About a year ago, I watched this 3Blue1Brown LLM tutorial (https://www.youtube.com/watch?v=eMlx5fFNoYc&t=367s) on how a model’s self-attention mechanism is used to predict the next token in a sequence, and I was surprised by how little we know about what actually happens when processing the sentence “A fluffy blue creature roamed the verdant forest.”
A year later, the field of mechanistic interpretability has seen significant advancements, and we're now able to "decompose" models into interpretable circuits that help explain how LLMs produce predictions. Using the second iteration of an LLM "debugger" I've been working on, I compare the hypothetical representations used in the tutorial to the actual representations I see when extracting a circuit that describes the processing of this specific sentence. If you're into model interpretability, please take a look! https://peterlai.github.io/gpt-circuits/
Emburse (YC W16) | Software Engineer | Full Time | ONSITE preferred, REMOTE | https://www.emburse.com/
Emburse is a corporate credit card platform that generates virtual and physical credit cards with specific budget restrictions. We manage both traditional corporate card expenses like T&E as well as backend vendor payments.
Emburse is built using Python/Django and is deployed on Heroku for development and AWS for production. We're looking for someone to join us as one of the first few employees and who would be comfortable embracing sizable infrastructure responsibilities which include:
Emburse is a corporate credit card platform that generates virtual and physical credit cards with specific budget restrictions. We manage both traditional corporate card expenses like T&E as well as backend vendor payments.
Emburse is built using Python/Django and is deployed on Heroku for development and AWS for production. We're looking for someone to join us as one of the first few employees and who would be comfortable embracing sizable infrastructure responsibilities which include:
Emburse is a corporate credit card platform that generates virtual and physical credit cards with specific budget restrictions. We manage both traditional corporate card expenses like T&E as well as backend vendor payments.
Emburse is built using Python/Django and is deployed on Heroku for development and AWS for production. We're looking for someone to join us as one of the first few employees and who would be comfortable embracing sizable infrastructure responsibilities which include:
For some companies such as those in manufacturing, enhanced oversight and better spend control are necessary card program requirements. I agree that business credit would add significant value. Thanks for the suggestion.
Founder here - looking for feedback on the product. You can think of Emburse as being like AmEx with Expensify-like features built-in. If you're in the process of transitioning from having all spend on a shared corporate card to having cards for each department or employee, definitely take a look.
Minor item: FAQ says that pricing will be available by end of 2015, which has just passed.
As for product feedback, I've literally never understood the appeal of debit cards over credit cards. I get cash back, delayed payments/free float, better anti-fraud, and better purchase protections with a credit card. So, I wish you the best of luck, but I don't "get it" unless there's a largish market that can't get (or can't handle) credit cards.
For now, we're starting with a debit card product because they're easier to launch. Eventually, as the product gains more traction, it will be easier to launch a credit product with the advantages you're looking for.
Which that said, there's plenty of opportunity as a debit card product. We believe that no financial product right now really prioritizes spend control and bookkeeping accuracy in the way we do within our product.
Hi there, co-founder of Crocodoc here. Crocodoc will continue being the best platform for document viewing and collaboration. Box is investing heavily in our platform and will make sure we stay awesome. Any updates to the API going forward should be totally seamless, but if you run into any problems, just give me a ping (I'm Peter). I'll make sure the transition happens smoothly.
Hey guys - Peter here from Crocodoc. In the official release, pages on mobile devices will be lazy-loaded to prevent the pages from using too much memory like it does on this demo here.
A year later, the field of mechanistic interpretability has seen significant advancements, and we're now able to "decompose" models into interpretable circuits that help explain how LLMs produce predictions. Using the second iteration of an LLM "debugger" I've been working on, I compare the hypothetical representations used in the tutorial to the actual representations I see when extracting a circuit that describes the processing of this specific sentence. If you're into model interpretability, please take a look! https://peterlai.github.io/gpt-circuits/