Some of the smartest people on the planet all in the same room, data at their fingertips… they randomly add it to the training set?
Labs at least must study prompts in an airgapped fashion. From there, consider how they could generate synthetic data to train another model. After, require trusted staff to do multiple levels of independent granular reviews of all fruits of the highest-value stolen inputs. (Or for model training data only, data never has to leave the airgap.)
Definitely risky, anyway. Surely some AI user has sent data, in confidential mode, with a unique shape they expect to be able to recognize if a later model recreated a facsimile even with heavy substitutions… but labs could bring risk of getting caught (over next few years) down quite low with extraordinarily ultraparanoid strategy. (But hopefully everybody is just behaving!)
Sounds like the boss's response was not to insist the proper procedure be followed, but to assert that the technology had to work as intended, and as soon as he figured out the issue on _his side_ the standard operating procedure could be followed.
And even if you spent a minute explaining the proposition to a user off the street, it still wouldn't be fair unless you laid out the drawbacks. Which leads me to a question.
There must be countless individuals all over the world who suddenly can't log into their Gmail or create any new accounts because a fraudster sent spam from their IP. I wonder: has anyone has tried to quantify that problem?
Has it become harder to find that information without setting off the fed’s alarm bells?
Was fascinated to learn the PRISM news reduced traffic to privacy-relevant Wikipedia articles, a chilling effect in that case, but indicates tech-savvy folks worry about doing anything on clearnet.
…then again since they’re using CLOUD models, guess my comment doesn’t make much sense…
Easy to send one’s clipboard to Microsoft Azure and have their DragonHD voices read the text, say with Keyboard Maestro (or presumably Alfred, Raycast, etc.). Should work with selected text too.
You’d definitely get to pay for it, not what I consider cheap. (“$15 per 1M characters”) But IMO just about best-in-class (maybe ElevenLabs has a voice I’d like even better).
Yeah, the app "tells you...whether today is a go", which is completely responsible and I'm not sure how HN could be able to miss it!
Jokes aside, couple changes come to mind:
- pitch more carefully as "may indicate when danger may be elevated"
- in app, on calm days insist "while prelim scan found nothing anomalous, click these links to find the original forecasts"
As of today, with this tech for this use case, the warnings should be the lede. Otherwise you're even more likely to have a "'Full' Self Driving" situation where we're lured into a false sense of safety.
Labs at least must study prompts in an airgapped fashion. From there, consider how they could generate synthetic data to train another model. After, require trusted staff to do multiple levels of independent granular reviews of all fruits of the highest-value stolen inputs. (Or for model training data only, data never has to leave the airgap.)
Definitely risky, anyway. Surely some AI user has sent data, in confidential mode, with a unique shape they expect to be able to recognize if a later model recreated a facsimile even with heavy substitutions… but labs could bring risk of getting caught (over next few years) down quite low with extraordinarily ultraparanoid strategy. (But hopefully everybody is just behaving!)