New AI Models - Gemini 3.0, Claud 4.5 Opus, ChatGPT 5.2, Deepseek V3.2, Grok 4.2

[obsidian]

That's why I think it's laughable when people say AI will we the end of humanity.  I feel like I have to hold it's hand when it programs and give very simple defined tasks.  I catch it making mistakes all the time and have to correct it. 

Yes, it can blow me away, but also be comically bad at other times.  On the same day some Getbigger says that AI is going to end all life on earth, Claude can't get a button to work on a program with just one button.  I'm like, sorry, Claude can't end all life on earth today.  He's busy fixing the button on my one button app.  Maybe tomorrow.

AI is a tool, just like a hammer or a car — it’s meant to augment your skills, not replace them.

[obsidian]

The open source models from China and France are incredibly important.  I downloaded and run Deepseek R1 and will do the same for Deepseek V3.2 when it comes to Ollama.  If you look at the benchmarks, the open source models are not far behind the close source. 

When Deepseek R1 came out it was so devastating to tech that Meta fired 75 percent of their AI staff. Deepseek was that good compared to where they were.

The fact that any person or company can run these models on their own computer is huge.  I'm already planning building an AI computer.  There are multiple Youtube channels dedicated to building AI rigs for theses AI models.  It basically a gaming computer with multiple graphics cards, a massive power supply and a ton of memory.   

Silicon valley startups are focused on all the open sourced models because there is huge business opportunity there.

Which OS and llama are you using? GPU specs? I've thought about doing this, because you have more privacy if you run it locally.

[Palumboism]

Which OS and llama are you using? GPU specs? I've thought about doing this, because you have more privacy if you run it locally.

No Gpu, just a Ryzen 3 APU with built in graphics.  32 GB of Ram and Windows 11. 

It's not super fast, but it's fast enough.  It doesn't even need to be connected to the internet to run.  No data leaves my computer, absolute data security. 

It uses the command line, which for me is not as fun as your typical ChatGPT interface.  You can use it with Docker which duplicates the ChatGpt interface. 

I actually learned to install it from a Youtube video:

Ollama is an open-source tool that allows users to easily run large language models (LLMs) locally on their own computers (Windows, macOS, and Linux). 

IBM explains what Ollama is:

[Van_Bilderass]

A few of the analyst podcasters I watch have been complaining recently about fake AI videos made with their images and voices. I think Mearsheimer was being congratulated via mail about a superb speech he had given. It was news to him. Some were saying Youtube has to start policing this aspect.

[Mayday]

A few of the analyst podcasters I watch have been complaining recently about fake AI videos made with their images and voices. I think Mearsheimer was being congratulated via mail about a superb speech he had given. It was news to him. Some were saying Youtube has to start policing this aspect.

Been rice for over 5yrs to general media people. Musicians write so vs, so.eine creates AI music and copywriter then flag them for copyright so all ad revenue goes to them. Massive scam.

The reaction videos are mostly BS. No effort required by those people who react, all effort done by the original creators. I had my stuff ripped off for mountain biking (I did some drone downhill stuff), brands just pasted my stuff and I never got anything. No thank you, no discount, they take it, paste it with their brand name on it and that's all she wrote.

My Assumption is the media companies let it get this bad because AI will take over it all anyway. You can have AI podcasters talk about anything for an hour. Unlimited content.

[Hulkotron]

I've had enough of this shit.