Thea Vales

ai without the hype cycle. just what works.
Biz inq 📧: [email protected]

[Thea Vales @codesthea]

Dave Chappelle says no one needs to think anymore because of AI: “Thinking is for poor people”

“I was talking to a friend of mine and we were laughing about how no one really would know what to teach a kid now. What skill sets are going to matter 10 or 20 years from now as the world becomes automated”

“We decided your character is your destiny. If you’re a good person that’s all you can ask from anybody and hope for the best”

“I don’t know if anyone needs to think anymore. It’s AI, quantum computing, it seems like thinking is for poor people”

[Kumail Sikandar @Aiwithkumail]

NVIDIA has lost it..

Chinese researchers open-sourced a model that writes CUDA better than humans experts.

And it completely rewrites the economics of AI hardware.

Writing low-level CUDA kernels to squeeze every ounce of performance out of a GPU has historically required elite, highly specialized hardware engineers. Standard AI models have always bombed at it, falling far short of traditional compiler systems.

Until now.

A joint team from Tsinghua University and ByteDance published "CUDA Agent”, a massive, large-scale agentic reinforcement learning system built to master GPU architecture.

Instead of relying on static prompts or simple multi-turn bug fixing, they built a closed-loop environment with automated hardware verification, profiling, and synthetic data pipelines.

The model learned how to write parallel, high-performance GPU code through trial, error, and reinforcement learning at scale.

The benchmarks are staggering:

It didn't compete with standard tools. It delivered 100%, 100%, and 92% faster execution rates over PyTorch's compiler on KernelBench Level-1, Level-2, and Level-3 splits.

On the brutal Level-3 benchmarks, it outperformed proprietary giants like Claude Opus 4.5 and Gemini 3 Pro by about 40%.

NVIDIA's moat has always rested on two pillars: elite hardware, and the proprietary software lock-in of CUDA.

If an open agentic system can automatically discover, write, and optimize production-grade CUDA kernels better than human specialists, the software moat starts evaporating.

The hardware matters less when the software can rewrite the metal itself.

[Samaira AI @Fatima_RP120]

respectfully, you've done one gen and formed an opinion. Wan 3.0's free to start on Qwen Create at https://t.co/iI2ppsov0u, go and do ten.

[Ram Singh Verma @RamSingh_369]

Leave Netflix tonight.

Watch this 2 h 34 min Stanford class.

It's the clearest, most complete, and brutally honest explanation out there on how ChatGPT and Claude are really built.

From Tokenization and BPE to the Transformer architecture, the training pipeline, and the next-token decoder. No fluff. No marketing. Just the truth.

Doesn't matter if you've never touched a line of AI code or if you spend your days launching Agents: by the end, you'll suddenly connect a ton of pieces you've been trying to fit together for years.

The real core boils down to this:

How text turns into numbers the model can "eat" (BPE tokenization)
The sole mission of a language model: predict the next token
How the Transformer uses Attention so tokens can pass messages to each other
In training, the NLL loss pushes the probability of the entire sequence
In generation, the decoder builds the response token by token

The big-picture view that takes most people years to form… this class delivers it to you complete in one sitting.

Free up your time.
This could be, no exaggeration, the most valuable class you watch this month.

[Kyronis @kyronis_talks]

Two Bulgarian friends killed the entire streaming industry.

It's called Stremio + Torrentio. You get 4K content from Netflix, Disney+, Hulu, and HBO Max combined for free.

Here's how it works.

Stremio is the player. Clean interface. Works on Windows, macOS, Linux, Android, iOS, and TV. You install it once and it looks like any other streaming app.

Torrentio is the addon. You add it to Stremio in one click. It scrapes content from every major torrent provider on the internet simultaneously and delivers the best available stream directly to your player. 720p, 1080p, 4K. You pick the quality. It finds the link.

→ No account required
→ No subscription
→ Works on every device
→ 4K and HDR supported
→ Subtitles built in

Netflix cannot shut this down. There is no central server to seize. No company to pressure. No domain to kill. It runs on your device and pulls from the open internet.

The entire streaming industry is built on one assumption. That you will keep paying $70/month rather than spend 5 minutes on GitHub.

That assumption just died in Sofia, Bulgaria.

MIT License. 100% Opensource.

https://t.co/zMbGW2UpJA

Get the addon here: https://t.co/DZQS4wVIEF

[Thea Vales @codesthea]

Apple is no longer hiding it. These machines are built for local AI.

Apple unveiled the new Mac Mini with M6 their first 2nm chip. Starting at $899.

What you can run on the $899 Mac Mini (32GB) 👇

→ Qwen 3.6 27B : fits comfortably, near-GPT-4 quality
→ Qwen 3 Coder 32B : full coding assistant on your desk
→ Llama 4 8B at full precision
→ DeepSeek R1 32B for reasoning tasks
→ 13.5x faster LLM processing than M1. 4x faster than M4.

What you can run on the Mac Studio with M5 Ultra (512GB) 👇

→ Llama 4 70B at high quality quantization
→ DeepSeek V4 with hundreds of billions of parameters
→ Frontier-scale models entirely in local memory
→ Connect four Mac Studios over Thunderbolt 5 cluster them to run trillion-parameter models locally

Apple said their Mac business grew 30% last quarter largely because people are buying these to run AI models on their own desks instead of paying for cloud tokens.

The future of inference is local. Apple built the hardware for it.

[Thea Vales @codesthea]

@Marryclaire_AI @idextratime Seedance 2.5 on Flova is the first time AI video starts looking like an actual production pipeline instead of a clip lottery.

[Marry claire @Marryclaire_AI]

With Seedance 2.5 in Flova, creators can build longer, more coherent sequences without prompting every shot from scratch.

Flova takes AI video beyond isolated generations and into a controllable, All-in-One production flow.

A quiet suburban horror story follows a lonely child, a voice behind the door, and a slow reveal that something is not human.

tutorial：

- Enter your idea and desired style into the Agent chat box, and ask it to write the script for you. Then click Generate.
- Flova will generate a detailed written script and let you review the Prompt Draft before generation, helping you avoid wasting credits on outputs you don’t want. After confirming, you can choose the models you want to use, such as Seedance 2.5** for video, Image 2 for images, and Suno for music, then generate step by step.
- You can leave comments to revise anything you’re not satisfied with, and save the visuals or characters you like for future use across projects.
- Export the finished video. Members can export without a watermark.

[Hayyan | Creative AI @hayyantechtalks]

China just open-sourced a compact OCR model that can process a full 100 page PDF in a single pass.

It’s called Unlimited OCR. Only 3B parameters. Runs entirely on your own hardware.

Most OCR tools still slice documents page by page and lose context across tables, formulas, and reading order. This one treats the whole document as one continuous pass.

What it actually does:

- One shot long horizon parsing with a 32K context window

- Multilingual support out of the box

- 93%+ on the standard parsing benchmark (about 6 points above the DeepSeek OCR baseline)

- Error rate stays under 0.11 even past 40 pages

- Completely offline

- Works with Transformers, vLLM, SGLang, Docker, Ollama, and llama.cpp

Cloud OCR (Textract, Google Vision, Azure Document Intelligence) usually costs $1.50–$15 per 1,000 pages.

This one runs on your machine for free, for as long as you want.

Baidu built it specifically to push past DeepSeek OCR. It already has millions of downloads on Hugging Face and is still flying under the radar for a lot of people.

100% open source. MIT license.

Link:

https://t.co/MUq6IFlAmN

If you deal with long PDFs, scanned reports, contracts, or research papers, this is worth trying. Local, cheap, and actually keeps the document together.

[Future Logic @the_futureAI]

🚨 BREAKING: Claude can audit your YouTube channel like a strategist.

Most creators only look at views.

That’s a mistake.

Here’s a better way to use Claude to find what’s actually holding your channel back: 🧵👇

[Perks @AiTechPerks]

Beyond Prompts: Flova Introduces an All-in-One, Agent-Native Video Workflow

Bring your boldest ideas Flova can turn every spark of inspiration into a creative, cinematic cola commercial. No complex prompts required. Just activate a master-level Skill with one click and start creating like a pro.

Detailed steps：

- Enter your idea and desired style into the Agent chat box, and ask it to write the script for you. Then click Generate.
- Flova will generate a detailed written script and let you review the Prompt Draft before generation, helping you avoid wasting credits on outputs you don’t want. After confirming, you can choose the models you want to use, such as Seedance 2.5 for video, Image 2 for images, and Suno for music, then generate step by step.
- You can leave comments to revise anything you’re not satisfied with, and save the visuals or characters you like for future use across projects.
- Export the finished video. Members can export without a watermark.

[Jay Bisen @JayBisen473370]

I surf the web daily,

and Here are 20 incredibly useful websites you probably didn’t know existed.

Trust me, you’ll want to bookmark these… 🔖

[Cooper Tech Hub @CooperTechHub]

Best YouTube Channels to learn AI in 2026:

1. AI Explained
👉 https://t.co/EA8Ko64HyE…

2. Andrej Karpathy
👉 https://t.co/LUCj2O231Z…

3. Cole Medin
👉 https://t.co/gtHjE58mAB

4. DeepLearningAI
👉 https://t.co/HDTqH7yQhe…

5. Futurepedia
👉 https://t.co/yFs7MEAwDf…

6. Matthew Berman
👉 https://t.co/sxyy2CFTkw…

7. Skill Leap AI
👉 https://t.co/7leRribHmO…

8. Tech With Tim
👉 https://t.co/2wcdDfHmQH…

9. Tina Huang
👉 https://t.co/xzL35tcIAA…

10. Two Minute Papers
👉 https://t.co/dmP9HtJQz1…

If you’d love to see more content like this, don’t forget to follow @CooperTech_UK for more AI magic.

[Jawad Rahman @Jawad_Rahman_]

Imagine finding a hidden spot by typing the way you actually talk:

“Find me somewhere scenic, chill, dog-friendly, and perfect for golden hour.”

@wanderboat_ai finds the vibe → you find the spot → and Tesla owners can use Wanderboat right in the car and navigate there.

From inspiration straight to navigation.

@wanderboat_ai
Try it now: https://t.co/675DQQQ9HM

[Iris Hayes @irisneural]

Wan 3.0 just entered the arena and I put it to the test.

I turned the viral “niulai” meme into an animated adventure, comparing the newly launched Wan 3.0 with Seedance 2.5 and MiniMax H3 inside Flova.

One idea. Three leading models. One Agent-native workflow.

How does Wan 3.0 compare? Watch and decide

[Kevin Parker @kevin_parker_ai]

Your files aren’t gone. They’re just trapped in chaotic directories named 𝐟𝐢𝐧𝐚𝐥_𝐟𝐢𝐧𝐚𝐥_𝐯𝟑, old hard drives, scattered cloud accounts, and endless camera rolls. 😅

If that sounds like your digital life, download MUZIM and sign up before Sep 30 to claim 1,000 bonus Credits.

Learn more here: https://t.co/ufoxwyateL

The irony is, our brains don't work like traditional file systems.

I rarely remember exact filenames.

I remember “𝐭𝐡𝐞 𝐝𝐨𝐠 𝐈 𝐦𝐞𝐭 𝐨𝐧 𝐭𝐡𝐞 𝐫𝐨𝐚𝐝” or “𝐭𝐡𝐚𝐭 𝐬𝐮𝐧𝐬𝐞𝐭 𝐛𝐲 𝐭𝐡𝐞 𝐬𝐞𝐚.”

That’s precisely why @𝐦𝐮𝐳𝐢𝐦_𝐨𝐩𝐞𝐧𝐬𝐨𝐮𝐥 grabbed my attention.

MUZIM functions as a 𝐥𝐨𝐜𝐚𝐥-𝐟𝐢𝐫𝐬𝐭 𝐀𝐈 𝐟𝐢𝐥𝐞 𝐚𝐠𝐞𝐧𝐭 𝐚𝐧𝐝 𝐩𝐞𝐫𝐬𝐨𝐧𝐚𝐥 𝐝𝐢𝐠𝐢𝐭𝐚𝐥 𝐦𝐮𝐬𝐞𝐮𝐦, designed to comprehend your library rather than forcing you to dig through endless folders.

You can run a Core Vibe Search entirely offline by memory, watch related files automatically group together, bring in selected items from supported cloud apps, and transform chosen moments into shareable formats. Cloud importing requires internet; it functions as an import, not a backup or live sync. 🔍✨

Ready to push it further? Connect your own Claude or GPT API keys (optional 𝐁𝐘𝐎𝐊) to trigger AI workflows while keeping raw files entirely local, or allow authorized tools to leverage context via 𝐌𝐂𝐏. These specific workflows operate online, and Agent/AI tasks might utilize Credits. You can also give your setup a physical base with the 𝐌𝐔𝐙𝐈𝐌 𝐋𝟏 𝐃𝐚𝐭𝐚 𝐇𝐮𝐛—boasting up to 24TB of SSD storage alongside a 10Gb/s high-speed dock. ⚡

Maybe we don’t need to build better folders.

Maybe we just need a library that inherently gets us.

𝐘𝐨𝐮𝐫 𝐟𝐢𝐥𝐞𝐬. 𝐘𝐨𝐮𝐫 𝐦𝐞𝐦𝐨𝐫𝐢𝐞𝐬. 𝐘𝐨𝐮𝐫 𝐫𝐮𝐥𝐞𝐬.

[Kumail Sikandar @Aiwithkumail]

A Chinese company just dropped a robot dog that switches between wheels and legs on its own.

Wheels on flat ground. Legs on stairs, rocks, streams, cliffs. It decides in real time. No human input.

25 kg. Carries 16 kg while walking. Runs at 6 meters per second. 30 km range on one charge. Works in -20°C to 55°C.

The demo shows it crossing streams, climbing cliff faces, and running through mud like nothing.

[Harsh Pawar @HarshBisen143]

Earn up to $3,000 monthly with Google, but not many are aware of this opportunity.

Ready to jump in? All you need is:

→ Internet access
→ A smartphone or computer
→ Time (2-3 hours)

Want the free guide? Here’s how:
1. Like
2. Comment "Google"

(Follow, So I'll DM you)

[Hazris Quinn @Hazris_AI]

Chinese models are already 112× cheaper than Anthropic per million tokens—right now.Chamath just laid it bare on CNBC: a “barrel of intelligence” costs $56 from Anthropic, $26 from OpenAI, $1.50 from Meta, $1 from xAI and Google… and a mere $0.50 from Chinese models.This isn’t a pricing quirk. This is the fastest, steepest commodity collapse in history—and it’s happening in real time.Oil took 40 years. Semiconductors took 20. AI inference is crushing those timelines in months.The companies stuck at $26 and $56 aren’t stupid. They’re desperately buying time, praying that trust, safety, and enterprise contracts can hold the premium a little longer while their costs scramble to catch up.But the clock is already running out.They cannot bet on the timeline—because the $0.50 model isn’t a demo. I’ve been inside the labs building it. It’s here.

[Kyronis @kyronis_talks]

A guy scanned an entire house with his phone. Uploaded it.

Now anyone on Earth can walk through it in a browser tab. No app. No VR. No agent. No appointment.

Click → you’re inside. Every room. Every angle. Every shadow. Photoreal.

The numbers are insane:

- Agent fee on a $500k home: $15,000
- Cost to make this scan: ~$200
- Time to “tour” 50 houses: one evening
- File size: smaller than a TikTok

The science is wild too:

It’s called 3D Gaussian Splatting instead of polygons (how games render), it uses millions of tiny glowing “splats” of color and depth.

AI reconstructs reality from your photos. The result loads on a phone and looks like you’re THERE.

The grift opportunity is even wilder:

Freelancers are already charging $300–$800 per scan for realtors, Airbnbs, venues, car dealers, museums.

One person + one phone + one weekend = a business.

100% Open source. Built on PlayCanvas.

[Holland Tech Marco @tec_marco10]

Kids animation is the most underrated money niche on YouTube 💸

No face.
No voice needed.
No trends.

Just stories + consistency.

One channel:
721K subs
24 videos

This is not luck.
It’s niche selection.

Want the blueprint?
Comment “KIDS” 👇

Follow +Bookmark must 🔖