Is the main point “let me mathematically prove that it’s impossible to do what I’ve already done 12 times this year?”
Yes, very long workflows with no checks in between will have high error rates. This is true of human workflows too (which also have <100% accuracy at each step). Workflows rarely have this many steps in practice and you can add review points to combat the problem (as evidenced by the author building 12 of these things and not running into this problem)
This looks very cool, will try it out on my next project.
There have been a number of solutions popping up to address this problem, and I think the need is very real. Decomposing these LLM tasks into subtasks seems to be one of the best ways to work around the shortcomings of LLMs in production apps (hallucinations, context window limits, etc). But then you end up with complicated pipelines that are difficult to debug, improve, reason about, etc.
Indeed - decomposition improves reliability but also makes the testing more challenging. That’s why we made the framework modular! Let us know of any feedback as you try it out!
He’s a couple days unemployed and is already producing code that looks like it’s meant to teach people. Very happy to see that! The world will be a better place with more Karpathy educational content.
> Sellers have the right to refuse your exemption certificate, even if it is correct and properly completed. A seller that refuses your certificate must charge you sales tax. You may apply for a refund of the sales tax using Form AU-11, Application for Credit or Refund of Sales or Use Tax. For more information, see Tax Bulletins How to Apply for a Refund of Sales and Use Tax (TB-ST-350) and Sales Tax Credits (TB-ST-810).
> Apple Store purchases will include sales tax based on the ship-to location and the sales tax rate in effect at the time of shipping unless proof of state tax-exemption is provided. To assure that your order is tax exempt, please call Apple Store Sales Support at 1-800-793-9378. Be prepared to provide proof of state tax-exempt status for the state where the product is shipping. The organization or individual name on the order must exactly match the state tax-exempt certificate.
I'm very happy to see that someone is working on this problem!
At my last startup we had a lot of pain points around financial modeling, and it felt like 80% of the work we were doing in our models was the exact same thing that every other startup was doing, all so that we could model the last 20% that was truly unique to our business. It was also very hard to keep these models up to date, test out multiple scenarios, share models with investors without them immediately going stale, etc. I think there's also a great opportunity for you to look at the data from different startups in order to help give guidance re: best practices (e.g., "80% of startups in your category have a lower conversion rate than you - consider adjusting your expectations").
I've actually considered working on this problem myself, but TBH I have some concerns about the business model (e.g., I think it's going to be challenging to get startups to pay a significant amount of money for this on a recurring basis). I'm happy to see that someone is working on it though, and I'll be rooting for you guys!
You're reading our minds; that is exactly where we want to take Finmark. As far as the last 20%, we think we've built our app in a way that is modular enough to get about 95% of the customization that you would have in excel. Over time we hope to to eliminate that last 5% altogether.
There's a reason these scammers choose gift cards as their preferred way to get your money: gift cards have become the easiest way to move large amounts of money while minimizing the chances of getting caught.
Every other business that deals with the transfer of money (banks, for example) is expected to put strong measures in place to make sure that the money is traceable and that they're not helping criminals. Why shouldn't large companies like Target be held to a similar standard?
Moreover, the incentives here are very worrying. Yes this guy got scammed out of $3k, but Target also got $3k in revenue. Without some sort of penalty/punishment, Target is incentivized to do the bare minimum to minimize their PR exposure (which they can do by saying "we take this very seriously and have adopted some new policies" but not really acting on it) but to otherwise allow this to keep happening. As the prevalence of these gift card scams has skyrocketed, so have gift card sales for these large companies.
There's also an interesting sub-class of this scam that targets employees of certain size companies. Due to anti-bribery laws/rules/regulations gifts of cash to other companies are usually forbidden (and every employee with an expenditures line should generally be very aware of that), but non-cash redeemable gift cards are often still seen as fair "gifts" for various types of incentives/thanks/congratulations efforts. So gift card scams are sometimes even more effective for corporate employees, because corporations already abuse gift cards as a grease currency, and some corporate employees are very used to buying strange amounts of gift cards and giving them out in strange ways.
From what I understand, a retail store doesn’t make any profit selling gift cards for other stores.
The Safeway in this story didn’t make any money off of that $2k purchase of Target cards and it could’ve saved/boosted its reputation if it had some better protective measures in place. It also means that maybe people who would’ve spent money at Safeway now can’t because they’ve lost several thousand dollars.
> From what I understand, a retail store doesn’t make any profit selling gift cards for other stores.
That seems dubious, considering they dedicate precious shelf and display space to selling them. Accounting can get crazy in the real world, but clearly they have a motive to move these things. The cards are on the shelf because the cards make money for the store, somehow.
I know that I have a lot of gift cards that go unused for months, or never used at all. Probably a lot of profit in interest on holding all the money and keeping anything not claimed after a few years.
It might be true that they make no money from the sale itself, but those displays are likely from a third party that is paying for the shelf space.
If you do a search for something like "gift card distribution", there seems to be a bunch of marketing firms that you can pay to get gift cards made and put in stores.
TapFwd empowers marketers to access and share data with complete control. With TapFwd, marketers form private data sharing alliances with complementary brands to unlock entirely new, mutually beneficial ways to acquire customers.
To learn more about who we are, our engineering culture, and whether this is the right place for you, read our Key Values profile: https://www.keyvalues.com/tapfwd
TapFwd empowers marketers to access and share data with complete control. With TapFwd, marketers form private data sharing alliances with complementary brands to unlock entirely new, mutually beneficial ways to acquire customers.
To learn more about who we are, our engineering culture, and whether this is the right place for you, read our Key Values profile: https://www.keyvalues.com/tapfwd
TapFwd empowers leading marketers to access and share data with complete control. With TapFwd, marketers form private data sharing alliances with complementary brands to unlock entirely new, mutually beneficial ways to acquire customers.
To learn more about who we are, our engineering culture, and whether this is the right place for you, read our Key Values profile: https://www.keyvalues.com/tapfwd
Yes, very long workflows with no checks in between will have high error rates. This is true of human workflows too (which also have <100% accuracy at each step). Workflows rarely have this many steps in practice and you can add review points to combat the problem (as evidenced by the author building 12 of these things and not running into this problem)