@JazzHandMeDowns@AndyXAndersen ChatGPT placed a tool in the hands of the ignorant. They don't know they lack correctness. So they merrily use them to write books or articles that are complete junk. The internet is now full of shit because it's much cheaper to get chatgpt to write an article than a freelancer
@JazzHandMeDowns@AndyXAndersen "The entire point of the thread is that people USED to be able to trust certain info & now they cant."
This for me is the problem. LLMs (and other probabilistic things) are super useful. People know weather forecasts are imperfect. But for now at least, they seem to trust LLMs.
@gwenshap One other thought I keep coming back to (and not just in relation to AI): why can't we have a global standard on privacy. GDPR was a start (any EU/UK company has to follow it globally), and it sets a reasonably high bar. So much money is wasted on compliance (often poorly).
@gwenshap Here's a summary of research: https://t.co/bEgS8qQbsE and this one about extraction is pretty interesting https://t.co/GEvDEnsuVN . The German DPA gave an interesting opinion on all of this. Model is not PII. Input and more importantly *output* can be. https://t.co/RLAyhofIlL
@gwenshap Models trained off that data are fine. E.g. medical models trained on anonymous data. There's no right under gdpr to forget that data. Llms are probably fine too if trained on sufficiently large dataset
@gwenshap This sounds very similar to gdpr. Anonymisation is key. So long as the data cannot identify an individual (including with additional data even if not in your possession) and assuming you're obtaining and processing the data legitimately, then you're ok.
@artiscoding @lospabloss @GergelyOrosz More pertinently, teams is fucking awful ux. Sad thing is, companies I've been at have moved to it due to needing office for working with corporate clients. Then it spreads like a virus until everyone has a license. And thanks to bundling, "why pay for Google, we have ms" ๐คฎ
@mipsytipsy ... and [when you're not greenfielding it] figuring out the implications and constraints of solving that problem in the context of the existing models
@johncutlefish Well yes. This article is very recognisable. Making the case for reducing WIP when the rest of leadership are fixated on 'doing more at the same time' and 'people just need to work harder' is really very challenging
@scottdwitt@sciolisticism@iTheRakeshP@rakyll I see a lot of SrICs focussing effort on things that will move the metrics tomorrow, not today. If they don't do this, tomorrow is going to suck. Good, experienced Srs know when to make the trade off. I find it hard to measure though
@GergelyOrosz It's no different to allowing jr devs to use stackoverflow 10 years ago. Or IDEs when I was at uni. Code review and/or pairing ensures learning and quality, not banning the tools
@MarioHachemer@rakyll@cramforce This feels a lot harder to navigate tho. "Why aren't you doing microservices... Oh we can't afford to treble your team"... "Make me some nfts... 2 days later: here you go, come back when you've sold some"
@MarioHachemer@rakyll@cramforce And it's not limited to developers. The number of non-technical people telling me how we should use AI without any clue what it implies is disturbing.
@mitchellh@mkheck I don't often have these issues in the UK. But the regional airport I fly from only gets 5-6 jets per day, and atc do a really good job lining everything up. Climbing out on rw heading to avoid noise sensitive areas at 80 can be painfully slow into wind