቸው fish.audio ላይ ገብታችሁ መሞከር ትችላላችሁ።

ይህን አስገራሚ AI ሞዴል ገብታችሁ ሞከራችሁት? ሀሳባችሁን በኮመንት አጋሩን! 👇

🚀 Introducing Dataset.ET

The future of AI should speak Ethiopian languages.

Today we are launching Dataset.ET — an open community initiative to build the largest dataset for Ethiopian languages.

Why this matters:
Most AI systems today barely understand Amharic, Afaan Oromo, Tigrinya, Somali, and many other Ethiopian languages. Without datasets, AI will ignore our languages.

Dataset.ET is changing that.

What we are building:
• Speech datasets
• Text datasets
• Community-validated language corpora
• Open infrastructure for Ethiopian AI

How you can help:
1️⃣ Record sentences using our Telegram bot
2️⃣ Validate recordings from other contributors
3️⃣ Help expand datasets for Ethiopian languages

No technical skills needed — just your voice and your language.

🤖 Start contributing:
https://t.me/dataset_et_bot

🌍 Learn more:
https://dataset.et

💬 Join the community:
https://t.me/dataset_et

Together we can teach AI to understand Ethiopia.

📢 Community leaders & channel admins:

If you run a Telegram group or community and believe Ethiopian languages should be part of the future of AI, we would really appreciate it if you could share this post with your community.

Your support can help thousands of people contribute their language and voice to this initiative. Thank you!

🚨 አስቸኳይ፡ በ AI ዘመን ቋንቋዎቻችን ከቴክኖሎጂው ዓለም ሊሰረዙ ነው! 🇪🇹
ዓለም

በ AI (Artificial Intelligence) በፍጥነት እየተለወጠች ነው። ነገር ግን አንድ እጅግ አስፈሪ እውነት አለ፡ አብዛኛዎቹ የ AI ሲስተሞች አማርኛ፣ አፋን ኦሮሞ፣ ትግርኛ፣ ሶማሊኛ እና ሌሎች ቋንቋዎቻችንን ጭራሽ አያውቁም።

ዛሬ ይህንን ዳታ ካላዘጋጀን ነገ ምን ይፈጠራል?

በህክምናው ዘርፍ ህይወት ይጠፋል፡ የነገው ዓለም በ AI ዶክተሮች (AI Doctors) የሚመራ ነው። እነዚህ ሲስተሞች የገዛ እናት እና አባቶቻችንን ህመም በቋንቋቸው መረዳት ካልቻሉ፣ በተሳሳተ ምርመራ (Misdiagnosis) ምክንያት የብዙዎች ህይወት በአሳዛኝ ሁኔታ ያልፋል።

ብሄራዊ ደህንነት አደጋ ላይ ይወድቃል፡ ወደፊት የደህንነት መዋቅሮች እና የጦር መሳሪያዎች ሙሉ በሙሉ በ AI የሚቆጣጠሩ ይሆናሉ። ቋንቋችንን የማያውቅ AI ሲስተም መከላከያችንን ሲቆጣጠር፣ እኛ በገዛ ሀገራችን ላይ ምንም አይነት መብትም ሆነ መከላከያ አቅም አይኖረንም።

ፍጹም የቴክኖሎጂ ባርነት (Digital Colonization)፡ የራሳችንን የ AI ዳታቤዝ ካልገነባን፣ ትውልዱ የሌሎች ሀገራት ቴክኖሎጂ ጭፍን ጥገኛ ይሆናል። ዓለም ወደፊት ሲገሰግስ፣ እኛ በዲጂታሉ ዓለም ሙሉ በሙሉ ተሰርዘን እንቀራለን።
ይህ የህልውና አደጋ እንዳይከሰት፣

አንድ በጣም አስፈላጊ ፕሮጀክት ተጀምሯል።

ዓላማው የኢትዮጵያን ቋንቋዎች ለ AI ማስተማር ነው።
ነገር ግን ይህ ያለ እናንተ ድምጽ ፈጽሞ ሊሳካ አይችልም።

ነገ በቴክኖሎጂው ዓለም እንድንረሳ እና ህይወታችንን አደጋ ላይ እንዳንጥል፣ ዛሬውኑ 5 ደቂቃ ወስዳችሁ የአፍ መፍቻ ቋንቋችሁን ለ AI አስተምሩ።

👉 አሁኑኑ ይህን ቦት ነክታችሁ ድምጽዎን ይቅረጹ:
https://t.me/dataset_et_bot
ምንም የኮዲንግ እውቀት አያስፈልግም—ስልካችሁ እና ድምጻችሁ ብቻ በቂ ነው። እንዲሁም ሌሎች የቀረፁትን በማረጋገጥ የህልውናችንን ትግል ማገዝ ትችላላችሁ።

AI ኢትዮጵያን ችላ እንዳይላት ጊዜው አሁን ነው!

⏳
🌍 ድረ-ገጽ:
https://dataset.et

💬 ግሩፕ: https://t.me/dataset_et

This week ai news

1. OpenAI launches GPT-5.4 & GPT-5.4 Pro — New reasoning/coding/tool-use upgrades across ChatGPT, API, and Codex. Better on docs, spreadsheets, software workflows — smoother professional tasks in-chat. Available now for users.

2. Google rolls out Gemini 3.1 Flash-Lite & Pro updates — Faster/lighter versions with improved multimodal and reasoning. Gemini app gets quicker responses, better for everyday quick tasks/search/creative.

3. Perplexity releases open-source embedding models — Efficient, low-memory models for AI search/retrieval (multilingual, quantized). Great if you're building or tinkering with custom search/tools — free on Hugging Face.

4. Google expands AI Mode Canvas to all US users — No more Labs waitlist: draft docs, generate code, build interactive tools right in search with real-time web data. Turns search into a productivity workspace.

5. OpenAI Codex Security agent — New tool scans your code for vulnerabilities (finds issues like missed bugs). Early tests show it catching stuff overlooked for months — handy for devs/power users in ChatGPT/Codex.

6. Criteo integrates as first ad tech partner in ChatGPT ads pilot — Conversational ads testing in Free/Go tiers (US first). Brands show up in chats — keeps free access funded, higher conversion potential for discovery.

7. NotebookLM (Google) adds cinematic video overviews — Turns notes/docs into narrated video summaries with visuals. Fun/engaging for study, reports, or quick content creation.

8. Apple Music AI transparency tags rollout — Labels/distributors must disclose AI-generated content (artwork/tracks/videos). Voluntary now but heading mandatory — affects tools like Suno if you're making music.

9. DeepSeek V4 teased/early buzz — Trillion-param efficiency gains (1.8x inference speedup). If it lands in apps/tools soon, expect cheaper/faster local-ish high-end performance.

10. Claude surges on App Store (user migration from ChatGPT) — Growing popularity as alternative (better trust vibe for some). Easy switch guides out — if you're shopping for chat agents, Claude feels stronger this week.

Week's vibe: Iterations on reasoning/tool-use (GPT-5.4, Gemini updates), search/productivity blending (Canvas expansion), and open-source bits (Perplexity embeddings) making everyday AI snappier/more useful. Agent security & creative overviews popping up too.

Which one are you checking first (GPT-5.4 in ChatGPT, Canvas in Google Search, or something else)?

This week ai news

1. Nano Banana 2 (Google DeepMind/Gemini) — New image gen model with Pro-level quality at Flash speed, search grounding (pulls real web refs), better subject consistency, configurable thinking levels, up to 4K resolution. Rolling out in Gemini app and tools — sharper, faster creative visuals from prompts.

2. Perplexity Computer — Autonomous long-running AI agent that executes extended workflows, scheduled tasks, and multi-step projects on your computer. Handles real actions over time — big for hands-off automation in research/planning.

3. Claude Cowork / Code updates (Anthropic) — New focus on workflow execution: scheduled tasks, better memory persistence, remote control options, multi-agent teams for office stuff. In Claude app/Pro — agents now handle ongoing work like file edits or reports without constant prompting.

4. Grok 4.2 beta (xAI) — Native multi-agent architecture (4 specialized agents collaborate/debate/synthesize), cuts hallucinations ~65%, weekly user-feedback tweaks. Available to premium users on web/mobile — more reliable answers for tricky/complex questions.

5. Claude customizable enterprise plugins — Job-specific plugins let Claude act directly in tools like Excel, PowerPoint, Google Drive, Gmail (autonomous multistep actions). Power users train custom ones — seamless for work tasks in chats.

6. Google Lyria 3 music gen — Turn text, photos, or videos into 30-second tracks with lyrics/auto-gen beats. Integrated in Gemini app/Labs — quick fun music creation from anything.

7. Qwen 3.5 series expansions (Alibaba) — Agent-focused upgrades (cheaper/faster multi-step workflows), 200+ languages, better text/image/video handling. Via Qwen app/API — stronger for autonomous agents/tasks in chats.

8. Cursor cloud-based autonomous Agents — Run agents in isolated cloud VMs, get video demos of code changes/tasks completed. Handy for coders/power users automating dev work.

9. Hermes Agent (Nous Research open-source) — Multi-level memory + remote terminal access for personal agents. Free/local run — privacy-focused tool-building for custom automation.

10. Samsung Galaxy AI multi-agent push — Perplexity agent added ("Hey Plex" wake), deep integration in Notes/Clock/Gallery/Reminder/Calendar for seamless multi-step tasks. On flagships — more natural phone AI helpers.

This week's vibe: Agents going longer/autonomous (scheduled tasks, computer use, workflows), creative multimodal boosts (images/music), and multi-agent collab everywhere. February wrapped with a ton of usable upgrades hitting apps — feels like AI is finally "doing stuff" reliably for daily use.

Which one sounds most fun to try (Nano Banana 2 images, Perplexity Computer for tasks, Lyria music, or Grok's multi-agent answers

This weak Ai news

1. Google launches Gemini 3.1 Pro — Major reasoning upgrade (double performance on tough benchmarks like ARC-AGI-2), better coding/multimodal/science tasks, same price as before. In Gemini app/API — stronger for deep thinking, long planning, SVG/code gen from prompts.

2. xAI releases Grok 4.2 beta — Native multi-agent setup (4 specialized agents collaborate/debate/synthesize answers), cuts hallucinations by ~65%, weekly feedback-based tweaks. Available to premium users on web/mobile — snappier, more reliable responses for complex queries.

3. Anthropic ships Claude Sonnet 4.6 as default — Faster/cheaper than Opus on many tasks, improved coding/long-context/computer-use (navigates interfaces like spreadsheets/web forms). Switch in Claude app — great for everyday office/coding without premium cost.

4. Alibaba drops Qwen 3.5 — Agent-focused model, deploys agents 5x faster than rivals, handles text/images/video in 200+ languages, multi-step workflows (forms/websites). Via Qwen app/API — cheaper autonomous agents for tasks/research.

5. Google adds Lyria 3 music gen in Gemini app — Turn text/photos/videos into 30-second tracks with lyrics/auto-gen. Fun/creative boost for quick music clips/podcasts.

6. Samsung expands Galaxy AI multi-agent ecosystem — New Perplexity agent integration ("Hey Plex" wake phrase), deep ties to Notes/Clock/Gallery/Reminder/Calendar for seamless multi-step tasks. On upcoming flagships — more natural device AI helpers.

7. OpenAI GPT-5.2 Instant style/quality update — Smoother responses in ChatGPT, plus expanded context window (256k tokens for Thinking mode). Noticeable in daily chats — better flow/memory.

8. Cohere Labs releases Tiny Aya family — Compact 3.35B multilingual models (70+ languages), runs locally/on-device. Download on Hugging Face — privacy-focused small AI for phones/laptops.

9. Google rolls out shopping ads + link popups in AI Mode — Sponsored placements in conversational search (product discovery), hover popups with summaries/images for sources. In Gemini/Search — more useful shopping/research with quick web jumps.

10. Anthropic measures Claude Code autonomy surge — Sessions now run autonomously longer (45+ min for top cases, nearly doubled recently). In Claude Code — agents handle bigger tasks without babysitting.

11. Manus agents hit Telegram — AI agents for chats/tasks, now in Telegram for easy mobile use. Quick setup — personal agents on the go.

12. Claude Co-Work / Excel & PowerPoint integrations — Agent teams + direct editing in Office tools. Pro users — automate spreadsheets/presentations in-chat.

13. Ollama adds subagents + web search in Claude Code — Run local subagents, search web inside sessions. For Ollama users — offline + online agent power.

14. Gemini app tier updates (Pro/Ultra) — More audio overviews/deep research reports, Nano Banana image editing. Check your Google AI plan — expanded creative/reasoning limits.

15. ChatGPT Voice instructions/tool use improvements — Better following commands, tool calling in voice mode. Rolling now — hands-free chats more capable.

Week's vibe: Models pushing agentic/multi-step autonomy hard (faster/cheaper agents everywhere), multimodal creative tools (music/video from anything), and everyday integrations (browsers/devices/Office) making AI feel like real helpers. No single mega-drop, but tons of usable upgrades hitting apps this week.

Which one are you trying first (Gemini 3.1 Pro reasoning, Qwen 3.5 agents, Grok 4.2 multi-agent, or something creative like Lyria music)?

Guys አሁን እየሞከርኩት ነበር Tigrinya  and ኦሮምኛ music

ምንም አይነት ግጥም ሳላስገባ
በዚህ prompt ብቻ

Prompt: create traditional Ethiopia Tigrinya /oromic music

እውነቴን ነው ምላቹሁ ሁለቱንም ቋንቋ ባላውቅም music is good