News for hackers. Stuff that matters.
Re:Show HN: Weight Lifting Algorithm Based on 5M Workouts by jakemor on Thursday April 11, 2019 @10:21AM (#19634771)
Fellow hackers,

After collecting an insane amount of data with my app Lift Log, I decided to design an algorithm that optimizes for accelerated muscle growth over a short period of time.

It currently outperform human trainers by 1-2x in the first 60 days of training, using a very simple algorithm. The opportunity to change how people, athletes, and soldiers train is huge.

We're cashflow positive, raised a small seed, and are hiring data scientists & engineers in NYC to hack the human body for efficiency.

Hope to answer any questions you may have!

Re:Show HN: Weight Lifting Algorithm Based on 5M Workouts by Stunting on Thursday April 11, 2019 @11:15AM (#19635360)
I'm all here for better lifting apps, but I have lots of questions.

How do you qualify the "trainers," that your algorithm outperformed? Most basic strength and conditioning books will outperform the average chain gym trainer.

How many of your the trainers in your sample size were Strength and Conditioning Coaches at reputable performance gyms? Would any qualify under the following statement from the National Strength and Conditioning Association's Website:

"The proper education consists of an undergraduate degree in an exercise science-related field. Such a degree is imperative because it forms the foundation of knowledge that underpins the profession. An exercise science-related degree affords the capability to practice an evidence-based approach to strength and conditioning."

What are your results for longer than 60 days? Do you continue to outperform human trainers at the 1 yr and 5 yr marks? For those that are uninitiated, 60 days is a incredibly short time in the world of Strength and Conditioning. Doing almost anything differently that your normal routine, consistently, over 3 months will yield positive results before plateaus set in.

How are you quantifying your results?

Are you raising the 1 rep maxes of your athletes 1-2x than more than Human Trainer? Their 10 rep max? Their max broad jump? Their 40 yard dash? Their 1 mile time? Their muscle mass? Lowering their Body fat percentage?

I want this to work. I crave a simple logging app that will allow me to enter a variety of exercises and an AI that would help with programming would be huge. I would be willing to you pay your subscription of $8 a month if it were true.

Re:Show HN: Weight Lifting Algorithm Based on 5M Workouts by jakemor on Thursday April 11, 2019 @12:31PM (#19636321)
>> How do you qualify the "trainers," that your algorithm outperformed? Most basic strength and conditioning books will outperform the average chain gym trainer.

Trainers use Lift Log to keep track of their clients workouts during training sessions. They were identified and their data was compared

>> How many of your the trainers in your sample size were Strength and Conditioning Coaches at reputable performance gyms? Would any qualify under the following statement from the National Strength and Conditioning Association's Website:

"The proper education consists of an undergraduate degree in an exercise science-related field. Such a degree is imperative because it forms the foundation of knowledge that underpins the profession. An exercise science-related degree affords the capability to practice an evidence-based approach to strength and conditioning."

Not sure! but they worked at or owned their own gyms so I'm sure they are certified. Only trainers who trained using the same exercises as FitnessAI were compared

>> What are your results for longer than 60 days? Do you continue to outperform human trainers at the 1 yr and 5 yr marks? For those that are uninitiated, 60 days is a incredibly short time in the world of Strength and Conditioning. Doing almost anything differently that your normal routine, consistently, over 3 months will yield positive results before plateaus set in.

I'm not sure! We've only been around for a few months so it's hard to say. That said, we have the data and users to optimize for this once users begin to plateau

>> How are you quantifying your results? Are you raising the 1 rep maxes of your athletes 1-2x than more than Human Trainer? Their 10 rep max? Their max broad jump? Their 40 yard dash? Their 1 mile time? Their muscle mass? Lowering their Body fat percentage?

% change in total volume lifter

>> I want this to work. I crave a simple logging app that will allow me to enter a variety of exercises and an AI that would help with programming would be huge. I would be willing to you pay your subscription of $8 a month if it were true.

try it out! And message me in app for a free 30 days

Re:Show HN: Weight Lifting Algorithm Based on 5M Workouts by Stunting on Thursday April 11, 2019 @01:55PM (#19637314)
No thank you. You have not qualified your trainers as anybody other than "but they worked at or owned their own gyms so I'm sure they are certified." That is naive or willingly ignorant. Spending time any time in a chain gym and interacting with the hired trainers would illustrate that.

Your methodology doesn't disqualify anybody who could have started with incredible small lifts and then added higher lifts at the end of 60 days. That would not reflect actual development.

Most importantly, I am aware of how to do steady slate programming, which seems to be what your App is selling. I have access to spreadsheets and the ability to program by percentages, as do most people.

Your app would be 100x better if I simply were able to input my One Rep Maxes and then have it spit out a 3 month plan of 5x5 training starting around 70% of my max increasing my loads by 5% a week, with a deload week every 4-6 weeks. Perhaps adding in weeks with 3x5, and 3x3 as it gets near max. However, as you likely know, no one is going to pay monthly for that.

Instead, you have packaged common knowledge as AI. You have taken data from users (that you charged a monthly subscription) that are not qualified to give educated information, failed to qualify which of them were experts, and are attempting to pass it off as some sort of hive learning. For that you want $8 a month (AND their data, to be used for who knows what in the future) from people who are looking to better themselves.

Re:Show HN: Weight Lifting Algorithm Based on 5M Workouts by jakemor on Thursday April 11, 2019 @02:15PM (#19637494)
Sorry you disagree mate!

There are thousands happy users who are seeing real results, I don't know what else to say.

Please don't forget that the average user doesn't know how to lift weights, and doesn't want to pay $50-150/training session. FitnessAI offers a real alternative.

>> Your app would be 100x better if I simply were able to input my One Rep Maxes and then have it spit out a 3 month plan of 5x5 training starting around 70% of my max increasing my loads by 5% a week, with a deload week every 4-6 weeks. Perhaps adding in weeks with 3x5, and 3x3 as it gets near max. However, as you likely know, no one is going to pay monthly for that.

FitnessAI basically does a variation of this, but it's informed by data rather than being anecdotal

Re:Show HN: Weight Lifting Algorithm Based on 5M Workouts by Stunting on Thursday April 11, 2019 @02:33PM (#19637696)
Man you just do not get it. a 5x5 or 3x5 plan is not anecdotal. It's based on empirical evidence, pioneered in most part by the Russians in the 70s, with knowledge existing for decades, if not centuries prior.

Your app does not invent a new way to program working out. Your app reflects proof that science we already know about has results that have already been proven to be true. And you charge people monthly for this.

Re:Show HN: Weight Lifting Algorithm Based on 5M Workouts by jakemor on Thursday April 11, 2019 @02:50PM (#19637907)
Sorry again Stunting, I guess we'll just have to agree to disagree.

Good luck in your fitness endeavors :)

Re:Show HN: Weight Lifting Algorithm Based on 5M Workouts by jdotjdot on Thursday April 11, 2019 @11:03AM (#19635175)
This looks exciting, but as other commenters noted, you need to provide more detail about the underlying data and ML strategy. There are many companies peddling both questionable "AI" and questionable "sure-to-work" training and fitness practices, and you sit in the intersection of both. Even if what you offer is real and truly works (and I hope it does!), you'll need to show a bit more to get people to trust you and invest their time in following FitnessAI's instructions.
Re:Show HN: Weight Lifting Algorithm Based on 5M Workouts by jakemor on Thursday April 11, 2019 @11:27AM (#19635543)
Noted! Thanks so much for your input — I'll work on a post
Re:Show HN: Weight Lifting Algorithm Based on 5M Workouts by halbritt on Thursday April 11, 2019 @11:37AM (#19635669)
I can outperform 90% of human trainers as well in the first 60 days of training as most human trainers are clueless and the novice effect is real.

Do any compound movement with as much weight as you can manage for a set of 5. Do a couple more sets. Add 10 pounds to the bar every subsequent workout until failure. Then add 5 pounds to the bar every workout subsequent to that.

That'll get most healthy young people through the first 8 weeks of training. Train two or three times a week and you'll add 80-100 pounds to every lift.

The caveat being that small people will run out of linear gains in their upper body during that time.

Re:Show HN: Weight Lifting Algorithm Based on 5M Workouts by jakemor on Thursday April 11, 2019 @12:35PM (#19636359)
>> Do any compound movement with as much weight as you can manage for a set of 5. Do a couple more sets. Add 10 pounds to the bar every subsequent workout until failure. Then add 5 pounds to the bar every workout subsequent to that.

The app basically does this very systematically. Give it a shot! It's simple and effective

Re:Show HN: Weight Lifting Algorithm Based on 5M Workouts by halbritt on Thursday April 11, 2019 @02:03PM (#19637388)
Sure, so is the program I described. No app required. One can just use a piece of paper to log workouts.

What's missing is how well your app works with people that are reasonably well-adapted?

Re:Show HN: Weight Lifting Algorithm Based on 5M Workouts by dyeje on Thursday April 11, 2019 @11:05AM (#19635191)
What exactly does the algorithm optimize for? I see you mention muscle growth, but does this come at the expense of other factors such as preventing injuries?
Re:Show HN: Weight Lifting Algorithm Based on 5M Workouts by jakemor on Thursday April 11, 2019 @12:32PM (#19636329)
% change in total volume lifted over time
Re:Show HN: Weight Lifting Algorithm Based on 5M Workouts by davidclark22 on Thursday April 11, 2019 @01:32PM (#19637016)
>> % change in total volume lifted over time

How do you define volume? Total reps per workout? Total reps x total weight per workout?

Re:Show HN: Weight Lifting Algorithm Based on 5M Workouts by jakemor on Thursday April 11, 2019 @02:24PM (#19637596)
reps x weight per exercise, as its directly correlated to the amount of work being done (since distance is constant).
Re:Show HN: Weight Lifting Algorithm Based on 5M Workouts by superpermutat0r on Thursday April 11, 2019 @11:09AM (#19635261)
I can easily outperform your app on any client by assigning more sets per during the week, more tonnage, than the app. It's the fine line of not getting adapted to bunch of work in the beginning to enable long term gains. People should be doing the minimal necessary dosage of stimulus.
Re:Show HN: Weight Lifting Algorithm Based on 5M Workouts by _Schizotypy on Thursday April 11, 2019 @10:29AM (#19634870)
What's your evidence for this "outperforms human trainers" claim? I only see one single blog post and not much else on the website.
Re:Show HN: Weight Lifting Algorithm Based on 5M Workouts by jakemor on Thursday April 11, 2019 @10:37AM (#19634933)
Sorry for the confusion — website is still new.

In analyzing people's percent change in total volume lifted over 60 days on FitnessAI vs users on Lift Log (a basic weightlifting journal app)

Re:Show HN: Weight Lifting Algorithm Based on 5M Workouts by tracer4201 on Thursday April 11, 2019 @11:04AM (#19635181)
Is the claim based on beating Lift Log alone or were there others as well. Assuming this was a scientific study, was there a paper published? I’m on mobile but didn’t see a link on your website.

App looks pretty cool either way - will check it out.

Re:Show HN: Weight Lifting Algorithm Based on 5M Workouts by Obi_Juan_Kenobi on Thursday April 11, 2019 @03:04PM (#19638050)
60 days isn't long enough to evaluate program performance; that's about the length of one mesocycle and thus it is not sufficient for evaluating how that performance carries forward. E.g. you could compare a peaking cycle to a volume cycle and falsely conclude that the peaking cycle is strictly better for strength gains. However, peaking is not sustainable and only provides temporary adaptations.

You're also analyzing increases in volume, whereas most people will be interested in increasing 1RM strength. Volume is obviously a key component of this, but volume as a metric is also tricky. The number of challenging sets is likely a better indicator of training stress and a driver of muscle growth.

https://www.strongerbyscience.com/can-we-predict-muscle-grow...

For instance, it's trivial to increase total volume by performing high reps of a light weight over a long period of time. I can spend all day lifting 35% of my 1RM, accumulate astronomical volume, but see little to no training benefit.

I also take issue with your claim that your algorithm 'optimizes for accelerated muscle growth', yet it does not appear you have measured muscle growth at all. It's an absurd claim to make; you've optimized for a setsXrepsXweight volume metric. This is perhaps a semantic issue, but deceptive nonetheless.

______

I imagine that the app results in reasonable programming, and it may even be effective, but you have a long way to go before claiming any kind of empirical validation.

Re:Show HN: Weight Lifting Algorithm Based on 5M Workouts by eranation on Thursday April 11, 2019 @11:13AM (#19635320)
If a user has some constraints such as lack of some equipments or some movements they can’t do due to health reasons, do you guys handle it? It would be amazing if your algorithms can work around those constraints (if there are some exercises the user can’t do, offer an alternative based on what the user can do and what equipment they have access to). I didn’t see it in the app.

Also I know that everyone does it, but the “free trial but we’ll charge you even if you don’t use the app and just forgot to cancel” is something that I really dislike. Maybe it’s an AppStore limitation but I rather do a trial and only pay if I actually use it, not if I forgot to cancel.

Lastly, maybe it’s after the paywall, but not everyone knows what all the routine names mean, if I have to google it, it’s something I expect the app to include. (A diagram / video of how to do it the right way)

Other than that, looks exciting and a really nice take on this topic.

Re:Show HN: Weight Lifting Algorithm Based on 5M Workouts by jakemor on Thursday April 11, 2019 @11:19AM (#19635412)
Hey! Message me in app for a free month :)

You can modify the routine but not based on available equipment and we have full step by step instructions along with videos for each exercise

Regarding the subscription thing — that's an app store limitation