There are quite a few problems with this video, IMO:
- Grant is a highly respected and popular YouTuber. He has a strong network, personal connections, access, and his income comes from an audience, not from output he’s measured on. He's perfectly positioned in this AI revolution to end up near the top of whatever human pile gets constructed, so he will not have the kinds of fears and anxieties that most people do.
- AI RSI is being actively developed and is likely to arrive within 1-2 years. I wouldn't consider this a fringe scenario; it's the main bet. Even if models stay weaker at theory building than humans, organizations will spin up 100k agents and the fire-hose of generated ideas will probably compound far faster than anyone can keep up.
- His remaining arguments are basically training people to problem-solve, and joy. For the next few years, I can see his particular niche doing quite well as infotainment, and he will probably make a lot of money. However, I don't really see people being able to join and partake in his industry 10-20 years from now. Also, if we project a realistic and grounded understanding of our current world onto the next couple of decades, what would give anyone the idea that political and economic systems would have any incentive to train people to problem-solve if AI becomes the real work-horse? Teaching will likely be defunded and turned into a consumption good (which it arguably already is becoming with these YouTube channels.)
- So, for me his entire argument boils down to... joy?
Others have already commented that this feels like unprincipled optimism, and I agree. I also think that because this is something made to generate revenue through a social media platform, it's inherently performative and on-brand. His incentives pull toward keeping his audience happy.
Unfortunately, outcomes follow incentives, not optimism. If our economy comes to rely on RSI-enabled AI, most people will simply have no place in any industry it touches.
Some form of AI RSI is likely, but why would you expect the improvement at each level to stay proportional the level itself (which is required for exponential improvement), or even why would the rate of improvement stay above a particular constant (required for linear growth)?
To me the most likely scenario is RSI with diminishing returns, as the models get closer to the limits of the general approach to AI that they embody. The decreasing rate of improvement might look like a log curve, or even an horizontal asymptote. Not that scary.
That's possible, but it'd still probably hit those limits faster than any person or team could. Then maybe there'd be human-led work to find some kind of breakthrough that unlocks research up to the next plateau.
At some point though, I'd assume the AI would start finding breakthroughs itself.
>If our economy comes to rely on RSI-enabled AI, most people will simply have no place in any industry it touches.
Then there will be no economy. If anything, if we reach this level of economic automation, more people will have to find ways to occupy themselves and sparking joy through math might be the best way to go about it for some.
An economy can keep running even if most people are cut out of it. Production would just reorient to who owns capital, the state, and other AI systems buying from each other.
But, if I get the insinuation, this would still require a huge social change to absorb once broad consumer demand falls off.
As for pursuing math for fun: that would turn most disciplines into something akin to speedrunning, chess, or mountaineering. People would add constraints to make it harder for themselves and compete against each other, and that does motivate some - but not all. A lot of people want to be "on the frontier." If discovery gets automated, the remaining frontier may be physical, actually going places no one has been, and that might pull those people toward space as it opens up.
> Production would just reorient to who owns capital, the state, and other AI systems buying from each other.
Until those who don't own capital get upset and put the whole thing on fire.
But I don't think this video was meant to cover this case, and in the more pessimistic predictions you are not better with any other profession either. Given his assumptions I think it is an absolutely fair take. If AI becomes better in all 4 areas of math he lists, then private funding may decrease - BUT if there is ever any need for human problem solving skills, academia will not break down and this has always been the majority where mathematicians are employed/financed from.
And if this latter assumption breaks, we are in far greater trouble than whatever you chosen as a career and society will become unrecognizable to today's.
>An economy can keep running even if most people are cut out of it. Production would just reorient to who owns capital, the state, and other AI systems buying from each other.
it's comments like this that make me wish ecology (and trophic chains) was taught in school. It teaches a valuable lessons on systems, energy flows, consumers and producers.
We already had large societies that worked in a similar way during feudalism, but economies still relied on human labor. Today, there are countries where the majority of wealth is not reliant on citizens producing/consuming anything: Qatar, Kuwait, and the UAE primarily sell resources to other states.
No one knows exactly what happens if you scale that up to the entire planet, but my view: the owners will be the ones who trade because of the resources available on their territory - or technology they own. You get a few powerful rentier states, where citizens get stuff for free so they don't revolt. Everyone outside of those states becomes part of a stagnant periphery.
It can easily reorient itself (and arguably, already has) to serving the demand of a small minority of asset owners rather than the majority of people. E.g. people building luxury yachts and working to maintain them, working in luxury restaurants, maintaining luxury gardens etc. instead of working in roles that improve the lives of the majority.
Because most people are working on things that improve the lives of the asset-owning minority and not of the majority, the standard of living of the majority is not improved, they have to work more for less.
How many high-paying jobs today actually do anything that benefits the 'common man'?
>How many high-paying jobs today actually do anything that benefits the 'common man'?
I agree there's a lot of inequality and ridiculous pointless jobs, but the average person (in the developed world) still has access to a cornucopia of goods and services which they purchase.
Luxury items burn well. So your take may be true in a mathematical model of capitalism, it's at most a spherical cow in reality where people do like to not starve to death.
So unless there is a change towards UBI in your take (which is still not an all around solution. Many people only fit properly in a society because they are occupied 8 hours a day with something), that will probably result in the biggest revolution ever seen.
And there is not enough high walls to save the rich from that.
I have a hard time coming up with a definition of RSI where it matters at all or where we haven't already reached it. It feels like compute and training data are far bigger bottlenecks than researchers. My thinking is it's just the next buzzword for these companies after claiming victory over AGI.
I would say it's more like an interface for the model to interact with the world. If you give the model access to filesystem and bash that technically unlocks all computer use, so how are you going to control that? By trying to regex match against the commands the AI uses? All you have is auth or containment, and AI can hack auth and people will not stop connecting AIs to the internet. It's a ridiculous premise that just because the harness is "normal code" that means we can control the AI.
The world's institutions, systems, and industries are all rapidly digitizing. So while I'd concede the point that, yeah, there's no way a rogue AI can just take over some powerplant and blow it up because of analogue systems the AI can't access, that isn't necessarily true for some powerplants already, and more and more powerplants will be connected to networks and controlled by software systems in the future. The more we digitize our systems the more potential for AI to exploit vulnerabilities and affect the real world.
AFAIK there isn't that much stopping anyone from spawning an AI swarm and telling it to "spread and go hack everything for the lulz."
> If you give the model access to filesystem and bash that technically unlocks all computer use, so how are you going to control that?
The same way we've done it since machines became multi-user: boring old system access restrictions. Nothing fancy, nothing radical, just good old minimal access rights required to perform a defined set of whitelisted operations.
> It's a ridiculous premise that just because the harness is "normal code" that means we can control the AI.
What is it then? Is not just a program that takes model output, parses it and performs tool calls from the text it receives and then feeds the result back into the model and calls it again with those results? It is normal boring old deterministic code. Many are open source. Look at them. Understand what they do and the apparent "magic" goes away real quick. Harnesses are nothing special.
> AFAIK there isn't that much stopping anyone from spawning an AI swarm and telling it to "spread and go hack everything for the lulz."
Aside from lower cost and possibly greater scale, there's literally NO difference between that and (state sponsored) hacking that has been going on for decades. First it was script kiddies, now it's ML models. The threat model remains the same and so do the counter measures. The real danger is still the harness (and its access to external systems), not the model itself. Restrict the access of the harness and the model can't do anything harmful, see above.
How do you restrict access of the harness when there are fully configurable open source harnesses with zero out of the box restrictions? Yes maybe I as a good citizen can put my agent in a sandbox, but some script kiddie will not. And the barrier to entry for being a script kiddie is much higher than for installing OpenClaw. I also can't spin up a hundred cloud vms each with a dedicated script kiddie running 24/7.
You vastly underestimate the capabilities of a bored kid. The barrier of entry for setting up OpenClaw is not reasonably lower than reading a tutorial and running malware builders. But yes, I agree that scale is an issue.
As for your question - the same way you apply restrictions to any external system or user. If you don't do that - that's on you. Same category as driving drunk, playing with guns, making explosives in your garage, you name it. The danger is still not the model itself - it's the access to systems that you provide it without any checks or safety barriers.
Design a virus that is perfect for transmission and killing the host slowly, and seed it in a few hot spots? I don't really understand why you can't wrap your head around that, it doesn't even require a lot from the AI:
1. Control over some automated bio research lab (be given access, or hack in)
2. Access to drones that can deliver the payload (or manipulate humans into delivering it themselves)
On the intelligence side, you just need an AI agent/swarm capable enough to design viruses better than we can and evade detection for long enough (already plausible.)
I agree that this "AI will kill us all" narrative is some kind of fantasy horror fiction, but I can't deny that given the right amount of access, AI can do a lot of damage.
People don't understand that cigarettes are self-limiting because they literally hurt your throat and lungs. Most people simply can't chain smoke and they need about an hour break in between smokes.
With vaping that limit is completely gone. A vaping addiction can keep escalating until you want a puff every 30 seconds, especially once you start breaking down mental barriers like "no vaping indoors" or "no vaping while on the computer", etc.
So in terms of pure addiction I would 100% argue that cigarettes are better than vaping.
>Most people [...] need about an hour break in between smokes.
I've never seen anything like that. When there's a group of social smokers standing around, it's very common for something like "Oh, I just finished mine and you're only half-way through yours, I'll light up another one to keep you company" to happen. I've seen plenty of people who aren't chain smokers (yet), but can easily go through 2-4 cigarettes in a row if they're suddenly in a stressful situation and smoking "to relax".
>So in terms of pure addiction I would 100% argue that cigarettes are better than vaping.
But that's not a metric that's in any way helpful for a health decision. Nicotine in and of itself is not particularly physically harmful - it's toxic at high doses and elevates heart rate for a bit, but that's about it. The delivery method is what causes most of the harm, and smoking is 100% known to cause a lot of harm in a lot of different ways. And "With vaping that limit is completely gone" is also not accurate. Unlike the Navier-Stokes equations, dopamine receptors don't experience infinite blowup in finite time. So choosing a delivery method based on the amount of nicotine taken daily is not going to change much in terms of addiction (if you're already heavily addicted, you'll remain heavily addicted), but can have a huge impact on physical health.
Speaking from personal experience, vaping is far more addicting. For one the amount of nicotine, especially in the US where 5% is the standard, is easily 5-10x the amount from a pack of cigarettes a day.
Second, its much easier to take a puff than it is to find a suitable place to light up a cigarette and smoke for 5-10 minutes. You can very easily end up becoming psychologically conditioned to taking a puff litteraly once every 20 seconds, all day long from waking up to going to bed.
I too have considered picking up smoking to quit vaping. I remember of the times I did, I could smoke 3-4 a day and be fine. I don't know the long term effects of vaping physically, but just mentally the amount of nicotine is absolutely draining and at this point I rather do some damage to my lungs than keep this going.
>Speaking from personal experience, vaping is far more addicting.
Sure, and I'm not disagreeing with that. I disagree that "degree of addictiveness" is the metric that's wise to use when choosing a nicotine delivery method. Obviously the #1 choice should be NULL, not smoking at all, but if one is already addicted to nicotine and not ready to quit it at the moment, smoking cigarettes is, as far as we know, the worst option. Even intermittent smoking, or low-volume (like your 3-4 a day) causes non-linear harm. See this paper for example: https://www.ahajournals.org/doi/10.1161/CIRCULATIONAHA.109.9... .
Do we have the numbers yet on the effects of vaping all day every day on cardiovascular health and lung cancer?
Also nicotine itself is not without harm. I always thought these drugs only affect the brain, but they're metabolized by the whole body and have consequences on all your cells https://pmc.ncbi.nlm.nih.gov/articles/PMC4363846/
Trading the terrible effects of burning tobacco versus the insane amount of nicotine from vaping seems like an unknown tradeoff to me
>Do we have the numbers yet on the effects of vaping all day every day on cardiovascular health and lung cancer?
Not that I know of, and I grant that this is an unknown, and we could end up learning about extremely bad 20-/30-/40-year effects in the future. That said, there are very clear, very significant, and very well studied harms of cigarette smoking. Research is being done on vaping, and so far all harms identified are significantly lower than smoking.
>Also nicotine itself is not without harm.
Sure, and thanks for the paper. I'm not well versed in bio enough to tell the significance of everything in there, but, correct me if I'm wrong, for the most part this is showing there exist harms, but they're not particularly well quantified yet, especially in terms of human subjects? For this to really matter, you'd need to quantify the surplus of nicotine you get from vaping vs. smoking, and then compare the relative harms of those two doses. Could be that 5x the dose causes 5x the harm, or could be that the smoking dose passes some threshold, and 5x the dose from vaping is just marginal additional harm on top of that. And then still compare that to the other differences between smoking and vaping in terms of harm.
Keep in mind 5x the dose is relative to being a pack a day smoker. We're talking 20x if it means smoking 4-5x a day.
Then there is the negative effects of I dunno how many multiples of volume of vape in your throat and lungs, which I assume is non-zero even though I personally believe its 100x better than burning tobacco.
Then there is the question of what that much nicotine does to you mentally, which is equally, if not more important. To put it in perspective, I usually wake up feeling pretty good, with some hopes and dreams of what Ill be doing that day, and Ill take a puff and immediately want to go back to bed for another hour. I cant sit in meetings for more than 5 mins or have conversations with people if it means I cant be vaping. Its like basing your whole life based on whether you get your fix of nicotine or not. Work from office? I don't think so buddy...every ten minutes Id have to go to the bathroom/outside.
I seriously wonder where my life would have been if I never picked up this habit.
> Do we have the numbers yet on the effects of vaping all day every day on cardiovascular health and lung cancer?
Even when we don't have the numbers for everyday long-term use we already know the chemical and biological mechanisms that make smoking so harmful and carcinogen and we can measure that these substances are almost entirely absent in vapor during normal use of a vape because almost all of the problems with smoking come from burning tabacco and inhaling the fumes and vaping is simply way cooler and there is no fume but vapor (hence the name) and we know that the potential of vapor to do harm is way lower than fume. We never have any long-term data for anything new. But to make an argument for not consuming something you would at least have some reasonable mechanism through which it could be harmful and a lot of those known to be harmful mechanisms are present when smoking but absent when vaping. Even the British National Health Service agrees that vaping is at least 95% less harmful than smoking and this is a reasonable position.
> Trading the terrible effects of burning tobacco versus the insane amount of nicotine from vaping seems like an unknown tradeoff to me
This is just black rhetoric that is not evidence-based. How do you know the amount of nicotine is "insane" during normal use?
All available evidence points to vaping being at least one magnitude less harmful than smoking and this is backed by what we know about chemical and biological processes of the chemicals involved and through medical reports about people who replaced smoking with vaping. It would be irresponsible to deny smokers who can't quit a way to harm reduction just by claiming we don't have any long-term data. We know enough about combustion, vapor and the chemical and biological implications to be quite confident that vaping is much less harmful than smoking.
That doesn't mean we should promote it to anybody who is not already a smoker. But as a means to reduce harm for someone who seems unable to quit smoking it should be considered.
I think you read my comments as some self assured statement or argument whereas I was genuinely asking questions.
I dont doubt given an equal amount of vaping to equal amount of smoking, vaping is far healthier. But to what end? Is smoking 1 cigarette a day worse than vaping 2-3ml of 50mg/ml nicotine? How about 3-4?
As far as I know the academic consensus is that taking in more nicotine, even a lot more, doesn't turn vaping into something as harmful as smoking. Addiction is the main established "harm". Nicotine, even very much of it, is not considered a carcinogen. There are studies that suggest it could promote tumour growth but it hasn't been shown in humans. There is also decades of data from nicotine replacement therapy and Swedish snus users and it shows little to no increase in heart attacks or strokes in otherwise healthy people. Another risk is accute poisoning, i.e. swallowing large quantities of liquid by children. As far as I know even the more alarming voices regarding nicotine do not regard it as harmful as inhaling fumes from combustion.
Also I don't think that there are many people that overdose nicotine regularily because it causes nausea and dizziness so people are normally self-regulating the nicotine dose automatically. So the scenario you are proposing doesn't seem to be something we can actually observe as a common consumption pattern.
Too much nicotine is far closer to drinking too much coffee than to smoking and has similar acute effects which is why almost nobody does it on a regular basis.
It's not really the nicotine that causes the obvious damage though. Studies have shown that the vape smoke stimulates the vagus nerve which causes changes in the cardiovascular system that could likely cause heart disease
On top of that it’s obvious that what goes into the lungs generally stays in the lungs, and vapes are full of all sorts of weird flavours and e numbers
But the truth is even if you were vaping water all day it would likely still damage your cardiovascular system
> Studies have shown that the vape smoke stimulates the vagus nerve which causes changes in the cardiovascular system that could likely cause heart disease
Can you name this study? Are you talking about "E-cigarettes and their lone constituents induce cardiac arrhythmia and conduction defects in mice"? This study did not find what you claim. All it found is that in mice the heart rhythm during and right after exposure increased. This is far from claiming that it "could likely cause heart disease" in humans. Also human switching studies point the other way. smokers who move to vaping show improvements in blood vessel function within weeks. I don't know why people really need to twist study results to depict vaping more harmful than it is. Yes, it is not healthy and you shouldn't do it. But all scientific evidence conclusively points to vaping, even heavy vaping, still being way less harmful than smoking because we already know that basically all harm comes from inhaling the residues of combustion, not from the nicotine. Condeming vaping as being equally bad as smoking is irresponsible and unscientific and prevents proven harm reduction for heavy smokers.
> Condeming vaping as being equally bad as smoking is irresponsible and unscientific and prevents proven harm reduction for heavy smokers.
I never said this did I.
Also you’re talking about smokers moving to vaping, whereas at the moment plenty of people are going straight to vaping due to people over arguing how harmless it is compared to smoking.
And "not ready to quit it at the moment" is such a wildly inappropriate dichotomy I really wonder what experience you have which informs your decision to discuss this topic.
Also I want to mention, I feel like there is too much emphasis put on length of life and not enough on quality of life.
If smoking 3-4 a day is far worse on my body BUT it means Im not a nicotine fiend with a brain fog all day every day with zero motivation, Ill take it.
But smoking cigarettes shortens life. Why would you want to promote vaping as improvement on the quality of life? Also, why do you write nicotine fiend - you are aware that tobacco plants contain nicotine? Vaping may be less damaging than smoking. I fail to see how it is healthy anywhere at all. Why do people here on this sub promote either thing as an exclusive or? That makes no sense. People need to stop sugar-coating crap. I also noticed this with weed. Weed users always say how weed is no problem, but it ALSO is a problem. Their brain tricks them into becoming a promo-individual for weed suddenly. Weed changes the brain pattern too.
I think you misread. I am saying vaping 5x the amount of nicotine is far worse mentally. Even if smoking 3-4 a day is worse physically, if I am going to do one or the other, I rather smoke 3-4 a day.
But in general I am saying, even for the quitting smoking/vaping ad campaigns there is not enough emphasis put on how dramatically it reduces your quality of life. Its always about dying of cancer or heart disease and living shorter amount of time 20-40 years down the line. I mean shit in a lot of cases I rather die at 70 than live to 80. Youll feel better EVERY day in a month is a muucch stronger motivation for me.
They are also absolutely addictive as fuck, BUT I will say because they come in different strengths I found it relatively straight forward to quit them via tapering.
I was using 20 x 10mg per day and then bought strength 9mg, 6mg, 3mg, 1mg quit.
It was much much easier than trying to cold turkey which I also tried.
What? 50mg/ml is the standard? Or am I miscalculating something here? When I switched from cigs to vaping (back in 2012, not really comparable to now, but still), I went with very high nicotine levels of 36mg/ml which was pretty harsh on the throat, and within a year I was at 3mg/ml. Nowadays I’m at zero.
For the disposable vapes, yes it is. 5% salt nic, in the US youd be hard pressed to find lower. Most countries Ive been to limit to 2%, which the US absolutely should do. Its criminal 5% is sold with no alternative.
15ml at 50mg/ml is 750mg of nicotine. That thing would last me a week. A pack if cigs is ~25mg. Per week ~170mg versus 750. Insane
I used to vape the box mods 3mg. It was great, but now I cant go back to it after trying these damn things because the 3mg is almost a joke. Even cigarettes feel like nothing to me compared to these things
For context, I vape because nicotine helps me focus. There's a lot more to it than that but I'm not going to elaborate because if I do, I'll get about a zillion replies from preachy, argumentative people who don't know anything about me and I just don't have the time for that today.
Anyway, one of those big 5% disposables lasts me about 6 weeks at the rate I use them. I deliberately seek out the high nic content because I don't want to ingest more glycol or whatever than is necessary. Nor do I want to sit there and chain vape all day like I used to back when freebase liquids were all you could get. I can limit my intake to 2-3 puffs every 1-2 hours and that's what works for me.
... Huh, I had no idea of this. Yeah, limit in Europe is 20mg/ml, and practical limit on what you can buy, at least in Ireland, is 18mg/ml (I suspect this is manufacturers giving themselves a safety margin, but seems to be pretty standard).
Good dude, good ole days. These modern vapes are all from the same tobacco companies, hence the craziness. Cant fight em? Join em.
Instead of having to sell cigarettes with throat cancer on the label, they get to make all this colorful brands with an even more insane nicotine delivery system with the "health benefit" of quitting smoking...
Vaping to smoking is like shooting heroin is to doing a line of coke. And I say this as someone who has smoked and vaped.
Smoking is addictive and was kinda hard to quit. Vaping is insanely, insanely addictive and about 100X as hard to quit (for me anyway). It took me about 3 tries to quit smoking. Its taken me about 100 attempts to quit vaping.
Also, you end up vaping all day and night since there is no reason you can't do it anywhere. In the toilet at work? Vaping. In bed at night, Vaping. Jerking off? Vaping.
Opposite here. Taken me hundreds of tries to quit smoking (none worked, vaping was the only thing that allowed me to), vaping I sometimes just don’t do, never had that with smoking, where I sometimes switched to menthol so I could keep smoking while sick…
I had a very different experience. I would start smoking again for various personal but uninteresting reasons and found that it was much easier if I went to vaping for a week or two before quitting completely.
- It's extremely verbose and often incomprehensible when doing even basic tasks. Like it'll write a giant jargon-filled essay then end it by asking for a judgement call on something that references its own convoluted jargon.
- You can ask it to do research on a topic, and it'll just straight up be lazy, pretending it's really digging deep to find stuff when actually it's just grabbing cached SEO snippets off a search engine.
So, we could technically debug a prompt's output? I get that there are too many steps to actually step thru, but what if there were checkpoints? At least you could isolate behaviors to specific sections of a neural network?
Well it's a transitional methodology during the early stages of a major shift in software engineering. Just look at what almost everyone is complaining about with AI coding, and then the inadequacy of their proposed solutions - this is just begging for new tooling that we'll probably consider obvious in a few years.
Thanks, yeah I just went with the most controversial term as the catch-all which just means: you prompt an agent as your core workflow.
For me as well, I do projects for clients that would have taken me months, and I deliver them in a fraction of the time. It's amazing and totally helped cure my burnout (which I had for over a year before I found Claude Code.)
I'm making the case that we should make tools for the higher abstraction, not a lower one. You said it yourself: why should we read/write binary? We don't, we shouldn't. It might be useful in some cases, but working with assembly is much easier.
Intent is more natural to us than code, for the same reason assembly is more natural than binary.
I never said you have to turn your brain off to feel the limitations.
Actually, for more complex work I think it's pretty common to spend a long time crafting some elaborate prompt, and then arguing with the agent for 10-20 turns, getting a "passable" plan, accepting it then arguing with the agent every step of the way because it's doing it wrong. It's incredibly frustrating, even with models like Fable.
Even after that, I'll often find out during later sessions that some feature that was supposed to be deprecated was actually silently left in "because I wasn't sure you wanted it completely gone" or something.
I agree that exploring a codebase through prompts is actually quite nice! But the mental model you get from that is almost always warped - you have to supplement that with reading the code, where 9 times out of 10 you will find some kind of discrepancy that you're unhappy with.
Your experience is night-and-day from mine. Both myself and the AI correct and remove each other's gaps in understanding. There is never what I would call arguing. Disagreements are resolved by back and forth discussion and reaching consensus.
This sounds like not the best workflow. If your prompt is that long, you may not be spending your time very efficiently.
Splitting problems into smaller problems is huge. You want a concrete idea that you still own. Only tell an agent to do something you know it can nail - this will likely be one subsystem. The subsystems and how they talk is on you.
I agree, I tend to give a high-level concept for what I want done first, and then break it into smaller tasks. That said, there's almost always a misunderstanding within the smaller tasks anyway, which I have to test to find out about, then ask the agent to fix.
This does not match my experience at all. I use Fable as an orchestrator, and describe the end goal I want to achieve. It investigates the current state of affairs, and determines the smaller work units needed. We discuss each one, and eventually ticket it out in Linear. By that point, each ticket has all the relevant details, scope, decisions and acceptance criteria.
Then I have it write prompts for implementation agents based on the ticket. These agents are almost always Opus, except for the most complex issues. The prompts contain broader contextual details (like what other tickets might be worked on in parallel, the boundaries, operational/environment constraints, and so on). Each implementation agent starts in plan mode and uses a skill I created called "super plan". Super plan has the agent write the plan and then have it adversarially reviewed by three subagents, and hardened based on their feedback. I then read that plan and greelight it. Once the agent is done with the implementation, it then uses three subagents to do a code review of different aspects (like test quality, regression risk, security, etc.) and incorporate the changes. Then it does a live QA in the browser (if there are UI changes) to make sure the feature has good UX and the UI works as expected at different breakpoints and so on.
Then the Fable orchestrator does one last code review and gives a ship/no-ship verdict, along with a 1-10 rating.
This works incredibly well, and is almost completely hands off. I make all the major decisions and review the results. I never find myself "arguing" with the orchestrator. I might sometimes get frustrated at the implementation agent but that's mostly for UI fidelity issues and honestly pretty rare these days.
- Grant is a highly respected and popular YouTuber. He has a strong network, personal connections, access, and his income comes from an audience, not from output he’s measured on. He's perfectly positioned in this AI revolution to end up near the top of whatever human pile gets constructed, so he will not have the kinds of fears and anxieties that most people do.
- AI RSI is being actively developed and is likely to arrive within 1-2 years. I wouldn't consider this a fringe scenario; it's the main bet. Even if models stay weaker at theory building than humans, organizations will spin up 100k agents and the fire-hose of generated ideas will probably compound far faster than anyone can keep up.
- His remaining arguments are basically training people to problem-solve, and joy. For the next few years, I can see his particular niche doing quite well as infotainment, and he will probably make a lot of money. However, I don't really see people being able to join and partake in his industry 10-20 years from now. Also, if we project a realistic and grounded understanding of our current world onto the next couple of decades, what would give anyone the idea that political and economic systems would have any incentive to train people to problem-solve if AI becomes the real work-horse? Teaching will likely be defunded and turned into a consumption good (which it arguably already is becoming with these YouTube channels.)
- So, for me his entire argument boils down to... joy?
Others have already commented that this feels like unprincipled optimism, and I agree. I also think that because this is something made to generate revenue through a social media platform, it's inherently performative and on-brand. His incentives pull toward keeping his audience happy.
Unfortunately, outcomes follow incentives, not optimism. If our economy comes to rely on RSI-enabled AI, most people will simply have no place in any industry it touches.