What is up with this? Chat GPT begins a response to my coding question with "Rock On"???

It isn't just the building that is expensive. Each 1GW of data center will cost about $9B/yr to operate between chip depreciation and energy costs. 17GW is significantly more power than NYC uses under peak demand. Even if they could afford to build it, operation would cost $150B/yr.

Does Elon think he has a cluster of a thousand arc reactors? 1TW could almost power the entire United States.

Clickbait is getting out of control. In my Youtube feed I see this, and I think, isn't that Natasha Lyonne? How could she possibly be homeless? Isn't her career going through a resurgence right now? Was she homeless at some point? Nope, it's about 4 people I've never heard of.

Someone online mentioned that whenever you ask an AI to generate a clock, it always sets the time to 10:10. Out of curiosity, I asked ChatGPT to create an image of a clock and this is what I got. The one that is weird to me is even the digital clock showed 10:10.

Does you have any details on the Meta's Hyperion project? Why would Meta need a datacenter the size of Manhattan that consumes almost as much power? Is this real or just vapor to impress investors? And what are "hurricane-proof" tents?

Working on a Nibbler like web game using #ChatGPT, #Gemini, and #CursorAI. Trying to get the #AI to work with a rendering engine I wrote is painful. GPT 4o is pretty good at creating pixel art, but I need to tweak a lot of it. I think AI improvements are needed for pure vibe coding of games.

I had some pretty good luck generating art of a generic asteroids game using the #openai 4o GPT model. I'm not having the same luck generating art for a snake game.

I made a little VS Code extension to use with my #Ollama server. It was pretty fun and easy to make. It would be nice if there were something a little more like #CursorAI that I could use against my Ollama server. Actually, what would be really great is if #CursorAI just let me use Ollama.

In recent sprite #gamedev I've had luck with K-means clustering for color reduction and lossless #webp. I was able to shrink a 112K .png sprite sheet down to a 64 color 21K .webp file. Often AI image generation creates unnecessary colors, which can be clustered. #openai 4o used for images

Nvidia has a new ComputeEval which a new method for testing how well LLMs can generate CUDA code. Right now #OpenAI o3-mini seems to perform the best. My understanding is that this is a new metric, so it is probably still pretty good, but I wonder how long it will take to reward hack it.

Playing with 4o to generate pixel art had definitely come a long way. Could still use some improvements. I added the lights at the bottom. they don't look as bad when this hasn't been scaled up. I probably need to spend a little more time on those.
#openai #LLM #AI

I set up an Ubuntu machine to do inference with Ollama. I would have thought that running a model would push the cards to use their maximum (or close to it) power consumption, but they only ran at around 50%. Is this a problem or a good thing?

Is that configurable?

A friend of mine had some #OpenAI weirdness. He asked #ChatGPT to generate a 3D model for him, and it actually seemed to think it created a link to a zip file. The link didn't actually download anything. I tried the exact same prompt and I got an explanation of how I could create the 3D model. Weird

Nothing makes you feel younger than seeing something like this at the Colorado History Museum: