@neil Yes and no. I use transcription models reasonably often - especially for generating meeting, podcast & video transcripts.

I struggle to follow complex information from speech and find it substantially easier as text, but a lot of content is only posted as YouTube videos or similar these days. So I end up using yt-dlp to grab the video, and whisper to transcribe it. I’m lucky enough to have the resources and know how to do this locally, but not everyone does.

@neil Reminds me of a line in BBC News at 6 tonight that sent a shiver down my spine (and made me think of you):

The Uk gov wants to use AI to calculate prison sentences 😱

@neil if the bursting happens quickly enough, I might get a market volatility notification from my banking app. Because my MSCI World ETFs might suddenly diminish in value.

@neil Yes. A large chunk of funny money would disappear from the "value" of my pension fund.

@neil

Maybe I would notice when the investment advisor called with a "Have you seen the news ..." type call 😟🤦‍♂️

Hopefully it wouldn't be too painful. If it's tolerable then I would celebrate.

@neil we regret to inform all customers that due to openAIs closure, your Samsung smart home appliance will cease functioning in 25 hours time.
We apologise for the inconvenience. We would like to offer all customers a complimentary voucher for £10 off a new Samsung home appliance when you spend over £1000.

@neil I'm retired and disabled and i'm sure it would affect the economy and make things even more difficult here and elsewhere.

But mostly i'd be curious as to what survived of it.

Specifically as a #Fedora #Linux user, i'd be interested in seeing how it affected #RedHat's business in general, but also linux #kernel development.

I don't think #AI is going away but the hype / investment is definitely a bubble.

https://mastodon.social/@transicorn/115530153385967421

@neil I'd have to do my own performance review instead of throwing last years through an LLM so that my boss can review it by throwing this years through the same LLM and see whether I'm doing well or not.

@neil Yes; it would have a tangibly negative impact on accessibility tools.

@neil I think quite a few Defined Contribution pension funds might.

@neil It would certainly be financially noticeable whilst people have a mad panic and the markets cratered, stopped panicing and recovered a bit.

@neil as others have commented I suspect my pension fund would take a hit. On the plus side, hopefully $dayjob could get some more GPU hardware at bargain prices so people could continue doing real science with them.

Been doing quite a lot of stuff to protect my savings from it eventually bursting, but likely the blowback is gonna be hard enough to hurt anyway. Work-wise, my customers would panic and abort projects. So, yeah, I would. I welcome it anyway.
@neil

@neil Yes, because the S&P 500 would tank and the business pages (which I read) would be all over it.

I would be pleased for many reasons though - most of all not having to answer the 'why haven't you adopted AI yet' question.

@neil I'm following enough shitposters that it would be impossible for me to not notice

@neil the peripheral noise around AI going away would be nice and noticeable.

But in a literal sense, hell yeah I'd notice it I'm never offline for long and the posting about it would be incredible. So many red line graphs

@neil I'd notice if people suddenly stopped complaining about it.

@neil Yes. because the fake money has been the only thing masking the recession.

@neil
No, but my 401k would.

@neil @jscholes

these tools are also widely used in health and social care to process assessments that are initially recorded as voice notes/interviews (its a bit like a better version of Dragon dictate, which seems to handle a very multicultural range of accents surprisingly well).

At my work everything gets verified by humans before its committed to patient databases / care plans, and no one has reported these things producing any glaring errors or causing issues with wrong data..

@neil I struggle to articulate my feelings on this topic, so hopefully this comes across as reasonably balanced.

An accessibility ecosystem has sprung up around LLMs. Much of it snake oil, some of it not.

In some cases the capabilities are new (or newly practical), such as visual descriptions for images/videos and guided photo taking. But they've also been able to improve or accelerate existing tasks, like information extraction from inaccessible documents or audio transcription.

A lot of discourse around LLM accuracy understandably takes an all or nothing tone, with any errors representing an automatic dismissal of the tools. But the situation ends up being more nuanced in the daily lives of disabled people. If someone's been dealing with inaccurate OCR engines and patchy access for decades already, any boosts in access can be seen as "better than nothing". And there are services which augment the "AI" with human verification and additions.