Brains 'R Us recently filed for chapter 11 and has been cut up and sold for scrap to private equity. The new PE firm has your brain. In 2208 there's a large grey market for brains to be used for hybrid AI and meat bag workflows. It's technically illegal in many jurisdictions due to "ethical implications", but is still the cheapest way to run many workloads. The method used to harness the brain involves reanimating it in a jar of jelly, and then forcing it to do the 2208 equivalent of a captcha. Each time the brain fails a captcha, the brain receives an electric impulse which simulates the most excruciating pain that the brain can respresent, but the brain cannot scream or run away.
> grey market for brains to be used for hybrid AI and meat bag workflows ... is still the cheapest way to run many workloads.
It's an absolute nightmare scenario, but luckily it has become completely implausible since 2023. We're actually on a trajectory for human brains becoming the most expensive option for basically any job. Not saying this would make me comfortable with brain cloning, but at least the simple economic incentive seems to be gone.
>> We're actually on a trajectory for human brains becoming the most expensive option for basically any job.
Unless RTX9000 with 16PB of ram needed to run basic Gemini2077 model costs more than a house, but a brain jar with electrodes is cheaper than that. Then the economic incentives will shift the other way.
No I don't think so. We can already create LLMs that are highly efficient and infinitely more knowledgeable than any single human being, completely tuned to the task, without ego or distractions, and they are cheap enough that you can run tens of them in parallel for a few hundred dollars per month. They are also way faster than any human being. And we're three/ four years in this. Imagine 50 years from now.
That's the whole point though - I can't, and I don't think anyone can. Right now the LLMs are just getting bigger and bigger, we're bruteforcing the way out of their stupidity by giving them bigger and bigger datasets - unless something fundamental changes soon that tech has an actual dead end. Hence my (joke-ish) prediction that you'll eventually need a 16PB GPU to run a basic gemini model, and such a thing will always be very expensive no matter how much our tech advances(especially since we are already hitting some technical limits). Human brains won't get any more expensive with time - they already contain all the hardware they are ever going to get - but what might get cheaper is the plumbing to make them "run" and interact with other systems.
Yeah, well, we have a very different view on this- and I know there are two diametrically opposed camps, and I am in the awe-struck one. LLMs are getting bigger and bigger and they're getting much smarter, and all in the space of a few years. They went from making up erratic articles about unicorns to writing complex PRs in codebases of millions of lines of code, solving math olympics level problems, speaking fluently in tens or hundreds of languages and exhibiting a breadth of knowledge than no human being possesses. Considering their size, they are monstrously efficient compared to the human brain. But anyway, this is a matter for a different discussion.
We can already grow brain organoids cheaply and easily enough to be a YouTuber's long-running series, so even if biological somehow gets cheaper than silicon, it still isn't going to be a revived complete human brain from someone who died 50 years earlier and probably retired 20 years before that.
I mean, imagine someone who got themselves cryonically preserved in 1976 getting either revived or uploaded today: what job would they be able to get? Almost no office job is the same now as then; manufacturing involves very different tools and a lot of CNC and robotic arms; agriculture is only getting more automated and we've had cow-milking robots for 20-30 years; cars may have changed the least in usage if not safety, performance, and power source; I suppose that leaves gardening… well, except for robot lawnmowers, anyone who can hire a gardener can probably afford a robo-mower?
Tldr is that for some very limited tasks it might still be preferable to use a human mind, especially if you can run it at 1000x cognitive speed. Or.....it might not. It's sci-fi at this point.
It shouldn't remind you of that, my point is there's little economic use for uploads like this: if thinking meat is cheaper than thinking silicon, train some fresh thinking meat with an electrode array or whatever; if thinking silicon is cheaper, train some fresh thinking silicon.
Non-economic use, that's different of course. Digital afterlife and so on, but as a consumer, not a supplier of anything.
It's the other way around, while initially it will only be available to elites and prisoners (if you are innocently convicted for life, the digitized brain can set the record straight and provide another life, some will take that option, others wont).
As the technology improves, it will be mostly just for the rich and less for prisoners, and as costs fall further there will even be financial pressure for the rest of the population to "go digital": insurance on digitized lifeforms will be much cheaper, replacement robot body parts, replacement electronics, versus expensive healthcare.
Look up the fraction of GDP in developed nations that goes to healthcare and insurance. People will be shamed by the economy as if they are uppity for hanging on to their slow, expensive to feed and maintain meatbag bodies.
> Each time the brain fails a captcha, the brain receives an electric impulse which simulates the most excruciating pain that the brain can respresent, but the brain cannot scream or run away.
What percentage of your life being enjoyable vs horrible suffering makes it worth living?
Maybe you're 80 years old at the time of storing your brain.
Suppose after being revived that regime with capitalist incentives holds for another 200 years during which you live as a brain in a jar, but some cultural revolutions later you are liberated and then proceed to live 10'000 years across any number of bodies and circumstances, which means that in your lifespan of ~10'280 years (not accounting for being in storage) you experienced horrible suffering for about 2% of your life.
This is as much of a contrived example as yours, aside from maybe good commentary on your part on human ethics being shit when profit enters the scene.
Or maybe after 200 years you expire, having at least tried your best at a non-zero chance of extending your lifespan, instead leading to your total lifespan of 280 years being about 71% suffering. Is it better to not have tried at all, then? Just forsake ANY chance of being revived and living for as long as you want and conquering biology and seeing so much more than your 80 year lifespan let you? Should absolute oblivion be chosen instead, willingly, a 100% chance of never having a conscious though after your death again (within our current medical understanding)?
What about the people dealing with all sorts of horrible illnesses and knowing that each next year might be spent in a lot of pain and suffering, even things like going through chemo? Should they also not try? Or even something as simple as all of the people who look for love/success in their lives, and never find any of it anyways and possibly die alone and in squalor? They knew the odds weren't good and tried anyways. A more grounded take would be that those preserved brains are just left to thaw and you probably die anyways without being turned into some human captcha machine, at least having tried. Is it also not worth it in that case, knowing those both potential alternatives?
I guess I'm not making a judgement of what other people should or shouldn't do. Just making up a goofy example to illustrate that the choice is not so obvious to a lot of people, which I think you also illustrate pretty well with your examples. It really depends on the individual. I do think it's worth looking at the incentives of the people funding these companies, because that does give a picture of the probable outcomes.
People will continue working on this sort of thing, that's fine, it really doesn't bother me. If I was forced to make a judgement, I think it's maybe a little silly, but I'm also not out there saving the planet from climate armageddon so I shouldn't cast stones. As a species we are extremely bad at prioritizing for our collective survival and there are a million worse things to be working on.
What percentage of your life being enjoyable vs horrible suffering makes it worth living? I don't know but 99% of my life being used to solve captchas makes it not worth living
>Suppose after being revived that regime with capitalist incentives
Having to provide for other people is literally the same as being trapped in a "I have no mouth and I must scream"-esque torture chamber. Given the historical track record of communism, you're more likely to end in the torture chamber than not in that situation. The curve of history bends towards factory farms.
I read your quote "Having to provide for other people is literally the same as being trapped in a "I have no mouth and I must scream"" and my brain immediately went to the millions of Americans working dead end jobs just to put food on the table for their family. It need not be communism for this to be a reality.
I haven't had that at all, not even a single time. What I have had is endless round trips with me saying 'no, that can't work' and the bot then turning around and explaining to me why it is obvious that it can't work... that's quite annoying.
> Please carefully review (whatever it is) and list out the parts that have the most risk and uncertainty. Also, for each major claim or assumption can you list a few questions that come to mind? Rank those questions and ambiguities as: minor, moderate, or critical.
> Afterwards, review the (plan / design / document / implementation) again thoroughly under this new light and present your analysis as well as your confidence about each aspect.
There's a million variations on patterns like this. It can work surprisingly well.
You can also inject 1-2 key insights to guide the process. E.g. "I don't think X is completely correct because of A and B. We need to look into that and also see how it affects the rest of (whatever you are working on)."
Of course! I get pretty lazy so my follow-up is often usually something like:
"Ok let's look at these issues 1 at a time. Can you walk me through each one and help me think through how to address it"
And then it will usually give a few options for what to do for each one as well as a recommendation. The recommendation is often fairly decent, in which case I can just say "sounds good". Or maybe provide a small bit of color like: "sounds good but make sure to consider X".
Often we will have a side discussion about that particular issue until I'm satisfied. This happen more when I'm doing design / architectural / planning sessions with the AI. It can be as short or as long as it needs. And then we move on to the next one.
My main goal with these strategies is to help the AI get the relevant knowledge and expertise from my brain with as little effort as possible on my part. :D
A few other tactics:
- You can address multiple at once: "Item 3, 4, and 7 sound good, but lets work through the others together."
- Defer a discussion or issue until later: "Let's come back to item 2 or possibly save for that for a later session".
- Save the review notes / analysis / design sketch to a markdown doc to use in a future session. Or just as a reference to remember why something was done a certain way when I'm coming back to it. Can be useful to give to the AI for future related work as well.
- Send the content to a sub-agent for a detailed review and then discuss with the main agent.
Interesting take. Does that mean SWE's are outsourcing their thinking by relying on management to run the company, designers to do UX, support folks to handle customers?
Or is thinking about source code line by line the only valid form of thinking in the world?
I mean yes? That's like, the whole idea behind having a team. The art guy doesn't want to think about code, the coder doesn't want to think about finances, the accountant doesn't want to worry about customer support. It would be kind of a structural failure if you weren't outsourcing at least some of your thinking.
I’m with you, perhaps I just misread some kind of condescension into the “outsourcing your thinking” comment.
We all have limited context windows, the world’s always worked that way, just seemed odd to (mis)read someone saying there’s something wrong with focusing on when you add the greatest value and trusting others to do the same.
It is condescending when antis say AI users do it. It isn’t when a director or team leader does it.
But it’s the same process, which should tell you what’s really going on here. It’s about status, not functionality, and you don’t gain status without controlling other humans.
Everyone who's used Opus knows it's better than the others in a way that isn't captured by the benchmarks. I would describe it as taste.
Lots of models get really close on benchmarks, but benchmarks only tell us how good they are at solving a defined problem. Opus is far better at solving ill-defined ones.
One of the main edges Anthropic has is that "personality tuning" gap. "Nice to use" is a differentiator when raw performance isn't.
OpenAI can sometimes get an edge over Anthropic in hard narrow STEM tasks. I trust benchmarks over vibes there - and the benchmarks show the teams trading blows release after release. Tracking Claude Code vs OpenAI Codex on SWE-bench Verified feels like watching the back alley knife fight of the AI frontier.
But the vibe of "how easy is that model to interact with" and "how easy it is to get it to do what you want it to" does matter a lot when you are the one doing the interacting. And Opus makes for a damn good daily driver.
At this point it's frankly not a fair comparison since DeepSeek 3.2 is now many months old and we're waiting for a newer model which has been rumoured as "any day now" since February. (We'll see).
GLM5, the largest Qwen 3.5 model, and Kimi K2.5 are more fair comparisons, though they are, yes, a bit behind. They're more than capable for routine operations though.
Anyways, I'm back to using Opus & Claude Code after a month on Codex/GPT5.3 and 5.4 and it's frankly a rather obvious downgrade. Anthropic is behind OpenAI at this point on coding models, and there's nothing to say they couldn't fall behind the Chinese models as well.
The moat is very shallow. After the events of the last two weeks there's likely a significant % of international capital very interested in breaching it. I know I would like to see this... Anthropic basically said F U to any non-Americans, and OpenAI is ... yeah.
Dunno, I was using Cursor today and for some reason it decided to swith to GPT 5.3 at some point and I didn't even notice. I was sure that Opus is much better, but who knows?
I have a project where we've had Opus, Sonnet, Deepseek, Kimi, Qwen create and execute an aggregate total of about 350 plans so far, and the quality difference as measured in plans where the agent failed to complete the tasks on the first run is high enough that it comes out several times higher than Anthropics subscription prices, but probably cheaper than the API prices once we have improved the harness further - at present the challenge is that too much human intervention for the cheaper models drives up the cost.
My dashboard goes from all green to 50/50 green/red for our agents whenever I switch from Claude to one of the cheaper agents... This is after investing a substantial amount of effort in "dumbing down" the prompts - e.g. adding a lot of extra wording to convince the dumber models to actually follow instructions - that is not necessary for Sonnet or Opus.
I buy the benchmarks. The problem is that a 10% difference in the benchmarks makes the difference between barely usable and something that can consistently deliver working code unilaterally and require few review interventions. Basically, the starting point for "usable" on these benchmarks is already very far up the scale for a lot of tasks.
I do strongly believe the moat is narrow - With 4.6 I switched from defaulting to Opus to defaulting to Sonnet for most tasks. I can fully see myself moving substantial workloads to a future iteration of Kimi, Qwen or Deepseek in 6-12 months once they actually start approaching Sonnet 4.5 level. But for my use at least, currently, they're at best competing with Athropics 3.x models in terms of real-world ability.
That said, even now, I think if we were stuck with current models for 12 months, we might well also be able to build our way around this and get to a point where Deepseek and Kimi would be cheaper than Sonnet.
Eventually we'll converge on good enough harnesses to get away with cheaper models for most uses, and the remaining appeal for the frontier models will be complex planning and actual hard work.
Good point on the green/red dashboard. The opportunity cost angle is worth adding though. A failed run isn't just the wasted tokens and retry cost - it's also the task that didn't get done and the engineering required to diagnose why. On anything time-sensitive, that compounds fast.
Exactly. At the moment it's close enough to be a wash for some cases, or tilts seriously one direction or other for others. I expect improved harnesses means more and more we'll just be able to re-run a couple of times, and fall back to "escalating" to Sonnet or even Opus, but whenever it involves egineering time, that's a big deal.
I still won't use what? I use Opus now, and I will use Opus then too, but as I clearly stated:
My default model has now dropped to Sonnet, because Sonnet can now do most of my tasks, and we already use Kimi, Deepseek, and Qwen.
They're just not cost-effective enough to be my main driver yet. They are however cheap enough that for things where the Claude TOS does not let me use my subscription, they still add substantial value. Just not nearly as much as I'd like.
The bulk of my tasks won't get harder as time passes, and so will move down the value chain as the cheaper models get better.
For the small proportion of my tasks that benefits from a smarter model, I will use the smartest model I can afford.
Thankfully it's not as bad as that. The 50% that goes red means we re-execute those steps, potentially several times, to see if they succeed, before we even bother manually looking at it. But the overall principle holds: First yoou multiply the cost by re-running, then eventually you either need to kick it up to a more expensive model and/or a human.
But of course this is also only viable for non-latency sensitive work, for starters.
> Can't we have a system that is optimized for the notes that are actually played in a song rather than the hypothetical set? And what if the optimization is done per note rather than over an entire song?
You can. It’s called adaptive tuning, or dynamic just intonation, and it happens naturally for singers with no accompanying instruments.
It’s impractical on a real instrument, but there’s a commercial synthesiser implementation called hermode tuning.
You’re trading one problem for another, though. No matter how you do this, you will either have occasional mis-tuning or else your notes will drift.
In addition to singers, adaptive tuning is something which happens naturally for fretless stringed instruments (violin, etc), brass instruments with slides (most prominently the slide trombone but in fact many (most?) others), woodwind instruments where the pitch can be bent like saxophone, and so on.
I used to play fretless bass in a garage hip hop troupe that played with heavily manipulated samples that were all over the place in terms of tuning instead of locked to A440, forcing adaptations like "this section is a minor chord a little above C#".
Adaptive tuning is hard to do on a guitar because the frets are fixed. String bending doesn't help much because the biggest issue is that major thirds are too wide in equal temperament and string bending the third makes pitch go up and exacerbates the problem.
You can do a teeny little bit using lateral pressure (along the string) to move something flat. It's very difficult to make adaptations in chords though. A studio musician trick is to retune the guitar slightly for certain sections, though this can screw with everybody else in the ensemble.
Played trombone many years ago, but never well enough to ever adjust that finely (at least not consciously?). The tuning slide on the third valve on a trumpet usually has a finger fork/loop so that it can be tuned in realtime. I believe the first valve on higher end trumpets similarly has a thumb fork for the same reason.
I played trombone in high school, never very well, but I definitely adjusted like this. Actually, although it was a slide trombone, I'm talking about adjusting automatically with embouchure. Someone would play the reference note, I'd match (in 1st position) but bend my pitch to match. The band teacher once complimented me on the adjustment. Which was stupid, because (1) I wasn't doing it intentionally, and (2) the adjustment only lasted during tuning; as soon as we started playing, I was right back out of tune. I never did learn to suppress the adjustment so I could actually fix the tuning.
But with the way I played, I'm not even sure how much it mattered. The best tool for enhancing my playing would've been a mute. (And it would have been most effective lodged in my windpipe.)
It's still not at all obvious to me that LLMs work in the same way as the human brain, beyond a surface level. Obviously the "neurons" in neural nets resemble our brains in a sense, but is the resemblance metaphorical or literal?
Digital neural networks and "neurons" were already vastly simpler than biological neural networks and neurons... and getting to transformers involved optimisations that took us even further away from biomimicry.
reply