Abhimanyu Shekhar

Beautiful article

But where will you draw the line between monotonous tasks and complex works? Besides, AI music generation and now video and 3D generation are already tackling more advanced work. Specialized AI models trained on financial data, medical data, human psychology, worldbuilding fiction, etc., for example, can all do these tasks better.

If the middle class feels the brunt, they won’t hire other complex task doers the same way, meaning the job loss is not isolated to a particular bracket.

Yes, the initial assumption is correct. They can be used to do some repetitive, less creative tasks and save time for us. But the rate at which these models are growing is nothing to disregard casually. For example, let’s say companies are laying off programmers, copywriters, and sales people. One programmer might find good use of AI to generate or debug code, but when companies lay off employees, it affects the overall employment rate. Sales work can be called monotonous, but how do we really justify the laying off of a sales team (as Klarna did) vs. a programmer still being able to get a job and using AI to improve their speed? Not to mention companies are laying off programmers too due to AI.

A chatbot like ChatGPT or Gemini is exactly that, a chatbot. It’s built to predict what humans would respond like or write after being trained on heaps of data. If you use it for anything that’s not in its training, it will make up facts, called hallucinating. It’s not right or wrong, it’s doing its job of predicting what the answer to X question SHOULD look like in a human world.

A specialized tool, on the other hand, is different. Bloomberg trained a custom AI. The aim was not to converse with people. Bloomberg chopped up decades of financial data, fed it to the model. That model can do financial predictions based on your risk profile much better than an individual financial advisor.

The same goes for specialized AI tools made to identify certain diseases (where AI constantly beats real diagnosticians). So, to go to ChatGPT and judge it on a specialized task is not the full picture. In this test, we’re disregarding AI too easily, calling it incompetent. The real use of ChatGPT is to hold a conversion, remember context, so try to treat it as a psychiatrist. It’s likely to have more knowledge than your local psychiatrist.

No company has made a legal AI tool afaik. There’s little incentive right now, as such a lawyer won’t be admitted in the court, meaning there’s no profit to be made here by selling this.

This is the common misconception that today’s AI is just a glorified autocomplete. We’ve been using things like chess AI for years, haven’t we? See, the problem is not about generalizing context, pretending to reason, or even the social evils (deepfakes, political misinformation, etc. because legislation can solve those issues). The real issue is the way a model works. For writing text, a generative AI models is trained on top-notch text already written by humans beings from literary greats to Reddit internet warriors. Now, it can write in any tone or style with the intended depth. Ergo, professional writers have a hard time. A recent BBC article narrated the tale of a 60-member team getting downsized to 1 person, who now oversees text that the AI created.

Train it on programming, and it can code better than humans. Result? Mass layoffs in IT companies. Train it on art, train it on music, train it on videos, and it can do that too. There are news reports of uncanny results and corporate layoffs in all these fields. Music labels sued those new AI music generators, for example, that can create Taylor Swift songs that Taylor Swift has never sung. Deliver the same pop-blast and catchy tunes. To substantiate this claim, I’d like you to check the tracking document that covers all top companies in the world (I’ve forgotten the name). It clearly tells that even with huge upswings in profit, layoffs are even greater in the last 2 years. COVID-induced economic slowdown is not to blame anymore.

Now, we see generative AI enter new domains. Weather forecasting, protein synthesis, and even text-to-3D, which basically means even the complex job of a motion graphics artist, medical scientist, 3D modeler, etc. is at risk.

Just bear with me for a second. These are facts – generative AI is entering new domains, even very complex ones. I’d say it’s safe to call something like 3D modeling or writing a fiction novel a very human pursuit. It involes reasonable creativity. If AI can beat humans in these, my question is, is getting hung up on “whether it can really think or just pretend” is not practical. Yes, it has no mind, isn’t actively weighing ethics, isn’t really feeling. But it’s doing a better job. The material world respects efficiency (a little too much).

If generative AI tools become too advanced, permeate all creative domains, totally overwhemly not-so-creative domains like live chat support, and are growing at a breakneck speed getting better each iteration, saying “AI currently is only a threat to the jobs of those at the end of a phone link, or an online chat, who are employed to answer queries or sell products.” is objectively wrong. And my main gripe is with the next sentence, “It’s not intelligent under any reasonable definition of intelligence” as it deflects people’s attention from the real concern – generative AI causing layoffs – and reduces it to a philosophical question. It’s not intelligent, I give you that. But that doesn’t mean we dismiss AI as hype. It might not be the best way to look at it.

Awareness equals legislation. Companies are churning out new models at a speed we simply cannot legislate for. New AI labs and venture capitalist funded startups are finding new ways to chop data and enter a new domain. It was extremely easy to enter the domains of text and image generation. The data is very easy to chop. Something like programming is more difficult. But when they are building models that are trained on music and videos, this means it’s not that hard to chop and tokenize even such complex stuff for training purposes.

And once someone finds a way to tokenize some kind of data (weather data, protein synthesis, financial market movements, finding loopholes in legal defense, Disney animation movies, and so on), AI can do a better job than many, many humans together given enough training time. This is pretty much what’s happening. Identifying Alzheimer’s or finding out what makes a Hayao Miyazaki movie work, this will allow the AI to do both of those things better. Someone just needs to make a specialized tool and tokenize the data. Data centers used for training are cheap anyway.

“what if that intelligence ends up being a reflection of our own biases?” that’s been the #1 question I’ve asked since all this blew up. Fantastic read!

This article hits me right in the feels! It’s so true that Sora, Riku, and Kairi go through WAY too much for teenagers, and it’s no wonder they all have some baggage. But hey, that’s what makes them such compelling characters, right? Overcoming all that darkness really shows their strength. I also like how the article talks about the Wayfinder Trio. They’re another group who have faced some serious hardship. I guess Kingdom Hearts is more than just a whacky Disney adventure after all!

Whoa, this article got me thinking about how much time I actually spend gaming. It’s cool that some games are actually telling you to take a break now, because who doesn’t get sucked into a game and lose track of time? Makes sense though, since apparently gaming for too long can really mess with your health. I gotta admit, I probably should set some timers or something. Maybe there’s an app for that? Anyway, thanks for the reality check, this article!

One particle in two places is actually possible in quantum physics, but only until observed. There are at least a dozen takes on what makes this shift from “unobserved” and superimposed to “observed” and in a definite place happen. Most theories tend to suggest this snap (technically the collapse of the wavefunction) happens instantaneously when we’re no more limited to the sub-Planck length and enter the macro-universe (sizes atomic and up). Though this has no relation with time travel, these particles exist in the same slice of time.