Hacker News Reader

Comment by pffft8888 | Hacker News Reader

Imagine having ChatGPT level AI running in an ASIC inside earphones. This could be like an always-on buddy, available offline and able to access resources when you're connected.

Or in Google Glasses. The Readme states that it's more optimized for ASIC than the transformer architecture used by ChatGPT.

View on HN

People will use this to create their perfect self. AI will give them witty, charismatic answers to help them get a job, find a mate, entertain others. Eventually they will become AI zombie husks.

Did the name "Dixie Flatline" just pop into anybody's head?

https://williamgibson.fandom.com/wiki/Construct

Until this tech moves into our brains. AI wetware ushering in transcendence.

> Imagine having ChatGPT level AI running in an ASIC inside earphones.

Oh, it’s going in brains. And no, don’t sign me up.

How will you know which voice in your head is yours? Maybe you are both.

> How will you know which voice in your head is yours? Maybe you are both.

As someone with DID I wonder this every single day.

Maybe that will teach us about fragility of our definition of consciousness and agency. For some reason many people think that they are insulated from outside effects unless they’re right in their brain. Which is not true.

I wonder if we will see LLMs on FPGAs or FPGA like devices.

Crypto went from graphics cards to ASICs, we may see something similar with LLMs given the hype.

I always think of FGPAs for things you want to update without having to repurchase the hardware. If the FPGA inside a gadget can be reprogrammed over the Internet then I think it's more suited than an ASIC for LLMs.

It could go even further, in theory. The kind of ops that the current crop of LLMs needs is very simple, and at the same time there's no hard requirement for precision (which is why 4-bit quantization works so well). This means that unconventional approaches such as analog computing are potentially in the play again - it's easy to do addition and multiplication in an analog circuit, if you don't care about the answer being precise, and in theory one could pack a lot more of those circuits in the same space.

There's already at least one company with a product: Mythic AMP https://mythic.ai/products/m1076-analog-matrix-processor/

And there's several research papers about other architectures. For example: https://arxiv.org/abs/2205.05853

Would an ASIC really be better than GPUs? I'm not 100% sure but aren't low power GPUs essentially ASICs for matrix multiplication.

Have you watched the movie “Her”?

No, but I think an AI is going to be Them, by default, not Her. for a while, until we figure out if there is a neurological basis for gender, and if so, how to mimic it.

Don’t…read a book by its cover, my friend. Her is not particularly about the gender expression of the robot.

What is Them?

like Her, but there could be many or there could be one

Reply on news.ycombinator.com