Oprah will screw up the AI story
anildash.com
anildash.com
Quite possibly, but likely not as bad as this article.
Complete clickbait title, assumes that the author's hobby horses are the most important thing in the world, bizarrely argues that crypto hype is an "attack on labour".
>I am just in this moment where I'm fascinated by how this technology AI is going to be a resource for change and improvements and making the world better and yet the other side of it who's in control of it and what happens when the bad guys get it.
> There will be no mention of the repeated attacks on labor that have been systematically funded as “tech innovation” in recent years, including the gig economy, blockchain and crypto hype, and attempts to push the metaverse and thoughtless return to work.
Well, what's her show like? Covering why popular but terrible ideas are terrible is a thing, but I have no idea if it's her sort of thing.
> Maybe LLM assitants would also be vastly more reliable managers of pension funds, too? I'd be surprised to hear a word mentioned about that.
Various forms of passive fund are already a thing, and AIUI do approximately as well as most actively managed funds.
From what I understand, things unique to pension funds are that they tend to have too little money put into them in the first place, and that they have regulatory constraints about sticking to safe (and therefore lower-performing) investments.
Does Bill Gates have a big financial interest in AI? He does own 1.3% of Microsoft which is of course alot of money but people seem to still equate Bill Gates and Microsoft - Bill has sold down most of his Microsoft ownership.
Cuban is also a big crypto pumper.
50% of the time, what they describe as AI could simply be auto correct, the other half of the time it's this fictional all knowing machine.
I actually got t-shirts for an upcoming conference that just say "Ask me about AI" so I can explain to these IP attorneys what it's actually about. Luckily I've been increasing my participation(s) on these panels so hopefully we can start to dispel some classic pitfalls.
For a bit of context, here's my startup https://ipcopilot.ai/
It's quite plainly research aid software for attorneys, as we can also see from their comment re: listening to lawyers talk about AI as talisman.
Do you have anything to cite about this company that leads you to believe this?
As someone who'd love to see the whole blasted IP paradigm flattened from orbit, I couldn't be happier.
One of my many problems with IP as a whole was that it creates a metadata burden: you have to be very aware of what your source material is, who owns it, and under what terms they offer it, because there's always the risk you're pulling in enough material to create a risk of legal action.
This was always a risk for scalability, but it was easy to pooh-pooh for human-made productions. "If you write the Great American Novel/Codebase, you can always hire a few lawyers to manage clearance or redaction as necessary." A human probably has a modest possible pool of source material to work with. The typical developer hasn't read everything on GitHub so there's little risk of incidental infringement against 99.99% of the projects.
In contrast, the current nature of GenAI is explicitly a product with tens and hundreds of millions of pieces of source material, and there's no way to negotiate clearances on that scale on an individual, human-centric way.
Hopefully, what we'll get out of it is the mechanism for large scale mandatory licensing. OpenAI might pay a kajillion dollars to keep their models online, but the same tools might be available for someone who just wants to create smaller scale derivative works without having to fight with some author's estate who considers it a personal affront.
Stopped reading right there.
What does AGI mean to you?
> The intentional blurring of lines around the definition of “AI” — encompassing everything from improving lighting in camera phone photos to drone targeting for missile strikes — will be ignored as if it’s an unimportant consideration.
These are a bit odd to see in the same list.
> no discussion of public models of ownership or control of AI technologies
Is this talking about the ability to run models locally, or about government restrictions on permitted models? From his association with the EFF I'd think the former, but reading it gives more of a feeling of the latter.
> repeated attacks on labor that have been systematically funded as “tech innovation” in recent years,
What exactly is Total Factor Productivity, anyway, and what's the mechanism behind how technology leads to economic growth?
> not one voice representing labor will be quoted
Really? That's be like discussing the social impact of automated looms without mentioning Mr. Ludd. Seems quite the oversight to preemptively accused someone of without evidence.
> fact that young workers are adopting labor organizing faster than they’re adopting AI.
That sounds interesting enough to want a link for.
> The word “consent” will never be mentioned, let alone the concept, when it comes to training data and surveillance data used to build large-scale models
The justifications I've seen for demanding that model training on public data require consent seen to be less about "having 'your' data included will harm you in particular in ways that data from other people won't" and more about "these models are bad in general and here's a good way to obstruct them".
> Errors, confabulations, and intentional lies and distortions produced by generative AI products will not be described as bugs or defects, but by the false marketing term of “hallucinations”, with no mention of any legal or moral obligation to make commercial software that functions in reliable ways.
Calling these things "hallucinations" is (usually - there are exceptions like the early intentional nonsense from Google Gemini when it first started doing images) less inaccurate than the terms he's asking to use.
Also if you're building something with accuracy requirements there are long-standing techniques for building reliable systems out of components that are unreliable in various ways.
> There will be no mention of the fact that large learning models are being used as a method of determining who gets killed in military (and militarized policing) engagements today
What is this referring to?
> or how “the AI makes the decision” is a tool for avoiding accountability
Um, "computer says no" isn't new and isn't specific to AI.
> Not one person who has had a loved one targeted for sentencing, violence or death will be quoted or spoken to in any context.
What is this referring to?
> In short, Oprah is making a commercial for some AI vendors, without disclosing whether she has any financial interest
This is quite a statement to make about a specific individual with no evidence whatsoever.