> they are very strongly tuned to value code that has no errors which can be picked up by linters or similar, while humans work the opposite way
This is a really valuable insight! It resonates well with my own experience revewing AI-authored code: I look hard at interfaces, architecture and performance, and merely glance over code that just "gets the job done", because if it works, I'm not worried about it. Now I can express that dichotomy much more clearly -- thank you!
That is my experience as well. I get a (subjective) speedup between 1 and 3 for parts of the code I'd consider critical and where I check the output tightly, and 5-20 for menial work OR for important code that's well isolated into its own module such that its quality doesn't matter because I can have it rewritten easily if it doesn't work as expected.
A bit off-topic, but it is interesting that you say this.
Just earlier today I listened to Startups for the Rest of Us podcast episode 842 [1] titled "The future of SaaS in an AI World", where host Rob Walling addresses the "SaaS is dead" question. One of his arguments is that it's rarely worth spending time on rebuilding your SaaS vendors' solutions, taking time away from your core mission. But I guess that:
1. the projects you hinted at are low-value services (e.g. CRUD web apps over DBs), and not SaaS services which have real moats, like CRM (connectors ecosystem) or email service providers (IP reputation, IT security).
2. His argument applies to startups, where you're trying to grow your valuation by growing revenue (valuation = 5-10x yearly revenue) as opposed to established corporations which usually have a higher incentive to limit costs.
You seem like a very smart person, so I'd be curious to hear your thoughts here. Were these low-value SaaS services, and is this in the context of an established business?
I agree in sentiment, I love knowing that another human made this, either because I fancy I could maybe do something as good as that, or because I just admire the talent, or simply because the lyrics or music touch me somehow.
That said, there are a lot of people who simply enjoy having something playing in the background, it doesn't matter what, and if you're into country music it's great to have 10,000+ hours of country music to play.
If Tidal provides a checkbox so you can choose whether to exclude AI content, I think that would work for both audiences.
Maybe if enough AI produces self-report their work as AI, and enough non-AI producers are honest about uploading non-AI work, they'll quickly have the necessary amount of good-enough data to train good classifiers?
> You want as much data as you can get about your health
The device looks very cool, but I strongly disagree with the premise, and think this statement is rather misguided.
1. Most people who feel unhealthy don't do so because of a lack of data but because of bad habits around meals, exercise, sleep, social interactions etc.
2. If you measure and scan all the time, every blip above or below the normal curve will start generating anxiety. One of the most frequent pieces of advice for people waking up in the middle of the night is to not look at the clock. Information can be stressful.
It's not about being against technology. It's that we know the simple rules that will keep us healthy most of the time, and they don't need any technology at all. Eat healthy meals. Exercise. Get enough sleep. Get enough rest. Don't smoke. Don't drink alcohol. Don't do drugs. Spend enough time with people. Serve others. Spend time outside.
Midjourney Medical looks amazingly cool. But it, and megabytes of data, is not what we really need.
We only learned how to do a lot of those things because we gathered and analyzed data to find out how best to do them! "Eat healthy meals" means something completely different today than 10, 20, or 50 years ago. How you exercise matters a lot. Smoking wasn't always considered unhealthy, nor was drinking. These "simple rules" and what they actually mean have been refined quite a bit.
You figure out this stuff by gathering and analyzing data. Whether or not this specific implementation will result in more meaningful actionable steps, I guess we'll see.
Exactly, if they do reach their goal of deploying tens of thousands of these machines allowing everyone to scan their body frequently and cheaply then even if there are limitations to this technology it will still provide a massive amount of longitudinal data about the human body and the impacts of lifestyle, age, pharmaceutical effects, etc.
The person you replied to mentioned diet and exercise, that seems like an area that would benefit directly from this type of scan. Being able to track the effect on body composition in a highly accurate way where we know exactly how much muscle and fat are gained or lost and where that's happening could tell us a lot about not only the effect on the "average person" but for each individual. I'm sure there are many other less obvious things that could be tracked using this technology.
True, we conducted science, part of which involved gathering and analysing data, to refine those "simple rules", though I'd say they've been pretty stable for a long time.
Both our unhealthy habits, and the "simple rules" to keep us healthy, have been around for decades. Building devices that give us gigabytes of data won't change anything. Dr Peter Attia makes a compelling argument in his book "Outlive" that science, as it is structured now, has achieved miracles when it comes to injuries and infectious diseases, but has been more or less powerless, for entirely systemic reasons, to do anything about neurodegenerative and cardiovascular diseases or against cancer and diabetes. His book is well worth reading to understand his argument--but the gist of it is that those require lifestyle changes.
Anthropic, as much as I think they are the soundest of the AI labs out there, still has a massive incentive to push things out that aren't saftey-vetted to the level we expect. They are very willing to "move fast and leave holes", to paraphrase M.Z. Hell, they leaked their own source code!
This is a really valuable insight! It resonates well with my own experience revewing AI-authored code: I look hard at interfaces, architecture and performance, and merely glance over code that just "gets the job done", because if it works, I'm not worried about it. Now I can express that dichotomy much more clearly -- thank you!
reply