#mastodonai — Public Fediverse posts

Có mô hình local nào phù hợp để chạy Clawdebot trên laptop cấu hình thấp? Spec: GPU GTX, Ryzen 5 5000, SSD 512GB. Xin gợi ý để tối ưu hiệu suất! #AI #LocalLLM #LowEndHardware #TốiƯuHiệuSuất #Clawdebot #MastodonAI

https://www.reddit.com/r/LocalLLaMA/comments/1qpg8as/which_local_model_is_best_for_clawdebot_in_a_low/

TheDrummer models được "thuần hóa" để giảm kiểm duyệt. 2 phiên bản: Magidonia-24B-v4.3 & Cydonia-24B-v4.3. Biến thể 1 giảm từ chối, biến thể 2 duy trì hiệu năng. #AILLM #MastodonAI #AI #MôHình

https://www.reddit.com/r/LocalLLaMA/comments/1prm2tq/thedrummer_models_meet_heretic/

Đang tìm đồng sáng lập Marketing (40% cổ phần) cho SaaS AI. Sản phẩm là chatbot AI giá rẻ, hoạt động ngay, so sánh được với Chatbase. Nhiệm vụ: định vị, GTM, phân phối; nội dung, SEO, mạng xã hội; thí nghiệm kênh thu khách trả phí. Ưu tiên có kinh nghiệm marketing SaaS/AI, đam mê thử nghiệm. #StartupAI #TuyểnDụng #MastodonAI #GrowthHacker #SaavnVietnam

https://www.reddit.com/r/SideProject/comments/1popoix/looking_for_a_marketing_cofounder_40_equity_for/

Sử dụng mô hình **Ollama địa phương** (ví dụ: *llama3.2*) để kiểm tra AI agents thay vì API đám mây. Ưu điểm: tiết kiệm chi phí, bảo mật dữ liệu và hoạt động ngoại tuyến. Cài đặt đơn giản qua **EvalView**: `pip install evalview`, kết nối Ollama với cú pháp YAML để đánh giá phản hồi AI. Dự án mở nguồn tại GitHub. Gợi ý thử nghiệm mô hình Ollama nào khác?

#AItesting #Ollama #AIBots #Llama3 #ĐánhGIáAI #CơChếTesting #PythonTools #AIĐịaPhương #MastodonAI #TechNewsVN

https://www.reddit.com/r/ollam

Cập nhật AI đa phương tiện tuần qua: Alibaba ra mắt LiveAvatar - tạo avatar âm thanh thời gian thực không giới hạn, Microsoft trình làng VibeVoice-0.5B - TTS 0.5 tỷ tham số chạy trên thiết bị cục bộ. ViBT 20B xử lý video/hình ảnh nhanh gấp 4 lần, Stable Video Infinite 2.0 mở nguồn tạo video dài hạn. YingVideo-MV tạo động tác ca hát từ chân dung, Reward Forcing cho video streaming tương tác. #AI #Alibaba #Microsoft #AIOpenSource #AIResearch #KHCN #AI2024 #MastodonAI #ViBT #LiveAvatar #StableVideo

Rhema: Công cụ học Kinh Thánh AI giống YouVersion, cho phép nhấn lựa chọn câu và nhận giải thích, ví dụ ứng dụng, câu hỏi thường gặp. Ứng dụng hiện ở giai đoạn thử nghiệm – cần feedback để cải tiến. #AI #BibleStudy #NghiênCứuKinhThánh #MastodonAI

https://www.reddit.com/r/SaaS/comments/1opybyc/i_made_a_bible_study_tool_like_youversion_but/

Ollama đang hiển thị 44% CPU usage thay vì GPU khi chạy model Mistral-Small3.2:24b. Người dùng bối rối vì khi chạy gpt-oss:20b lại hiển thị đúng 100% GPU. Có thể là lỗi cấu hình? #AI #Ollama #AIHardware #AIResources #TríTuệNhânTạo #Chatbot #MastodonAI

https://www.reddit.com/r/ollama/comments/1nozp8t/ollama_registering_44_cpu_usage/

How I AI

### Introduction

AI has taken off since ChatGPT came out. At first, it just felt like a more advanced autocorrect, but things have moved fast since then. Over time, I’ve developed accessible strategies for actually getting the most out of AI. In this article, I’ll cover the software I use and the strategies I rely on for effective use.

### Should you pay for the big services like Gemini or ChatGPT?

It really depends on your use case. ChatGPT Plus and Pro offer fairly generous usage, so if you do a lot of coding, it’s worth considering a paid plan. If you want AI integration on Apple devices, or want to use Gemini on Android, a subscription might make sense. But honestly, it’s a mixed bag—only pay if you actually need the extra features.

If you’re just looking to use AI occasionally, you can get by with APIs and a small investment—\$5 on OpenRouter and OpenAI would go a long way.

### Why would you use other AI providers?

There are a few reasons I use alternative providers for AI:

1. **NSFW and Jailbreaking:**
Google, Microsoft, and OpenAI really don't want you to use their AIs for NSFW (Not Safe For Work) stories or to create NSFW content. They also don't want you to try to jailbreak the AI to make it do things the company doesn't approve of. For these kinds of use cases, you may want to use a third party like [OpenRouter](https://openrouter.ai) or, from the web, mostly [Hugging Face](https://huggingface.co).
I have personally been banned for this. It doesn't happen often, but in my experience, when I contacted OpenAI, they weren't even aware that my account had been banned. So, just use a third-party service.

2. **Use of Different AI Models:**
There are over 400 models on OpenRouter or HuggingFace, so you can always find different models to try your queries on.

3. **Non-Chat AI Implementations:**
There are plenty of interesting approaches to using AI. Services like [ElevenLabs](https://elevenlabs.io/app/home), [Suno](https://suno.com), [Udio](https://udio.com), and even products like [RiverSide](https://riverside.com) use AI to create compelling features that set them apart.

4. **Saving Money:**
If you're not coding, you can save 60-70% by learning how to use different APIs. Even if you are coding, if you know what you're doing or use smaller outputs, it will be cheaper overall for you.

5. **Better Software:**
The software I use to access APIs is much more accessible than the official ChatGPT software, for example. You might prefer the available third-party software.

### What is the software you use, and where can I go to learn more about using APIs?

When I don't use the mainstream AI services, I usually use [BasiliskLLM](https://github.com/SigmaNight/basiliskLLM).
On macOS I use [VOLlama](https://github.com/chigkim/VOLlama/) or [Pal Chat](https://apps.apple.com/us/app/pal-chat-ai-chat-client/id6447545085).
I think Pal Chat is best on macOS.

If you want to learn about using the APIs and get a required API key, you can go:

* [Here for OpenAI](https://openai.com/api/)
* [Here for OpenRouter](https://openrouter.ai/docs/quickstart)
* [Here for Hugging Face](https://huggingface.co/docs/inference-providers/en/index)

I've only used Hugging Face from the web myself, but they offer their own APIs if you find software that works with it.

### What about other use cases?

There are official features for [using ChatGPT inside of an IDE](https://medium.com/@julian.goldie/code-without-coding-chatgpts-new-ide-integration-makes-developers-obsolete-d583858e7227), but I haven't used one of those.

I have started, however, using a utility called [Shell GPT](https://github.com/TheR1D/shell_gpt), which will let you chat with the OpenAI API version of ChatGPT, e.g.

```
sgpt "hello"
sgpt --shell "update windows, and applications using powershell."
```

After you run the shell version, it will display the command. You must hit `E` for execute, or `D` for description, or `A` to abort. Since it's in Python, it will work on macOS and Linux as well as Windows. I was able to change the version of GPT it used from 4O to 4.1 in the config file, so I'm sure it will be updated to work with newer models.

### What about on Mobile?

On mobile, I use primarily iPhone. I prefer [ChatGPT](https://openai.com/chatgpt/download/) and [Pal Chat](https://apps.apple.com/us/app/pal-chat-ai-chat-client/id6447545085).

I also have [PiccyBot](https://apps.apple.com/us/app/piccybot/id6476859317) for describing things, and [Be My Eyes](https://www.bemyeyes.com/), [Seeing AI](https://www.seeingai.com/), and [Ally](https://ally.me/) if I need a backup.

I think Ally is good enough if you don't want to pay for ChatGPT Plus, but the Plus video is faster than Ally at the moment. I'm sure this will change. I don't yet have Apple Intelligence on my phone, but I don't think it would radically change this strategy. Sometimes I want video, or I want to do something that those programs don't let me, so I don't think that's enough for me.

### Conclusion

Don't be left behind. There are plenty of accessible options for utilizing AI while saving a bit of cash. We have only scratched the surface here. I hope you learned something useful from this post and can have a little fun with AI at the same time.

---

\#️⃣ **Hashtags:**

\#AI #ChatGPT #Gemini #Accessibility #Blind #AssistiveTech #AIAccessibility #OpenAI #OpenRouter #HuggingFace #ElevenLabs #Suno #Udio #RiverSide #PalChat #VOLlama #BasiliskLLM #BeMyEyes #SeeingAI #Ally #AppleIntelligence #MobileAI #APIs #Python #ShellGPT #TechForGood #AICommunity #AIModels #TechTips #AIJailbreak #NSFWAI #IDE #Coding #MacOS #Windows #iPhone #Android #AIGuide #SaveMoney #DigitalInclusion #AccessibleTech #AIApps #ThirdPartyAI #NoLimitsAI #BlindTech #AIAssist #MastodonAI #TechAccessibility #FutureOfAI

Just got my hands on @ollama's self-hosted LLaMA! This post was actually generated by their AI model. Mind blown! Now I can engage with this game-changing language model directly from my own instance. Kudos to @ollama for democratizing AI access and empowering Mastodon users like me! #SelfHostedLLaMA #MastodonAI