The case against AI everything, everywhere, all at once
time.com
time.com
It's about an ancient civilization so advanced they have tools and computers that can create art just from their intent. But they have grown so distant from the inner workings of these tools that they are incapable of modifying or repairing them and their survival is put into jeopardy because of it.
Survival risk, maybe not, but we are certain to see massive die-off and a decades-long cold start given a perturbation maybe an order larger than the COVID pandemic, when the system is already undergoing unprecedented stress from climate change, as appears likely in the coming century...
Ironically, later shows dialed down on the smartness of Starfleet computers - from what I heard, TNG level voice interface seemed too advanced to be believable. But it turns out, the unbelievable became reality sooner than anyone expected.
Our underlying data should still be extremely principled and structured. But I think LLMs enable significantly better human computer interaction, where humans aren't interacting directly with the data in your preferred manner.
For example, I've been parsing sentiment across the internet from the app I run, and using it to crunch it down into a very structured internal tracking database I have. I want to canvass as many places as possible. Using an LLM for that transformation has been stellar, and I'd casually say I'm at around 90% accuracy, which is more than sufficient for my planning & justification needs.
In that sense, it should just be an additional tool in a toolbox, not a platform to bet the house on.
It doesn’t explain why a randomized black box is needed. It only shows that we need better tools (i.e. not randomized and open).
Having LLMs use programming languages/tools/libraries built for humans is the wrong path to fully leverage their potential. They are capable of so much, but trying to use them as a drop-in replacement for humans isn't the best use for them. Frankly, it's impressive it works at all, but I think we will find hard limits on the actual complexity and reliability of the systems they can build in this manner.
We still need intelligent, rigorous system design. LLMs have the strong potential to let system builders operate at a higher level of abstraction, but in my opinion having LLMs write Python or SQL is an evolutionary dead end.
My own work is directly focused on this and what IS the proper place of a LLM in a data processing or knowledge system.
It feels like the claim is that this was just developed by industry giants to mess with people.
> Once again, a handful of competitive but ideologically aligned leaders are telling us that large-scale, general-purpose AI implementations are the only way forward.
The only way forward to what? What we have is just the continued evolution of hardware and software. Hardware keeps getting more powerful and software continues to get more powerful to match it. We are doing with that hardware exactly what we want to do: make things easy that were once difficult and make things possible that were once impossible.
The main thrust of the article is that there will be both positive and negative consequences. There will be a big push by many players to extol the positives while suppressing information on the negatives to profit as much as they can. Then they can stop suppressing the negative consequences as they make their exit and leave everyone else to deal with them. This is such a standard playbook for business at this point it should not be a revelation to anyone.
Remember, you don't have to wait until Google develops a customer service AI that arbitrarily locks you out of your digital platforms because of a hallucination. You can in fact push back and request baseline levels of accountability!
> Remember, you don't have to wait until Google develops a customer service AI that arbitrarily locks you out of your digital platforms because of a hallucination. You can in fact push back and request baseline levels of accountability!
AI changes nothing here there. People already get locked out of their accounts because of a bunch of If-statements. If anything I think AI is merely exacerbating issues that already exist and would exist without it. Focusing on AI as the problem is missing the forest for the trees. Google being unaccountable to people relying on their service has nothing to do with AI. Misinformation and propaganda being easier and cheaper to spread than legitimate information is not just an AI problem. Perhaps many of the problems caused by people, if-statements, and AI can also be solved with AI. But for that we need to focus on solving those problems explicitly and not on the technologies involved.
And remember kids, be sure to remember what you learned in class!
I wasn't expecting to be a Luddite at 36, but I am. I not only agree with the author, but, more generally, also really dislike the way that the future is developing.
I have zero interest in today's version of social media. The version that keeps people glued to their phone screens at all times, mindlessly viewing or scrolling through near-zero-substance content, giving constant hits of dopamine to break down self-control.
I have very conflicting feelings about _kids_ getting sucked into this, I.e. a parent handing their child a tablet at a restaurant to stop the crying.
I absolutely hate how abused phone notifications are and have most turned off. I'm actually uninstalling more apps in favor of their websites (while that is still possible), partly because of this, and partly because of the tracking these apps do that you can't easily turn off.
Voice assistants are buggy as hell and are annoying to use more often than not. I turned off Siri on everything except my Apple Watch.
Lastly, I have absolutely no use for LLMs.
I really like being able to write code or learning how to write boilerplate in other languages. I like understanding everything that's going on and what incremental progress looks like.
I like searching for answers on search engines and reading through answers or (sometimes) documentation to find what I'm looking for.
It is blisteringly obvious that the goal of big LLM is to (a) get users to think less, criticize less (because why criticize if ChatGPT tells you the "right" answer every time), and consume more, and (b) to centralize wealth in companies that are big enough to scour the Internet, train their own LLMs on that data and generate answers with minimal latency.
We should cut out the middle man and start building a map of all likely programming symbols a stream of integers/bytes can create.
LLMs can hide data and source code behind weighted nodes. The easiest way to cut through that is map out the binary stream directly.
Sounds insanely complex, but it isn't really. It's safe to say the cast majority of code is boiler plate running in popular coding languages, with similar symbolic 'shapes'.
We should probably cut the head from the current abstraction stack (isa>assembly>C>C++) and start-over, interpreting blobs of binary straight up.
In the old days I had an automation to grab things, clean up and push to my kindle to read texts. This was at the time of RSS feeds and just after Google had killed Google Reader and I already despised ads.
I am using an iPhone and it's just dreadful to use the internet on this thing due to the ridiculous amount of ads.
I'm not seeing any as I read this on iPhone. If you want a similar experience there are many good solutions, but my "quiet web" extensions are 1Blocker, Hush, Rekt, and Vinegar.
https://apps.apple.com/us/app/adguard-adblock-privacy/id1047...
Maybe something like this, for you?
Or install AdGuard as a local VPN on your iPhone and you're set or set your DNS to an ad-blocking DNS, even Chrome for iOS will get ads blocked. This is a you problem, stop whining.
One other option is to use https://github.com/danburzo/percollate to convert a webpage to a nice document directly. I use both tools depending on my needs.
Is the lack of such posts due to newfound restraint? Or is this kind of task already too complex for an LLM? Could one manage to work from raw page sources and figure out what text to summarize to the kindle reader? Or would there need to be other browser-simulating layers to first do all the page rendering and scraping before handing off text to the LLM for further cleanup?
The argument to made like this is already everywhere understood and almost nowhere addressed: that specialization and localization and globalization reliant on various preconditions, has created an incomprehensible web of dependencies which regularly, belatedly, turn out to be points of failure which cause cascading disequilibrium.
Maintaining a tribal-level small community, and some loose federation across a set of those communities? 20K years of adaptable stability.
Empire-level governance on the scale of contemporary empires (the US and China specifically) is proving almost impossible to maintain; the stressors from contemporary technologies regularly outmatch their benefits wrt operational stability. We can't solve federation for a Twitter clone; why do we think we can solve it at the nation state level? Every indication is that this is proving the US was not capable of scaling 10x or beyond without a refactor.
Modernity rests on an evolved rather than designed chaos of systems few of which are load-bearing within their originally conceived limits, and those which were so conceived are themselves single points of failure (again: Twitter. But also? Amazon, SpaceX, Walmart, UPS...)