I've been wanting to build something like this for myself, but partnering/integrating with banks seems to be the main difficulty. How do you solve this?
And which cards / banks does it support?
Also, what does the name mean? It might be a tad difficult to google, unfortunately, since I imagine that googling "lang" would come up with a lot of other results.
It doesn’t integrate with banks. You log as you spend. It’s a common question, but I think there are many reasons to keep it manual. It keeps you aware of what you have left. And automation won’t ever be perfect, so you must keep an eye on it and adjust things.
The app focuses just on your everyday spending. You don’t log bills and subscriptions. And it’s not about being exact. You can add the rough total of what you spent. It acknowledges that when you plan your spending, it’s really just a guess. And you’ll adjust the plan as you go.
What did you have in mind when you thought about building something like this?
The name means ’long’, and is pronounced similarly. Naming is of course hard. I’m hoping that it will be something you remember.
Because LLMs are based on the abstract ideas of neural nets from brains. Say what you wish, but some problems were completely unsolvable before we adopted this paradigm. On some level, we must've gotten some ideas close to the right ballpark.
Yeah it is to some degree. I tried to use it as much as possible, but there's always those annoying edge cases that makes me not trust the results and I have to check everything, and it ended up being faster just building some simple UI where I can easily classify the name myself.
Part of the problem is simply due to bad data from the websites. Just as an example - there's a 2-week contact lens called "Acuvue Oasys". And there's a completely different 1-day contact lens called "Acuvue Oasys 1-Day". Some sites have been bad at writing this properly, so both variants may be called "Acuvue Oasys" (or close to it), and the way to distinguish them is to look at the image to see which actual lens they mean, look at the price etc.
It's true that this could probably also be handled by AI, but in the end, classifying the lenses takes like 1-2% of the time it takes to make a scraper for a website so I found it was not worth trying to build a very good LLM classifier for this.
> It's true that this could probably also be handled by AI, but in the end, classifying the lenses takes like 1-2% of the time it takes to make a scraper for a website so I found it was not worth trying to build a very good LLM classifier for this.
This is true for technology in general (in addition to specifically for LLMs).
In my experience, the 80/20 rule comes into play in that MOST of the edge cases can be handled by a couple lines of code or a regex. There is then this asymptotic curve where each additional line(s) of code handle a rarer and rarer edge case.
And, of course, I always seem to end up on project where even a small, rare edge case has some huge negative impact if it gets hit so you have to keep adding defensive code and/or build a catch all bucket that alerts you to the issue without crashing the entire system etc.
What is this biased way of reading CIs? Are you paid by 3M?
Yes, it shows that with 5% probability the risk is only 1.18 - why choose to focus on those 5%? According to the study there is an equal chance that the relative risk is 12.7 times normal, why should we ignore that?
You really think pointing out how large a CI is, is "biased"?
Your reading of the CIs is off. It isn't telling you that there's a 5% chance that the real risk is 1.18. It's saying there's a 5% chance the real risk doesn't lie in the CI, and could be anywhere outside the CI, including 1 (no effect).
Here's the main point. Increase the significance level slightly, say to 97% (if this is a 95% CI, or 92% if this is a 90%, etc). Then a CI that large and that close to 1 at the left edge will turn into a CI that includes 1. The effect is not significant at all at that level. It wouldn't be saying the true effect is significant with some probability or whatever you seemed to be interpreting CIs to say. It would be saying there's no effect.
This is markedly different from the case of cigarettes, where the p-values are astronomically small. Since earlier comments seem to compare these chemical to cigarettes.
But who knows? I think the objective function is so vague that it can come up with basically anything. I would be super interested to see it actually running. I imagine someone could set up a Twitch stream with this - perhaps with other objectives - and it would probably get a large following
My suspicion in reading the abstract and intro is that this is a translation issue. Inference informed by:
1) All the authors have South Korean institutional affiliations
2) "Detection of human biofluids such as blood, tears, saliva, sweat, and urine is important for clinical analysis of various physiological patterns" (First sentence, seems to be missing a word)
3) "differentiate patients from the normal group with high sensitivity and specificity" (abstract, patients and normal group is an odd way of phrasing this)
That being said...it had never occurred to me that nueral networks might be a useful way of interpreting spectroscopy data, that is a really cool insight
Neural nets operating on mel sprectograms (spectrograms shifted to bias frequencies that are of interest to humans) have remarkable ability for audio classification and synthesis. It stands to reason it would not be difficult to adapt methods towards nearly anything which can be mapped to an image in a similar way. I don’t think its particularly novel in concept and kind of surprised it isn’t significantly further along.
I think we are at the technology level to really make medicine significantly more affordable and available, at the cost of many high paying doctor’s jobs. Same can be said for lawyers. These are powerful groups that will not take being automated lightly.
> The dataset was randomly divided into 70% training subset (1420 for prostate and 692 for pancreatic), 15% validation subset (304 and 148), and 15% test subset (304 and 148).
