AI is most important tech advance in decades
bbc.co.uk
bbc.co.uk
It's an alien, a remote expert system, that 0.1% of the 0.1% of AI practioners have even the faintest real-ish understanding of. It is a fantastic expert, of deep experience, but it is an idiot savant, and no peer in lifting us up as well.
I have never been sadder to see a technology. I lived through the seemingly incredibly short term of Personal Computing which seemingly is at it's final ends, first eroded by online-services & Service-as-a-Software-Substitute (SaaSS), and now fatalitied in a Mortal Kombat style by idiot machines fed on the immolent corpus of the internet of posts. A system which seems all too destined to radically outstrip the pop-culture-will-eat-itself idea with a new ultra-hyper-self-consuming internetworked online-content-will-be-begat-by-AIs-begat-by-online-content Oroborous.
None of this shit informs. None of it makes clear. None of it raises human consciousness. It's the best monkey on a type-writer the universe-we-know has ever seen, but it's utterly without soul or character or purpose, is a refined regurgatory mimic devoid of purpose. This is the worst thing to happen to technology.
And humanity will spend the rest of my life obsessing over it & fascinated with it. It's all bad magic. It's all against-the-cause. This is the worst shadow-puppet machine humankind has ever invented, and it's visage chains us to the wall. Do Not Want.
Ask it to inform and make clear, ChatGPT can do explanations.
Prompt: "I have heard people say "c'est ne pas un pipe" has deep meaning, can you explain simply to me?"
Answer: "The phrase "c'est ne pas un pipe" is a French sentence that means "this is not a pipe." It is a famous statement by the Belgian surrealist artist René Magritte, who painted a picture of a pipe with the words "Ceci n'est pas une pipe" ("This is not a pipe") written underneath. The meaning behind the phrase is often interpreted as a commentary on the nature of representation and reality. Magritte's painting is not actually a pipe, but a representation of a pipe. The painting can never be smoked, and therefore it is not a pipe in the literal sense. The phrase "C'est ne pas un pipe" challenges the viewer's assumptions about the relationship between words, images, and reality, and invites them to question the nature of representation. In other words, the phrase suggests that there is a difference between an object or idea and its representation, and that the representation is not the same thing as the real thing. This idea can be applied to many aspects of life, such as language, art, and even our perceptions of ourselves and the world around us."
It connected my broken French with the correct sentence, translated it, put it in context of who said it and when, and expanded into a coherent and correct explanation of some deeper meaning. That's a better explanation than a monkey-on-type-writer or Markov chain simple token-predictor has any right to give, isn't it?
> "it's not intellectual augmentation, it doesn't extend how we think, it thinks for us"
Ever had the experience where you recognise a famous person's face on TV but can't think who they are, wait for a while, and then the answer 'comes to you'? That is "thinking for myself" but it's not me actively doing anything, it's magic information retrieval that happens to me. I'm not sure what the act of thinking consists of ("when I went to bed, I lay awake, doing something I call thinking but probably is not." - JOB, Robert Heinlein); what do you have in mind for a machine 'extending how we think'?
This is a topic that the academic & internet worlds have both tread a million times over. That the AIs can find some relevant selections in it's memory banks & regurgitate some synopsis does not impress me.
And worse, none of this alien intelligence's conclusions are in any way original or distinct: they're all blank boring averages. A ranked selection of the variety of existing models that it's embedded within itself. I could not be less impressed. I could not be sadder to see such vacuous crap hyped.
> That is "thinking for myself" but it's not me actively doing anything, it's magic information retrieval that happens to me.
The case you describe is just recall. What is exceptional about humans & intellect isn't recall, but connection, & association. I too am not sure what the act of thinking consists of, but in almost all cases I respect there's at least some reminisce of the influences, some distillation of points of view where ideas came from. And machine learning is bereft of this point of view, is it's own weird expert universe, that does nothing to express itself usefully. It's all barren. It's stochastics. Without referent. Shame on us.
Do you want a shitty lo-fi hyper-reality? This is how you get a shitty lo-fi hyper-reality. Where regurgitation of prominent symbols overtakes the real. By having idiot machines who have no conception of real fictate to us their idiot perspectives.
It’s just such a strange time right now. I’m certainly impressed with the technology-side of things but it just leaves me feeling flat. Meanwhile at work today a manager broke the beginning silence of a meeting with “so who’s gonna try out the new Google Bard?”; blink, table flip. But also HN audience is probably way oversaturated on this stuff by now.
“This is not intelligence”
> "The case you describe is just recall"
Natural language recall, over a memory bank larger than any human could ever memorise.
> "they're all blank boring averages"
Average of a near-infinity of topic you didn't know about is a significant improvement over how things were.
> "What is exceptional about humans & intellect isn't recall, but connection, & association."
"I refuse to be impressed until it's as exceptional as humans" is not an interesting take, it's a cynicaler-than-thou emo teenager take.
> "Do you want a shitty lo-fi hyper-reality? This is how you get a shitty lo-fi hyper-reality. Where regurgitation of prominent symbols overtakes the real. By having idiot machines who have no conception of real fictate to us their idiot perspectives"
The Sunday Times bestseller "The Martian" by Andy Weir is fictating to us an imaginary idea of surviving on Mars, in a way that would never work in reality. That didn't stop it being popular, enjoyable, influential, or make it a 'shitty low-fi reality where prominent symbols like the word Mars overtake the reality of the planet'.
it will get better, and it will get better faster and faster. this is what terrifies most people the most. its limitation is mostly resources. If we hooked it up to a robot and funds... we'd have a Matrioshka brain / Dyson Sphere probably by the end of the century.
I can really foresee a situation coming in the not too distant future where scarce amounts of high quality, properly verified information are very highly regarded and shared amongst experts in fairly closed loops.
Remember captcha ? Which is a huge source of training material for self driving cars ? Yeah that’s gone.
Newspapers with verification chips built in so you know it’s from a trusted printing press.
really?
I honestly don't believe this was ever a thing. it makes no sense at all.
I'm fine being wrong here. But just being a mean dispirited bastard rejecting my premise out of spite & providing zero helpful indicators is a barren shitty pathetic attempt to dismiss me. Say something real we can talk about, show some possibility. Of substantiate-able substance.
There is literally nothing concrete I can reply to in your fluff. Maybe your experience really means something & is real. But there's no case made for or against that. You haven't made any case for yourself. There's no clear correlating examples. You've crowed up a belief, without support or cause.
So far I'm not sure which if any of these systems even can look at your code. What have you seen that can?
Someone has to present a case. My only case so far is that there is no case, there is no value, this is un-creationist & regurgatory.
We can both accept that my premise might not be right. Who has the obligation here to present evidence? Is having no good evidence but maybe no clear negatives an acceptable counter to my claims? Are there more forward stances I need to consider & deal with? Or is your whole premise about as it seems, a shade-throwing you-don't-know "trust-me"? What would possibly satisfy you, how far would I have to go trying stuff before you could say, OK, yeah, your right, it's indeed predominantly mostly bunk?
Who do you think has the decision? Do you have evidence to contribute? Is lack of anti-evidence, the possibility of maybe all you want? How do we decide this?
> There is literally nothing concrete I can reply to in your fluff. Maybe your experience really means something & is real. But there's no case made for or against that. You haven't made any case for yourself. There's no clear correlating examples. You've crowed up a belief, without support or cause.
> So far I'm not sure which if any of these systems even can look at your code. What have you seen that can?
I think the replier has been quite concrete, and I think it would be quite clear what they are referring to if you have given these tools an earnest try. Nevertheless, let me give what I've been using them for:
I've used copilot to autocomplete boilerplate.
I've used ChatGPT to start brainstorming gift ideas, to look for documentation, to look for suggestions for study resources, and to get (always incorrect here and there) code examples of usages of public code libraries. In all these cases I can also get somewhere with some focused thought and research, but AI helps me do it faster with less brainpower. Then one's work is lightened just like how an unreliable student or intern can lighten a workload. This is what I suppose the replier means by autocomplete, where the discovery (which can be hard) as well as easy work is done mostly right for you by the AI.
Actually, in my experience, these tools are better as assistants and as smart rubber ducks than in helping to produce good code, where IDE tools still feel superior.
I'm pretty sure the EXACT Same argument was made after the invention and proliferation of the assembly line and industrial mass production.
> It's an alien, a remote expert system, that 0.1% of the 0.1% of AI practioners have even the faintest real-ish understanding of
Just because a novel new technology seems alien to you, and people are still figuring it out when its brand new doesn't mean it's ungrokkable or that we can't train and develop ways to better understand, design, use, diagnose, etc AI systems.
> None of this shit informs ... it's utterly without soul or character or purpose,
Outright false and demonstrates a profound misunderstanding of what Machine Learning is and can be used for. Again, just because you don't understand doesn't mean no one else does or can. A tool does not need a soul to find a purpose, and the purpose you find for it need not be the purpose that it was originally intended for it to still be of use and provide value.
> And humanity will spend the rest of my life obsessing over it & fascinated with it.
Yes, welcome to technology. This particular one is going to change society, in big ways, for better or worse, but is absolutely extremely useful, purposeful, and full of potential.
My experience with the modern tools is that they are capable of assisting with many tasks in relatively surface level ways. It’s still extremely impressive but it hasn’t done anything truly new or creative. GitHub Copilot helps with boilerplate, comments, and tests, but it hallucinates so much that I leave it disabled a good 50% of the time. ChatGPT will throw words at me when I’m blocked writing song lyrics, which never gives me anything usable but it does help me think from different perspectives. They can help rapidly spin up a prototype, using “understanding” of docs and prompts to write code. But none of it can be trusted because of its hallucinations, none of it makes me fear for my job. It’s a utility akin to autocomplete, or intellisense, a super advanced compiler.
(Somewhat aside, it’s telling that we live in an age where the press and the industry are falling over themselves raving about ChatGPT while we’re still struggling with troll farms and spam bots manipulating discourse. We shouldn’t be surprised that many can’t differentiate between extremely capable word association coming from code and humans crafting ideas when we’re unable to recognize bad actors deliberately engaging in bad faith arguments online.)
My downgraded interpretations and expectations aside, I honestly do think it’s exciting and brilliant, I reach for it all the time, and I can’t wait for a more capable Siri and better integration into productivity tools. But it’s not intelligence and I think we should be careful to separate hype from reality.
If AI helps us focus on Robotics, Spaceships, and underwater cities, I am game.
Does LLM help "Social Media, Influencer culture, Creators, and a Ad-tech driven surveillance" or "Robotics, Spaceships, and underwater cities" more?
That is my prophecy.
So I guess GPT-4 may still be most useful at destroying the modern information economy, thanks to being an infinite generator of garbage indistinguishable from content marketing. But GPT-5 might be just good enough to enable some qualitative breakthroughs in robotics, spaceships and underwater cities...
(NOTE: this is not be believing that OpenAI solved consciousness. This is me recognizing the surprising ability for generative language models to navigate abstract concept space, just by virtue of being exposed to enough association implicitly encoded in everything we say or write...)
I think it does help those. You've seen how the early models had to be taken off because teenagers trained models to be bigots. If there's a widely-deployed model that uses near-contemporary data, it would be the prime target for all kinds of promotion, influence, disinformation and product placement just like how Google became one such. The reasonable, useful replies that current models give you may become a thing of a past once people start to target LLMs for the same things that they today target Google's search, and you will have the same SEO arms race, albeit probably harder to combat considering how much more opaque LLMs probably are compared to Google's search algorithms.
Perhaps the good useful information that we get from LLMs today are more a product of them being trained on paid content of high quality that aren't widely distributed.
Without the ad-driven Internet forcing the world to generate so much content, garbage or otherwise, I'm not sure if and when we'd have generative models.
I wonder if the Internet Archive are aware of the importance their job is about to gain. They're not just maintaining an archive of the Internet - they're maintaining an archive of pre-generative AI Internet, meaning they effectively hold what's about to become the main source of pristine, human-generated content for training corpora.
I think this is "the promise" but I don't think it will be the outcome, like many things out of Silicon Valley.
Remember when Facebooks mission was to "connect the world", wow, what a fail.
Remember 3-4 years ago when we were all going to live in a magical symphonious world brought to eternal harmony with blockchain built into everything? Now with AI actually promising a world like that (if things go right) people don't seem as excited as when the killer app for the new tech was pictures of apes with hash codes pointing to their urls.
Obligatory reminder that modern public knows shit all about anything on aggregate - they get excited over things the advertisers tell them to.
Cryptocurrencies were somewhat special, because by enabling a bunch of get-rich-quick schemes (rehashes of old fraudster tricks, for the digital era), they became awash with money that could be used to lure more victims. A positive feedback loop.
It doesn't challenge our norms, it is designed (maybe unintentionally) to repeat them. It repeats them in a friendly, safe, unchallenging voice; and with precision and completeness that signals to us real knowledge.
How well do we evaluate its response when it keeps telling us, in its safe voice, 'you're so right boss! That's brilliant!'
Always with the poor countries. How does he know this won’t totally wreck poor countries ?
Climate change is going to absolutely ruin poor countries long before they need an AI doctor which gives false information.
Meanwhile wheat does he think these AI doctors run on ? Air?
Fossil fuels.
So the end of fossil fuels will probably wreck rich countries, too. Especially given that no country is remotely getting prepared for that.
Right now we can develop photovoltaic and wind farms in a world that has a lot of fossil fuels. But remove the fossil fuels today, and everything collapses. The question is therefore: can we transition to clean energy before the end of fossil fuels? And the answer is really not clear. Actually it seems more likely that we cannot transition: we will probably have to do with less energy.
Which in turn means that all the tech projects that try to replace "unlimited fossil fuels" by "unlimited clean energy" are just counting on the fact that somebody will solve the problem for them. What's the point of having electric sports cars if we don't have enough energy to use them?
AI is in the business of doing more with more. Chances are that we will have to do less with less.
We won't though because of the same trope Bill Gates et al regurgitate...this tech will solve all the problems, poor people need it...
What's a shame we need to rush everything, it's a shame couldn't have just done less with less for longer, built more efficient understandable systems and just enjoyed the ride a little more.
"Market forces won't naturally produce AI products and service that help the poorest," he [Gates] said. "The opposite is more likely.
"With reliable funding and the right policies, governments and philanthropy can ensure that AIs are used to reduce inequity.
"Just as the world needs its brightest people focused on its biggest problems, we will need to focus the world's best AIs on its biggest problems."
What we have, right now, is enough to turn many things in the world upside-down, and we only owe it to human and legal inertia that this is still not absolutely apparent yet, unless you're following what's happening.
Hindsight bias.