Out of honest interest, what workflows have you integrated ChatGPT into?
577 karma · joined May 21, 2009
My biggest problem with AI safety is that, simply, the problem they envisage doesn't exist yet (generally, at a minimum, relying on the existence of "AGI"). Hence discussions about it have to make a huge amount of assumptions about a whole range of aspects of what the AI threat will be - what the AI will be capable of, what it's impact will be - before getting on to what possible solutions might be relevant to preventing it. But given the first two are so undefined, the later is pure speculation - one that is difficult to criticise directly, because any specific critisms can usually be easy deflected by adjusting any of the above assumptions without making a substantial change to the "inevitable" conclusion.
That's why it feels like an apocalypse cult to me - it's a conclusion, that has little strong evidence today, stacked on top of a constantly shifting set of assumptions, allowing adherents to avoid backing their arguments with evidence.
Then, more recently, I had to run a different project that lacked any documentation as to how I was to run it, had a setup.py file, a Pipfile, and more in it. In trying to get this to run, I managed to make a real pigs ear such that (no doubt thanks to my lack of experience with those tools) I eventually had to delete all my virtual environments, as none of them worked anymore…
So yes, I am 100% in the “one tool to rule them all” camp these days - and although PDM does look promising, right now it isn’t offering me anything above Poetry that I care that strongly about.
As Cato might say, PyPA delenda est.
However, that does mean they can attempt to “brute force” the encryption, trying any number of passwords as often as they like - and it seems some earlier versions of LastPass used rather poor choices with that cryptography, meaning the amount of effort needed to make an attempt is lower than other similar services (plus some users may have rather poor master passwords, making them easier to guess).
Many definitions of security include integrity and availability as a properties to protect - and storing data in multiple locations can definitely help protect both of those.
Except other national bodies like NCSC [1], and long-standing academic research e.g. [2, 3], that is!
1. https://www.ncsc.gov.uk/collection/passwords/updating-your-a...
2. https://dl.acm.org/doi/abs/10.1145/1866307.1866328
3. https://link.springer.com/article/10.1007/s10623-015-0071-9
Whilst I agree that she (and most other science reporters!) use simplistic models for immune system responses (possibly more for her readers), I’m yet to come across anything she’s written that has been very clearly wrong or she hasn’t later corrected/clarified.
1. https://www.nytimes.com/2021/11/18/opinion/covid-winter-risk...
Of course if you want to distribute it via their App Store you have to apply (it was free when I did it), and the way changing settings used to work (through a convoluted path on the connected mobile device and via the store) it was often best to do that. But you could also use the settings file software Garmin included in their IDE to do it (or now do it directly on the device).
The authorisation it talks about here is not the users - it’s their authorisation. The privacy concerns are the advertisers (yes, really - they claimed at one point some adverts include names and contact details for the advertiser, and hence constitute private information). That’s why they say “people’s privacy” without explicitly identifying the people they are protecting.
I frankly doubt there’s any sort of cost-benefit analysis being done here. Certainly in my experience it was much more driven by legal uncertainty and risk-aversion.
Secondly, being released under a permissive open source license definitely helped with its adoption. I was working as a senior developer in a UK Government department in 2013 and we had need for a full-text search engine for a project - and even before v1, ES was a contender, and it was eventually selected once v1 was released. This was largely due to a) ease of set-up/use and b) it’s release under an open source license. If it had been under AGPL, would we have still used it? Yes, probably - our specific use case wouldn’t have been affected by such a license, and the dept. was relatively open to more complex OSS licenses - but I have worked for several other orgs. where even just the AGPL would have resulted in a hard “no”.
Thirdly, ES has had contributions from a wide range of people. I honestly don’t know how I’d even begin to evaluate how much value ES has got from community contributions, but I feel it’s likely to be greater than the costs of managing those contributions.
But of course eventually Elastic got funding and had shareholders to placate. I don’t really have much sympathy for them about this conflict - their early choices were in part clearly made to maximise their value, and they decided to cash in on that value at a later date - the fact that those decisions had implications for their value should have been somewhat obvious to any investor who did any due diligence. I’m still not entirely convinced that the open source model is antithetical to commercialisation, but I think it does highlight how early decisions around OSS licensing can affect such processes.
The latter is particularly strange to me, since Douglas Adams was a huge fan of Wodehouse, used very similar humour, and even wrote forewords for Wodehouse re-printings.
1. https://eprint.iacr.org/2019/779 2. https://eprint.iacr.org/2020/823
Training their recognition model on a range of pre-recorded, staged audio sources would likely result in it being very good at generating subtitles - but not much else. As for the telephone, again, it's relatively different (the customer reps will likely be using a script in some cases, and are going to be putting a lot of effort into being comprehensible). And neither would effectively simulate the situations (for example, a noisy family home) an Alexa will likely be used in.
This may have been rectified since, and I'm still not sure how so many of these people haven't noticed that their "new" email address is not working - but I guess they put it down to spam filtering or similar.