The webpage linked is an example of everything I wish people would stop doing in web design.
Fortunately, at the bottom there is a link to the "technical documentation" (https://squeezlabs.github.io/handcrank/) which is vastly improved (aside from being light-mode-only and linked from a dark-mode-only marketing page). It also gives me much more interesting information (specifically: models that can apparently run acceptably on a Pi 5).
Please let me read your content with a scrollbar that works the way scroll bars are supposed to, rather than turning everything into a weird slide show where you don't actually know when the next slide is coming. Please let me just click on buttons that look like links to more information, without JavaScript.
Why can't technical people appreciate that us, the silent majority, love having our scroll hijacked? I can't remember the last time I used a scroll bar to navigate a website, but using it to navigate between choppy javascript keyframes fills me with joy.
Totally agree on the atrocious landing page. The technical one is much better, although the power supply circuit by using a resistive balancer and a linear regulator wastes some good power for nothing.
My partner just got a rowing machine that offered "watts" as a unit of how hard you're going (like "calories" or "mph") and got me wondering if they made rowing machines that could slowly charge a battery, and how much I'd need to row to power one of them fancy newfangled M5 Max MacBooks answering prompts.
All that to say, CrankGPT, I am your target demographic and if you don't respond to my request for a demo I'll be cranking my keyboard with bad reviews online. Or cranking a rowing machine that powers an LLM to do it for me. Wait...
I took several biomechanics classes as electives back in my undergrad, and in one assignment I remember comparing the energy outputs between the human and robot equivalents of different tasks, whether or not the robot was humanoid in how it was designed. The most impressive think that stuck with me is that humans are incredibly efficient, from an energy perspective, in anything we do, compared to machines. Every time we delegate a task to a machine, we are using several orders of magnitude of energy to do the same thing. For most tasks, it feels wrong, but it doesn't make me any more willing to give up my car. Maybe if I lived outside the US.
If you live in most places in the US other than the urban heart of a few very large cities you have to take a huge hit to your ability to get places in a reasonable time frame without a car. I have hope some more cities other than NYC are improving the situation, but as it is the closest I got to using public transit for a commute was when I was going to one of our other offices in a different downtown area I would drive my car to the park n ride to take the train the rest of the way. The train saves time and sanity because traffic downtown is a nightmare, but that drive takes 5 minutes, and it would add 20+ minutes if I had to walk to the closest bus stop so I could take the bus up to the train station.
No, it's evolution. Mammals burn a ridiculous amount of energy just existing, so evolutionary pressures tend toward more efficient muscles and body geometry.
Electrochemical reactions in your muscles combined with the mechanical advantage from the geometry of your joints and ligaments is simply more energy efficient than most mechanical or electromechanical systems. On top of that, our learned and evolved kinematic algorithms result in vastly more efficient control. Humans tend to be pretty good at using only exactly as much energy as required for a given action. Overshoot is quite limited compared to robots.
Your suggested actions seem inefficient, but if you look at the actual energy expenditure, mechanical means are much worse simply because mammalian muscle is so efficient.
There's a difference between consistency and efficiency.
I read efficiency as "Energy inputted to accomplish a task", in which case, biological systems are far more efficient than current-day mechanical ones. It's a tradeoff.
If you're comparing raw calories to output, yes. Even gasoline has a caloric value, but humans can't drink gasoline. Growing and preparing food for human consumption uses a lot more energy than pumping and refining gasoline, so at the end of the day, human efficiency gains are not that impressive.
For more reference how insane 700W is, the average FTP of a world tour pro road cyclist (i.e., Tour de France) is ~350-420W/6-7W/kg. FTP (Functional Threshold Power) being the avg you can sustain for an hour without fatiguing.
My own is ~250W @ 3.12W/kg. I can't even hit 700W yet, let alone for over a minute. My 5 second power is ~640W.
It’s basically well known to cyclists that training with a power meter that tells you “watts” more accurately gauges effort and caloric expenditure. (Heart rate gauges subjective effort however, taking into account stress, caffeine intake, etc.)
It’s also interesting that the industry has settled on using watts to mean rate of useful work whereas calories to mean the total work including inefficiencies, despite that calories is just a unit of energy. A rule of thumb for cyclists is that in addition to usual unit conversions, the “calories” figure should be multiplied by four to account for energy expended by the body but not used for rotating the pedals. I don’t use rowing machines but I’m sure they would have a similar conversion factor in order to calculate calories.
However you can expect around only 3 watts of output at normal speeds and you will need to put in around 5-7 watts of power for the same speed. This is barely enough to trickle charge modern phones.
Annoying. Blame the Germans and their lighting laws for bicycles. I want human powered USB-C with enough oomph to power a modest sound system or lights, whilst charging my phone. The allure of USB-C is what interests me, but 3W is not much to work with. Also annoying, no rear dynamo for my bike, so I can't even double up to 6W.
I will probably end up with no sound system and just expensive dynamo lights, using a USB speaker that doubles up as a power brick.
There is a nice USB battery kit for dynamo that fits in the steerer, so it is soldering iron time for that, so might as well learn how to do USB-C power things.
One day there will be structural solar panel batteries that can be 3D printed into lightweight bicycle frames, so maybe I will stick to throwaway lights until then!
An untrained cyclist is not able to maintain 200 watts.
For an average untrained male cyclist who is 175lb, they should be able to maintain 1.5-2 w/kg over an hour, or 120-160watts.
A beginner cyclist who's been cycling recreationally over over a year should be able to attain 2-2.5w/kg which is 160-200 watts.
A recreational cyclist who's be training for several years should be able to maintain 200 watts.
Trust me, I'm a cyclist, and I cycle with a power meter.
As an average male who is ~175lbs and untrained at cycling, this is hugely validating for my terrible idea; 140 watts is the max charging speed for 16" M5 MacBooks. I can finally stop thinking for myself and have my computer do it all for me, powered by my big beefy legs.
140 watts is the FTP. That means you can do it for an hour, and it will be an extremely exhausting hour and you will want at least two days of rest to recover from this workout before doing it again.
This exactly. FTP is functional threshold power, it's the maximum you can physically maintain in zone 4 heart rate for 40-60 minutes and it's physically exhausting. This is hard work, hardest work honestly. If you have gas left in the tank after you could had a higher FTP. I'll get out on the bike and do a zone 4 workout and keep a steady state of 180-200 watts through the first 20-30 minutes, by the end of the hour or hour and a half of my workout, even with breaks to break, reset heart rate, have some water, salt and glucose, I'm down to 120-150 watts at the end of the ride.
Basically, you can probably charge your macbook at peek power for an hour every other day, or every day for a short while if you're okay with burning out eventually.
Expect to need to eat 400-600 calories and a lot of water each time you do this.
I just rode with an untrained cyclist (new to cycling) yesterday. The person averaged 80W over five hours. It’s about right for an actual untrained cyclist.
I've had the same thought and it's been a topic I've been this close to talking about at parties. If I did I'm sure I'd bore everyone to death.
Considering the difficulty of sustaining 700 watts vs 350 watts, we could've had some very well-burnt toast if they uninstalled the heating coils for the 2nd piece of bread!
I thought toasters took ~1.6 kW, just like any other resistive heating device (space heater, oil-filled radiator, microwave oven, oven oven, hairdryer, kettle, under-sink water heater: it's allways 1500-1850W!) except for the ones on special circuits (shower, stove). Turns out, our toaster draws 850W!
While watching the video, I was wondering how they modified the >1kW device to produce a toasted toast in that short amount of time (I guess you could substitute instantaneous power for time up to a point, but the video wasn't that long), thinking maybe they removed one of the sides' circuits. Now I'm disappointed as well. Thanks xD
I can do high level thinking for around 6 hours with just two scrambled eggs and a cup of coffee.
What I need is something to prevent me from context drift. /starts googling how many scrambled eggs are equivalent to the energy consumed by a data center. Google how many chickens are in the world.../
Consider that you can do this as a working day because someone else is plowing the fields that grow your food, probably burning stored energy in fuels for the process
We're sadly not that efficient. The 150kcal/6h=600kcal/day you've mentioned aren't enough, and it takes more than 600kcal to create 600kcal plus transportation into your home
Besides, we won't stop existing, so any math about "chatgpt uses X kW and so it's better than hiring another human" doesn't work out. The human doesn't stop burning fuels when not in use: any LLM usage is additional energy that needs to be generated while staying within CO2 budgets
Or we could... hear me out... build more power plants. The de-growth stuff is pretty evil when you take it to it's logical conclusion of population control.
There's a bit of a difference between population control and reaching a sustainable equilibrium. One can also argue the death of all life on earth is a pretty evil logical conclusion of infinite growth on a finite planet.
Do not confuse curiosity and caring for extremism.
If someone said they pick up trash on the side of the road to help the environment you wouldn't say the logical conclusion of their ideology would be that they become the unabomber
It's not more sustainable though. It takes more energy to grow the food to get the calories to metabolize to turn the crank than it would to run an electric motor or a engine to do the same work.
instead of complaining about the website, which wouldn't be allowed, i am just going to link this much better reading experience instead: https://squeezlabs.github.io/handcrank/ (from the "technical documentation" link at the bottom of that... page thing)
> We chose a cheap off-the-shelf switchable voltage 20W hand-crank generator marketed for emergency USB charging. The Pi normally draws around 1.5A, but when it’s working hard (as it does when doing inference on the CPU), its current requirements can increase substantially, causing the generator voltage to sag below the Pi’s required 4.8V or even, in the case of a momentary 5A spike, to trigger the generator’s internal overcurrent protection and shut off the voltage output entirely, causing the Pi to brown out.
> To ensure the Pi sees a steady voltage when the full inference stack kicks in (and to afford crankers a little rest), we built a custom capacitor board [https://docs.google.com/spreadsheets/d/1Zv_Hsinvx_sWtdur4iWY...] to smooth out the generator’s output and act as a short-term (~20 second) power reservoir.
Somewhat off-topic, but could this capacitor board work with a small-ish 5V USB solar panel? I'm not great with electronics, but it seems like just the solution for a device I want to run with the panel.
My vague impression was it's not kosher per the USB spec to just stick a capacitor across the supply to even out the brown-outs, but this looks like it's doing some other stuff.
Playdate (the came console) is amazing for this! I was really bored once and built a Claude Code remote control for Playdate.
Voice recognition was done via parrot + handy.computer
Basically: different key combos were tied to different actions, e.g. \
- hold A to speak
- move the crank slowly to navigate
- crank super fast to send the prompt
Eventually this became a universal remote control for the computers in my home (YAML file with bindings from Playdate UI → A11y events).
Using the crank to control movies is fun!
(I can share the source -- just let me know if this is actually useful)
Also, I feel like the author and me have similar hobbies. A few years back I almost won a (re-sellable on Ebay) award for https://meat-gpt.sonnet.io !
I really wanted to like the Playdate but it's such an unbelievably overpriced and underspecced toy. But the crank, which really looked like a gimmick, is oddly quite nice to have and I wish my kids' Retro Arcade had one.
EDIT: To elaborate a bit, if you are burning oil or gas in a turbine, you do not need a cooling tower, the waste heat goes into the atmosphere with the exhaust. If you use fossil or nuclear fuels to produce steam for a steam turbine, you either need a river with enough flow to not boil all the fish if you reject the waste heat into it or you need a cooling tower to reject the heat into the atmosphere.
When I was at Peloton a long time ago, someone proposed an April Fools’ joke where we could announce a dynamo add-on that would let you power your house from your bike.
The math isn’t as bad as you might think: 200Wh (about a 60 min, somewhat intense ride) seems to be about 20 minutes for a H/B100. Still 3x, but not bad at all!
I remember the idea being dismissed quickly because people would likely actually want it.
200W is pretty good FTP for an "untrained" cyclist. FTP is the maximum you can sustain for a single hour, but you're not going to do multiple back-to-back.
Obviously the peloton crowd is biased towards people who will have better endurance and higher FTP, but basically the upshot is you could run one card for one hour with the effort of a 100km long ride which most recreational cyclists do once a week.
I bet if you took one of Taalas' cards which consumes 200 watts for 14,000 tokens/second [0] and slowed it down by a factor of 10, it would actually be quite reasonable to power by bicycle.
This is so cool; I wonder what the roadmap includes? I'd love to see if this could evolve from gag product/gift idea, to a solar-powered inference box on slightly more robust hardware, for instance, completely freeing the owner from any need to purchase a subscription or rely upon the grid for inference. That would be freaking cool and I'm sure would sell a ton if the models that could run on the hardware were sufficiently capable and could be piped into a laptop (which seems like the easiest part). What would the power requirements be to run a more capable model than the ones used, maybe a DeepSeek open-source model? Really curious but I'm unfamiliar with the technical details that might go into such calculations.
Not sure it's possible to say this without being pithy, but haven't there been stories told, perhaps TV episodes and films made, regarding the concept of using human bodies to generate power in support of machine intelligence?
Seems a little on the nose to me, but I guess some days it's hard to tell what's a gag and what's a legit pitch.
ACTUALLY xD
Transferring power from gym bicycles is one of the smartest things. But they probably won’t even be enough for the AC to run - still a lot of power and heat lost.
They're not raising at $1bn valuation until they pull out a slide deck with hockey stick curves showing the point that human brawn can out-think human brain...
I thought this was going to be about LLM-generated conspiracy theories, but the website was funny anyway.
There's a technical documentation link at the bottom of the page that documents an actual working hand-crank-powered Raspberry Pi that runs a local model.
Oh that's a fun idea... I've been poking at a "ghost phone" that synthesizes a personality and matching voice, which rings at random times on an old candlestick rotodial handset.
"Conspiracy call-in" on a CB radio would be a good variation!
Whether it's a parody or satire or a fake product page for a DIY project, emphasizing the climate cost of AI while using AI to generate every video there kind of ruins it.
Well, I assume this isn't real, but.. I'd want to know the actual number - how much more (than in the demo) work'd we need to do (energy to produce) to actually power a CPU/GPU which could use real small on-device models..
I really want to know, no matter how big that number unfortunately is.
I watched the "is this real" video and I'm still* not sure if it's real lol.
Regardless of this website presentation, the idea is sound and I'm behind it. We need to stop giving everything away to a hyper concentrated group of wealthy super elites that do not have our best interests at heart. We already have disappointing politicians that are elected. Now we also have disappointing unelected rich decision makers altering our lives based on what bar they had their next back of napkin scheme at.
Sorry, I watched it sans audio. I don't have twitter (or insta or fb) so their links default to muted on click. That said, it would still be good if we could address general comprehension, exercise, and private AI usage all with this one hand crankable device!
Fortunately, at the bottom there is a link to the "technical documentation" (https://squeezlabs.github.io/handcrank/) which is vastly improved (aside from being light-mode-only and linked from a dark-mode-only marketing page). It also gives me much more interesting information (specifically: models that can apparently run acceptably on a Pi 5).
Please let me read your content with a scrollbar that works the way scroll bars are supposed to, rather than turning everything into a weird slide show where you don't actually know when the next slide is coming. Please let me just click on buttons that look like links to more information, without JavaScript.
You can scroll normally, with all your favorite keys, or go super fast to the bottom
It’s just scroll animations. Bad ones, admittedly.
All that to say, CrankGPT, I am your target demographic and if you don't respond to my request for a demo I'll be cranking my keyboard with bad reviews online. Or cranking a rowing machine that powers an LLM to do it for me. Wait...
Might this just be selection bias? I mean, if humans can't do a task efficiently, we're not going to do the comparison with a machine.
Some actions we do seem (to me) very inefficient when compared with machines. For example: grating carrots and brushing teeth.
Electrochemical reactions in your muscles combined with the mechanical advantage from the geometry of your joints and ligaments is simply more energy efficient than most mechanical or electromechanical systems. On top of that, our learned and evolved kinematic algorithms result in vastly more efficient control. Humans tend to be pretty good at using only exactly as much energy as required for a given action. Overshoot is quite limited compared to robots.
Your suggested actions seem inefficient, but if you look at the actual energy expenditure, mechanical means are much worse simply because mammalian muscle is so efficient.
I read efficiency as "Energy inputted to accomplish a task", in which case, biological systems are far more efficient than current-day mechanical ones. It's a tradeoff.
My own is ~250W @ 3.12W/kg. I can't even hit 700W yet, let alone for over a minute. My 5 second power is ~640W.
Crazy numbers.
Lost the opportunity to say "bread".
https://youtu.be/7qxTKtlvaVE
I would suspect my equivalency to be about 1/3rd a Robert [unit of measure from vidlink].
Even without a battery, I could easily imagine designing an efficient single slice toaster that could handily brown a pop tart on a 300W budget.
If you like then "golden," perhaps the entire box.
It’s also interesting that the industry has settled on using watts to mean rate of useful work whereas calories to mean the total work including inefficiencies, despite that calories is just a unit of energy. A rule of thumb for cyclists is that in addition to usual unit conversions, the “calories” figure should be multiplied by four to account for energy expended by the body but not used for rotating the pedals. I don’t use rowing machines but I’m sure they would have a similar conversion factor in order to calculate calories.
However you can expect around only 3 watts of output at normal speeds and you will need to put in around 5-7 watts of power for the same speed. This is barely enough to trickle charge modern phones.
I will probably end up with no sound system and just expensive dynamo lights, using a USB speaker that doubles up as a power brick.
There is a nice USB battery kit for dynamo that fits in the steerer, so it is soldering iron time for that, so might as well learn how to do USB-C power things.
One day there will be structural solar panel batteries that can be 3D printed into lightweight bicycle frames, so maybe I will stick to throwaway lights until then!
The Concept2 rowing machines can power itself using the power you generate by rowing, so we're partly there.
It's the only unit that makes sense tbh
For an average untrained male cyclist who is 175lb, they should be able to maintain 1.5-2 w/kg over an hour, or 120-160watts. A beginner cyclist who's been cycling recreationally over over a year should be able to attain 2-2.5w/kg which is 160-200 watts. A recreational cyclist who's be training for several years should be able to maintain 200 watts.
Trust me, I'm a cyclist, and I cycle with a power meter.
Basically, you can probably charge your macbook at peek power for an hour every other day, or every day for a short while if you're okay with burning out eventually.
Expect to need to eat 400-600 calories and a lot of water each time you do this.
Considering the difficulty of sustaining 700 watts vs 350 watts, we could've had some very well-burnt toast if they uninstalled the heating coils for the 2nd piece of bread!
While watching the video, I was wondering how they modified the >1kW device to produce a toasted toast in that short amount of time (I guess you could substitute instantaneous power for time up to a point, but the video wasn't that long), thinking maybe they removed one of the sides' circuits. Now I'm disappointed as well. Thanks xD
What I need is something to prevent me from context drift. /starts googling how many scrambled eggs are equivalent to the energy consumed by a data center. Google how many chickens are in the world.../
We're sadly not that efficient. The 150kcal/6h=600kcal/day you've mentioned aren't enough, and it takes more than 600kcal to create 600kcal plus transportation into your home
Besides, we won't stop existing, so any math about "chatgpt uses X kW and so it's better than hiring another human" doesn't work out. The human doesn't stop burning fuels when not in use: any LLM usage is additional energy that needs to be generated while staying within CO2 budgets
I feel like it is not only an interesting engineering challenge but one that might lead to a more efficient and sustainable framing.
What are you balancing if not human growth? And how do you plan to do that?
If someone said they pick up trash on the side of the road to help the environment you wouldn't say the logical conclusion of their ideology would be that they become the unabomber
> To ensure the Pi sees a steady voltage when the full inference stack kicks in (and to afford crankers a little rest), we built a custom capacitor board [https://docs.google.com/spreadsheets/d/1Zv_Hsinvx_sWtdur4iWY...] to smooth out the generator’s output and act as a short-term (~20 second) power reservoir.
Somewhat off-topic, but could this capacitor board work with a small-ish 5V USB solar panel? I'm not great with electronics, but it seems like just the solution for a device I want to run with the panel.
My vague impression was it's not kosher per the USB spec to just stick a capacitor across the supply to even out the brown-outs, but this looks like it's doing some other stuff.
Voice recognition was done via parrot + handy.computer Basically: different key combos were tied to different actions, e.g. \
Eventually this became a universal remote control for the computers in my home (YAML file with bindings from Playdate UI → A11y events). Using the crank to control movies is fun!(I can share the source -- just let me know if this is actually useful)
Also, I feel like the author and me have similar hobbies. A few years back I almost won a (re-sellable on Ebay) award for https://meat-gpt.sonnet.io !
(I lost to a gallery of 3d sandwiches)
What I love is this quote is super-imposed with a background image that has gas-burning smokestacks but also nuclear cooling towers in the same field.
This is a bit representational of this particular line of protest against AI - just super confused about it all and thrashing out.
Green energy has been (technologically) solved, but instead we want to go back to manual labor as a source of power? Hilarious.
[1] https://en.wikipedia.org/wiki/Coal-fired_power_station#/medi...
EDIT: To elaborate a bit, if you are burning oil or gas in a turbine, you do not need a cooling tower, the waste heat goes into the atmosphere with the exhaust. If you use fossil or nuclear fuels to produce steam for a steam turbine, you either need a river with enough flow to not boil all the fish if you reject the waste heat into it or you need a cooling tower to reject the heat into the atmosphere.
The math isn’t as bad as you might think: 200Wh (about a 60 min, somewhat intense ride) seems to be about 20 minutes for a H/B100. Still 3x, but not bad at all!
I remember the idea being dismissed quickly because people would likely actually want it.
Obviously the peloton crowd is biased towards people who will have better endurance and higher FTP, but basically the upshot is you could run one card for one hour with the effort of a 100km long ride which most recreational cyclists do once a week.
[0]: https://www.sdxcentral.com/news/chip-designer-taalas-bets-on...
And also it mades me realize that we would all be way more healthy if we powered our laptops from bike power.
Seems a little on the nose to me, but I guess some days it's hard to tell what's a gag and what's a legit pitch.
No way there are still so many human devs in the office.
(although I salute the attention it brings to an important cause)
There's a technical documentation link at the bottom of the page that documents an actual working hand-crank-powered Raspberry Pi that runs a local model.
"Conspiracy call-in" on a CB radio would be a good variation!
I really want to know, no matter how big that number unfortunately is.
Edit: I confess, kind of tempted, 41 EUR for a "big" one https://www.bol.com/be/fr/p/zimoros-handslingergenerator-kra...
Regardless of this website presentation, the idea is sound and I'm behind it. We need to stop giving everything away to a hyper concentrated group of wealthy super elites that do not have our best interests at heart. We already have disappointing politicians that are elected. Now we also have disappointing unelected rich decision makers altering our lives based on what bar they had their next back of napkin scheme at.