IrritableOcelot

Not somebody who knows a lot about this stuff, as I'm a bit of an AI Luddite, but I know just enough to answer this!

"Tokens" are essentially just a unit of work -- instead of interacting directly with the user's input, the model first "tokenizes" the user's input, simplifying it down into a unit which the actual ML model can process more efficiently. The model then spits out a token or series of tokens as a response, which are then expanded back into text or whatever the output of the model is.

I think tokens are used because most models use them, and use them in a similar way, so they're the lowest-level common unit of work where you can compare across devices and models.

Hmmmm milk is slightly acidic, and concrete will dissolve if the pH is lowered from its normal high alkalinity, so given a large enough volume of milk...I suppose milk would dissolve concrete substantially faster than water would.

There's going to be a temperature range somewhere between "fridge" and "corona of the sun" where that milk is the foulest-smelling thing in the universe.

I think its because while its under water it doesn't have a chance to diffuse into a larger volume of air -- normally farts are pretty dilute by the time it makes it to anyone's nose.

My favorite overheard undergrad story:

I was walking past the lecture hall right after an organic chemistry midterm, and there was a cluster of 4-5 students talking about the exam. One asked about question 8b, and another one said "you're not supposed to mix nitric acid and ethanol, that makes TNT, right?" I had to stifle a chuckle as I walked by.

So close, and yet so far! Nitrated acetone is explosive, and TNT (trinitrotoluene) is also made with nitric acid, but toluene is a much more complex molecule than acetone. If those undergrads could figure out how to turn acetone into TNT efficiently, they'd get a Nobel!

Agreed! I'm just not sure TOPS is the right metric for a CPU, due to how different the CPU data pipeline is than a GPU. Bubbly/clear instruction streams are one thing, but the majority type of instruction in a calculation also effects how many instructions can be run on each clock cycle pretty significantly, whereas in matrix-optimized silicon its a lot more fair to generalize over a bulk workload.

Generally, I think its fundamentally challenging to generate a generally applicable single number to represent CPU performance across different workloads.

Ehh, its not actually a big jump, and its oversold here. It's useful as an alternative to hand-designing genes, but its just a summarization tool for the gene sequences we've already annotated -- and a ballpark of 10-20% of those annotations are wrong, as it's actually very hard to annotate genes' functions correctly. I would be wary of trusting that this will work outside the most-studied protein families.

Lol I was trying to play Dragon Age games a couple months ago, and the EA app is so terrible that I couldn't get them to run on windows. But on Linux in the proton sandbox? No problem, worked right out of the box. 😂😂

I mean, sure, but largely GPU-based TOPS isn't that good a comparison with a CPU+GPU mixture. Most tasks can't be parallelized that well, so comparing TOPS between an APU and a TPU/GPU is not apples to apples (heh).

The changes to how account data is processed will effect anyone who logs in with their account. If you want to avoid changes, you cannot use a FF account.

Thats Britain lol, Ireland is on the left edge of the frame.

Why is that article so hard to read? Its not grammaticaly wrong but the sentences are structured so oddly...