Forgot your password?
typodupeerror

Comment Re:Open training data matters too (Score 3, Insightful) 28

Yep. Open weights is closer to having precompiled binaries only (though you can do a little bit more with open weights). All training data with labeling, plus a detailed build log including any special fine-tuning logs, is needed, especially to help show how the final weights were arrived at (since there can be randomness involved). Unless someone else can both replicate your final output plus make their own changes to it, it's not fully open source.

Comment Re:Hmmm.... (Score 1) 47

We're entering the exciting age of AI-powered cybersecurity attacks. I imagine it won't be too long before something like a malware worm which polymorphs to evade detection and adapts itself as new CVEs get published, able to utilize tuned edge models, shows up. Wireless IoT could mean we end up with physical "e-quarantine" zones, wherein devices keep attacking anything in radio range...

Comment Re:TLDR (Score 1) 114

The price thing is important. What I suspect is that when companies can no longer subsidize bot use via investor funds and per market-share fights, users will balk at prices and look for cheaper alternatives, which may result in better model compression and/or overseas commodification, meaning the US data-centers won't be competitive, and those investors will get screwed, possibly triggering a Wallstreet crash.

Slashdot Top Deals

Vital papers will demonstrate their vitality by spontaneously moving from where you left them to where you can't find them.

Working...