> I've come to the conclusion that it wasn't coding on assembly level that made development in the 80s slow, but old vs modern tooling
Well, perhaps a bit of both, coupled with the fact that there were extremely restrictive hard limits to deal with - processor speed, amount of memory, as well very limited instructions sets such as the 6502 (although you got used to that).
As far as tooling, macro assemblers have been around for a long time. When I used to work for Acorn Computers in the early 80's, some people chose to use MASM, while others just used the BBC BASIC assembler. I'm not sure that it really made much difference. Fighting the limited hardware - speed and memory - was always the challenge.
Another factor that made development difficult/different was just the slow speed of iteration, especially if you were dogfooding as we were at Acorn and developing on the same impoverished hardware you were developing for. A combination of slow build times and a less expressive language (assembler) made for a very different feel to development - you were making more of a commitment to design choices rather than them being fluid and easy/fast to change. When TurboPascal came on the scene many people found it revolutionary just due to the speed.
I do think that language and libraries (which only becomes possible with capable enough hardware that efficiency isn't the main concern) makes a big difference to development though - you can just think at a higher more abstract level. How great it is in C++ to be able to use something like std::map with no concern for efficiency, when back in the day you'd be hand coding a custom hash table and sweating every cycle of the hash function.
The major reason for the never-ending disagreements on the nature of consciousness is that pretty much 100% of the time no-one ever rigorously defines what (things) they are actually talking about, and the word is so overloaded and poorly defined that any discussion therefore devolves into people talking about different things, as well as the discussion being so vague as to be meaningless.
There are of course other reasons too, with things like religious beliefs and human ego meaning that people come to the discussion with a major bias and fixed views rather than even being open to any rational discussion.
Finally, everyone is conscious and has an opinion, but only a tiny fraction are actually knowledgeable about the brain and have spent any large amount of time thinking about things like evolution and brain development .. they have an opinion, but are just not qualified to discuss it!
If you break down all the different things that people are referring to when they talk about "consciousness", and define them individually with as little wiggle room as the english language and underlying taxonomy of concepts allows, then I really don't think there is much mystery about consciousness at all, but of course those with an agenda who want there to be a mystery will still argue about every part of it including the definitions that remove all the wiggle room.
The nature of consciousness has long been a contentious subject, and one of interest, but it seems that the rise of AI has intensified the discussion with the new question being whether AI is or could be conscious. I do think this can be answered in a principled way (=yes), but in the end you can only PROVE that something, or someone else, is conscious if you accept a functional/testable definition of it in the first place.
One can build scientific theories without rigorously defining terms: a stipulative definition is enough.
Best example is Darwin's "Origin of Species"; here, Darwin didn't rigorously define "species": 'No one definition has as yet satisfied all naturalists; yet every naturalist knows vaguely what he means when he speaks of a species.'
Many in the social sciences fetishize definitions, operating under the false notion that formulating a precise definition is the primary goal of inquiry. In reality, a robust scientific theory is a structured set of hypotheses; when combined with auxiliary theories, it derives a specific set of testable consequences.
Even within this framework, one must remain vigilant against ad hoc explanations. An ad hoc explanation fails to provide genuine systemic insight because it is engineered solely to fit the target phenomenon; it eliminates the explanatory gap by simply re-stating or absorbing the explanandum without offering any independent predictive or falsifiable power.
But with "consciousness" we don't have even a stipulative definition, more like 40 different competing definitions (https://philpapers.org/rec/VIMMAT) which are in many cases only loosely related, and generally are talking about very different things. When the situation is this bad, I am skeptical you can make scientific progress.
Stipulative definition is all about fixing the reference. Or if we want to use the talk of meaning, it is abou the extension/denotation. If 100 people fix the reference of the word 'consciousness' to 100 different things, it just means that these people are engaging in philosophical discussions without having any object level theories (theories from neuroscience, etc).
If 10 different scientists who work at the object level, using the same word but with a different reference, then we have 10 different theories about different phenomena. This is what we need today: more research, not philosophical discussions.
I'd say that comes down to a matter of degree, and as a quibble it's certainly not a scientific theory if it's not worded sufficiently precisely to be testable and falsifiable. With Darwin it was really about where did all the variety of animals come from, and with speciation/variation being a process rather than an event, it really doesn't make much difference how one defines "species" (can't interbreed is a useful starting point). Of course Darwin was also just speculating about some mechanism of hereditability existing, so at the time this was more of a thought experiment than theory.
The trouble with discussion of "consciousness" is the sheer degree of ill-definedness - it is such a hand wavy and multi-facted concept, that it's not possible to even begin any meaningful discussion without defining a better vocabulary and breaking the concept down into pieces. Are you talking about subjective experience, mental awareness, free-will, altered state of consciousness, or what, or more likely simultaneously some mish-mash of all of the above and more!
> The major reason for the never-ending disagreements on the nature of consciousness is that pretty much 100% of the time no-one ever rigorously defines what (things) they are actually talking about, and the word is so overloaded and poorly defined that any discussion therefore devolves into people talking about different things, as well as the discussion being so vague as to be meaningless.
You're right, but that rigorous definition is a significant part of the problem. We have a very difficult time rigorously defining and then debating certain attributes about consciousness or related concepts precisely because the definition and exploration of the definition is what is being debated.
This makes it a very fascinating topic.
For my own pet theory I think consciousness as we like to understand it is an emergent and evolutionary social construct for cooperation amongst humans, and different people may have different levels of conscious thoughts, similar to how mammals are conscious in a different way amongst other species. It's a spectrum. There are, in fact, philosophical zombies.
> with things like religious beliefs and human ego meaning that people come to the discussion with a major bias and fixed views rather than even being open to any rational discussion.
You're forgetting that attempting to have a "rational" discussion is itself a bias inherited from the many centuries of intellectual development that occurred between the middle ages and now - the parts that the article conveniently skips over entirely.
The "debate" here doesn't function to generate an answer, but to narrow down the scope of the question into the very constrained domain. When ppl debate "consciousness" they are re-affirming their opinion that humans are inherently rational agents (hence "scire" -> "to know"), rather than agents that can live, feel, think and will, which would require a different term, like "soul".
> rather than agents that can live, feel, think and will, which would require a different term, like "soul"
You're just substituting one ill-defined, and overloaded, word ("consciousness") with a bunch of others ("live", "feel", etc), and asserting that to you they mean something different.
It's impossible to have a discussion on this basis, and I'm sure to many people "soul" is exactly what they mean by "consciousness", or at least part of it. It's no less reasonable that an AI has a soul as that it is conscious - it depends on exactly what you define those words to mean.
People have made due with conceptual fuzziness, I think it's disingenuous to suggest that discussion is impossible without absolute conceptual clarity. All you are saying is that using these terms does not allow you to have a specific kind of discussion - which, a lot of the time, is one that reduces humans to mathematical objects that perform computations.
Yet, if that is your goal and the definition of "soul" or "consciousness" are entirely arbitrary decisions that you don't care about - then it's worth remembering the adage "you may not care about politics, but politics cares about you".
Yes, but as noted elsewhere in this thread it's a matter of degree. Consciousness is such an ill-defined and overloaded word that it's hard to say it really means anything - it's more of an all-encompassing term for a bunch of largely subjective phenomena.
I don't disagree - for a long while I believed that the study of "consciousness" was the result of a linguistic mixup and best dealt w/ via Wittgenstein.
However, if you move past surface appearances, you can think about what kind of cultural "work" is being done spending effort on this mixup. What's happening is that, in the noosphere, the cloud of concepts around "consciousness" is battling it out with the cloud of concepts around "soul" over which cloud of concepts best describes what it means to be human, with big implications on what it means to be good person. Now we're dealing with a mix of Aquinas and Latour.
I'm not a professional neuroscientist for sure, but I've certainly spent decades thinking about things like intelligence and evolution, perception, qualia, have read dozens of papers on subjects like cortical microarchitecure, etc, and would like to think I'm well enough informed to be able to discuss things like intelligence and consciousness.
I don’t really care about whether you are “qualified” by your criteria, because I don’t hold your view that there is some specific level of education that allows one to hold and express an opinion. My point was more to point out some potential hypocrisy.
No one is obligated to humor or debate amateurs because no one is obligated to humor or debate anyone[1]. But being an amateur does not mean that the person couldn’t hold a valid or meaningful viewpoint. Dismissing someone’s opinion specifically because they are an amateur is just a special case of appeal to authority. If their opinions are fundamentally flawed, then those can be addressed head on without resorting to insisting that they don’t have the right qualifications to participate in the discussion.
[1] The “faster than light” neutrinos come to mind, too. Scientists involved in the experiments explicitly said they didn’t have time to entertain amateur theories, a valid statement.
Being an amateur has no bearing on how knowledgeable someone is about a subject, but for an opinion to be useful (or just interesting) it needs to be backed up with some grounded rationale, otherwise it is just that - an opinion.
Trouble is you don’t how much time anyone else has spent studying. How can you tell who else has “spent decades thinking about things like intelligence and evolution, perception, qualia, have read dozens of papers on subjects like cortical microarchitecure”? You can’t but you still dismiss others as unqualified.
> it needs to be backed up with some grounded rationale
This was my point. Address the claim and not the qualifications.
There is no gatekeeper to internet discussions, so nobody is stopping you from joining in.
If you feel qualified to add something useful to the discussion, or just to throw your unqualified opinion in for that matter (whichever the case may be), then go ahead. I'm certainly not stopping you.
As may be expected the conversation is already a garbage fire.
It's a strangely/poorly written article. I've read it a couple of times and am still not 100% clear what the author is trying to say.
As best I can tell, the author is saying that it's unrealistic to expect that a vibecoded photoshop would YET exist since just because you can use AI to help doesn't make the task much easier or quicker. If this is the right take, then I guess he's really talking about AI-assisted development (i.e. AI coding used as a tool by a human developer), rather than "vibe coding" in the sense of "here's some specs - write this for me". With AI as a coding tool, then all the hard work is still left to the human - coding it up once you've specced and designed it was never the hard part.
With Karpathy-style vibe coding - just tell the AI what you want it to build - it's either going to succeed fast or fail fast, so "where is the vibecoded Photoshop" is then a reasonable question, albeit a rhetorical one, reflecting that this type of "gimme X" vibecoding isn't able to produce something of that nature, so of course if doesn't exist.
Yeah it has a confusing clickbait title that gives the wrong impression of what the article is about. His point is that the bottleneck to making a complex app like photoshop is architectural rather than just writing basic code… and he argues the LLMs don’t magically make the architectural part easier.
I totally agree - the phone as a form factor is not going away. People are always going to want to have a mobile communicator/computer, and want one with a screen and all-day battery life. The phone is not going to be replaced by smart glasses or some other wearable or screen-less pocket device.
It may well be that the user interface of your "phone", and how you use it, changes over time as we progress toward AGI, but as long as Apple keep to the Job's aesthetic of making well designed products that get out of the way and just "do the thing", they should be fine. Of course Apple will eventually fall, as all companies do, but I don't think the reason for it will be that the "phone" market was rendered obsolete by AI.
Perhaps if phones becomes more of a "pocket assistant" than a device to run discrete apps, then they will becomes harder to differentiate based on software, and more of a generic item rather than a status/luxury one ... who knows? Anyone else have any theories of how Apple may eventually fall?
There is one potential AI risk to Apple, that they are at a disadvantage due to not having their own frontier models and datacenters to run them on, but I think there will always be someone willing to sell them API access, and they will adapt as needed. Good enough AI is only going to get cheaper to train and serve, and Apple not trying to compete in this area may well turn out to have been a great decision, just as Microsoft seem to be doing fine letting OpenAI take all the risk.
That's really the point of the article. As long as the phone is the (or at least a significant) conduit for our use of AI technology, Apple is in a good spot, and it's the same spot where they have historically done very well.
I think the vision of pocket assistant versus discrete apps is very much Apple. Remember the original iPhone had no app store. The app store is kind of a pain to deal with. If I had to bet, this starts with Apple pivoting Swift Playground into Playground releasing it across all devices. The programming language becomes invisible. The live canvas is the document.
>want to have a mobile communicator/computer, and want one with a screen and all-day battery life.
Well before the iPhone flew off the shelf, using the the previously established smartphones I never had to settle for less than a week of battery life.
Plus anybody could just slap in another spare battery whenever they wanted to, whether they were off the grid for an extended period or not.
Never thought it was going to end, only get better not worse.
Perhaps, but it still amazes me that something like an iPhone with a GHz processor blasting pixels to the screen all day, can run on a wafer thin battery (Mr. Creosote reference for the cognoscenti) for a full day.
Given that humans sleep at night, recharging the phone at night is a reasonable price to pay for the benefit of a smartphone vs flip phone, but a device that needed charging during the day as well (e.g. due to a form factor with a tiny battery) would almost certainly be a product killer.
I guess the good news may be that if/when there is a major pricing correction, that many of the people using free or $20/mo subscriptions to generate social media commentary may balk at the real cost and go back to writing it themselves.
Something I have noticed is that the people who are using it to write everything are the same people who had a poor level of English writing a year or two ago.
I've never had a problem with direct translation... but the 3 paragraph choppy structure with subheadings full of AI-isms is not ESL users using it faithfully
Would make sense ... writing is a skill, and one that I think most people are proud of if they are good at it.
Maybe it's different if you are doing technical/commercial writing, but for social media where you are writing for fun, and to express yourself, it'd be odd to let AI be your voice unless you realize your own writing is very poor.
> for social media where you are writing for fun, and to express yourself, it'd be odd to let AI be your voice unless you realize your own writing is very poor.
A lot of people post for clout, so something that can skip the difficult process of becoming a good writer (and original thinker) is more than enough. They can churn out think pieces about any topic at an unlimited pace, basically.
It doesn’t add much to the world, but they get a lot of traction (which I cannot understand, given the quality of content.) And that’s what matters to them.
I think if you gave most people the choice between (a) being a thoughtful and original writer (b) being seen as a thoughtful and original writer, the vast majority choose (b). Especially when it is zero effort.
I noticed this from former coworkers who I know couldn't write beyond first grader level a few years ago. They weren't good at their native language either.
Now they write "competent" blog posts on LinkedIn that seem 100% AI slop. Some are employed at AWS, too.
I'm not a native English speaker as I'm sure my writing shows. My point is that I'd rather read genuine posts full of grammar errors instead of slop.
I can't tell from your post that English is not your native language, outside of the Americanisms (I assumed that American English was your native language) :-)
I think there will always be a free tier that they'll be willing to use. Even if it sounds hackneyed, those folks will still use it because many people are not discerning readers anyway.
Despite what I just said, I do hope so, because I'm really not inclined to pay for it, at least not very much. I don't need another $100-200/mo bill in my life, and it doesn't provide that level of value as a chatbot. Google is enough.
I'm not sure that free tier will necessarily continue forever though, unless there is a way to monetize it (presumably by advertising, or by selling data they've gleaned about the user), or perhaps if there is no privacy and the provider is treating you as a source of free data. Right now we're still in the market-share grabbing "never mind the profits, count the users" stage.
A free tier will almost always exist. Mostly for the reasons you already describe. That's a training ground for their small models as well as a way to get full access to new training data (and advertisements). As well as funnel new paying users. Why would you ever give that up?
The paper is about a way to do SFT will less chance of catastrophic forgetting and performance regressions.
The idea is that SFT on new data that was NOT generated by the model (aka "off policy" data) is likely to cause problems due to the statistical mismatch between the new data and what the model has already learnt. As I understand it, their solution is to statistically align the new data with the old by feeding it to the old model, which will hopefully grok it via in-context learning, then have it regenerate it in its own words such that "off policy" data now becomes "on policy". The model can then be SFT trained on this regenerated data (i.e self-distillation).
To me SFT and "continual learning" are two distinct things.
Human/animal continual learning is always-on learning that removes the need for, and distinction between, training and inference, and it initiated by prediction failure. It's as much about skill acquisition as it is about knowledge acquisition. Continual learning can happen in any context from trying to do something (or just observing something passively) and being wrong about the outcome of your own actions, or what some external entity does next, to curiosity/boredom driven exploration and play which is more along the spectrum of pure learning with less expectation of outcomes.
Continual learning is what, one day, will let the AGI intern pick up new skills on the job by trying to do things and failing/learning/practicing until they get better. This is not the same as sending the intern home with a textbook to read, or a transcript of the conversations you had with it today, and having it take these onboard overnight, which is basically what SFT is designed to do - intermittent addition of new declarative data.
Yeah, and they are not self-distilling in the sense of training on the outputs of the model, but instead operating on the pre-output probability distribution - trying to minimize teacher-student differences.
Still, at the end of the day this is a type of SFT, not continual learning in the human/animal sense.
> I don't think it's super clear what we'll find out
I think some companies will find out that their senior engineers were providing more value and software stability than they gave them credit for!
Corporate feedback loops are very slow though, partly because management don't like to admit mistakes, and partly because of false success reporting up the chain. I'd not be surprised if it takes 5 years or more before there is any recognition of harm being done by AI, and quiet reversion to practices that worked better.
No, the behavior is undefined. That means, quoting the ISO C standard, "behavior, upon use of a nonportable or erroneous program construct or of erroneous data, for which this document imposes no requirements".
A conforming implementation could reject it at compile time, or generate code that traps, or generate code that set a to 137, or, in principle, generate code that reformats your hard drive. Some of these behaviors are unlikely, but none are forbidden by the language standard.
I'm not sure where precisely this sequencing exception to the default "eval order undefined" rule is given, but after the 24(!) sequencing rules they do give this "++i + i++" as an explicit example of undefined behavior.
Interestingly that page says that since C++17 f(++i, ++i) is "unspecified" rather than "undefined", whatever that means, and presumably plus(++i, i++) would be too, which seems a bit inconsistent.
A lot of crops need nitrogen. What has been impacted by Trump's Iran war is the supply of Urea through the Straight of Hormuz.
If the closure persists then no doubt other sources can ramp up to fill the void, but it's going to be too late for this season. Some Asian farmers have already chosen not to bother planting rice crops since the increase in fertilizer (urea) cost has meant they'd be losing money.
Fuel prices are also impacting imported produce prices.
Well, perhaps a bit of both, coupled with the fact that there were extremely restrictive hard limits to deal with - processor speed, amount of memory, as well very limited instructions sets such as the 6502 (although you got used to that).
As far as tooling, macro assemblers have been around for a long time. When I used to work for Acorn Computers in the early 80's, some people chose to use MASM, while others just used the BBC BASIC assembler. I'm not sure that it really made much difference. Fighting the limited hardware - speed and memory - was always the challenge.
Another factor that made development difficult/different was just the slow speed of iteration, especially if you were dogfooding as we were at Acorn and developing on the same impoverished hardware you were developing for. A combination of slow build times and a less expressive language (assembler) made for a very different feel to development - you were making more of a commitment to design choices rather than them being fluid and easy/fast to change. When TurboPascal came on the scene many people found it revolutionary just due to the speed.
I do think that language and libraries (which only becomes possible with capable enough hardware that efficiency isn't the main concern) makes a big difference to development though - you can just think at a higher more abstract level. How great it is in C++ to be able to use something like std::map with no concern for efficiency, when back in the day you'd be hand coding a custom hash table and sweating every cycle of the hash function.
reply