Claude is great for code and apps - I do not use it anymore for text as it watermarks
Chat GPT currently does not watermark - so now I have another AI expense
Haven’t looked into the Chinese models… anyone with experience using them?
Thanx
-Mike
Why are you using the big AI, for pay models for simple tasks? Download LM Studio, install the Gemma 4.0 model (Google’s), and run your AI locally. QWEN 3.8 (Chinese open source) is also quite good.
Didn’t want to tie up the Mac mini m4 pro - will look into those thanx Thom
Use a lot of disk space?
Got 64gb ram…
So you choose your model as to how much storage to use and how much RAM it will use. QWEN 3.8 27B is 16GB in size. Seems to run fine on my 64GB MacBook Pro as LM Studio initially configures it, but still exploring.
Gemma 4 31B is 18.85GB in size, and it is said to be optimized to keep memory use down. Haven’t yet run anything through it yet, though.
Wow ½ hour of research later… omg
Had no idea this existed…
Thought you needed a large computer - hard to implement - not worth it….
Learning quickly - many models I’d like to add - didn’t know this was possible …
This is amazing - thanks for opening my world 
You should see the response when I did, in real time, a speech for the Minister of Tourism in Botswana. In Setswana. On my laptop. Draws drop.
I thought this was only for large systems… had no idea it was so accessible now…
Well, it would be nice if media could keep up with what’s really happening (said someone trained in media).
I think it will be nice for designing sites to be able to do both Elements and a LLM that understands MCP on the same machine, without having to keep going out to the net (I often work where there isn’t a net). But at the moment I would do heavy-duty programming with the local models. Claude is much more efficient at a full programming task, particularly when done from scratch.
Hi, I am using Perplexity Max which kind of selects the correct AI model for the task I am throwing at it. From creating images to elements components to accessible checking of websites or adding the necessary htaccess and scripts. Absolutely amazing and saving me tons of hours of time or external specialists.
Definitely hot 
computer is running 99% GPU - only 20 GPU cores on the M4 - the CPU cores are barely touched… now I understand why people want 80 GPU cores
For me - so far - let it work in the background and do other things - this takes time
First run redoing three articles - 30 minutes so far
Mind you - using a heavy model - there are others that would blast it out quickly - which I haven’t downloaded or used (yet)
Fan takes off periodically - have Mac Dan Control for extra cooling power… and a raised stand and a small fan blowing on the mini
Learned from all my years of teaching - heat kills electronics
Really really nice to have my own local AI
Just finished 
let’s see what it made…
Edit not done - now making each re-write
Lost track… an hour later? Came back with three very good re-writes…
Very good - local control etc lots of benefits - just takes time to create - but again using a big model also..
Interesting first run….
Well - I asked it for research also - find the 2-3 staccato articles and rewrite
So it was doing a lot more research and work than I realized…
Find several Staccato articles → read them → understand them → synthesize them → rewrite them.
So I’m a little crazy 
trying to install ComfyUI and LTX 2.5 and try a five second video clip…
My thought is:
Script a music video - create 8 second clips overnight - assemble clips into complete video by morning
Testing models - time and rendering first - but first need to get through hugging face and installations
Hi Mike,
What do you mean that Claude watermarks text?
Never mind Mike, researched it. Watermarking is being driven by Article 50 of the EU AI Act which now requires it. 200 companies have signed the EU Code of Practice on Transparency of AI-generated Content. Open AI signed it but has not yet implemented. It is likely to in the near future.
So the practical landscape right now: Claude marks text, Gemini marks text, ChatGPT doesn’t yet but has committed to, Grok isn’t committed at all, and local open-weight models like LM Studio are outside the system entirely.
glad you figured it out Ed
got all the bits to make an image into a movie - we’ll see if it errors or not…
If it went from 4% → 8% in about 7 minutes, you’re looking at roughly:
- ~4% every 7 minutes
- ~14–15 minutes per 10%
- Potentially 2½–3 hours for the whole generation
So the computer is going nuts fans a movin’ - been up all night going to bed and see what I get in a few hours - chat gpt says the win was getting it started and working - we can optimize later
5 seconds of video at 3 hours = pain point for 180 second music video
Goodnight from Illinois or good morning
I’m afraid all the LLM’s are going to start watermarking. Claude was just the first to do it.
As long as the content is good and helpful to the reader I don’t see why Google should care, but Google does many things I don’t understand…lol
Hopefully some will not watermark and be different in the marketplace…
Update: switched video resolution - did 10 seconds in 20-25 minutes
Video is possible!!!
The issue with having AI write content is simple: it’s essentially just taking known patterns and repurposing them. Consider this: take all the book authors in the world and tell them they can only use previously published books for sentence and source material. Yeah, no new information gets exposed that way, and essentially the value of creating AI content declines, as it starts seeing its own content as patterns to repurpose.
Historically, we’ve seen stale-ization happen in pretty much every medium. Then someone tries something new and different and breaks through that with fresh content. I don’t have a lot of time left on the planet (I’m 74), but I’m not going to use AI to generate content; I’m going to try to present fresh content until my fingers stop typing, my mouth stops working, and I can’t play any of my instruments or tools any more.