News AI language model runs on a Windows 98 system with Pentium II and 128MB of RAM — Open-source AI flagbearers demonstrate Llama 2 LLM in extreme cond...

[Admin]

EXO Labs has demonstrated a modern LLM running on a 26-year-old Windows 98 PC.

AI language model runs on a Windows 98 system with Pentium II and 128MB of RAM — Open-source AI flagbearers demonstrate Llama 2 LLM in extreme cond... : Read more

 

[acadia11]

Very king cool! Something has to be done power consumption required to train todays models, … I wonder how precise their approach happens to be?

 

[bit_user]

Of course, a 260K LLM is on the small side

That's putting it mildly! If you look at the output it generated, it's like one step above complete gibberish!

​

In my mind, this utterly fails as a PR stunt, because the result is both useless and unimpressive. They should've gone for the largest model the machine could handle, unless that would've barely moved the needle on output quality.

Thus EXO hopes to "Build open infrastructure to train frontier models and enable any human to run them anywhere." In this way, ordinary folk can hope to train and run AI models on almost any device

Did he say the "any device" part? Because no - you need a lot of data, a large cluster, and lots of communication bandwidth for training. This isn't going to happen on "any device".

The way I read the "open infrastructure" comment is just that you would have a training infrastructure you could more easily port from one hardware to another, not that it's going to vastly lower compute requirements than existing training solutions.

 

[evdjj3j]

I wonder how much upgrading to a P3 which would add SSE would speed things up?

 

[bit_user]

I wonder how much upgrading to a P3 which would add SSE would speed things up?

The original SSE only added 4x fp32 operations. So, if they're using a lot of 8-bit integer arithmetic or similar, then you might not benefit from going beyond MMX, until you get to SSE2.

 

[Notton]

All this proved to me was that an LED light bulb seller can slap on the AI moniker so long as it has a processor and some storage.

 

[awake283]

This is just flat out cool.

 

[Mindstab Thrull]

Suddenly I wonder how much AI LLM you could get running on Knight Rider's KITT... from the 80s.

 

[awake283]

How how about

Suddenly I wonder how much AI LLM you could get running on Knight Rider's KITT... from the 80s.

Or this?