My HN username is also my gmail. I've got most of the stuff you mentioned, including unencrypted copies of US tax returns (with SSN) and house buying paperwork.
> I used to reach out and tell them I didn't sign up for their service. But honestly, after doing it for a few years I gave up.
Same here. It's surprising that most of the services don't use double-opt in before sending emails.
Some day, I want to use an LLM to identify those emails and label them.
The deal isn’t always about the price. For example, a $1M house bought with $100K down and $900K mortgage is a worse deal for the seller as compared to $500K down and $500K financed. Assumption here is that it is more likely to get a $500K loan irrespective of the appraised value of the house.
A lower all cash offer (say $975K) is likely a better offer for the seller because it reduces the risk for them and closes the transaction much quicker than a mortgage transaction.
I have been a buyer in two transactions where my offer was slightly lower than the highest bidder, but with better terms.
HTML emails were there before, but I agree that fast UI and threads was a step up from Yahoo or Hotmail. Labels instead of folders, username+wildcard@ email addresses, filters (at least the UX was better than traditional rules). They also enabled POP and IMAP. IIRC, IMAP used to be a premium feature for email providers at that time.
I’m not sure about revolution, but e-mail can certainly be evolved.
20 years ago, not as many emails were sent, especially transactional emails. User behavior has evolved since then.
For example, storing email receipt of a random Amazon order from 2 years ago doesn’t make much sense these days.
Hey has addressed some of these changes, but there is a lot of room for improvement.
To be fair, if the GPT creator doesn’t do a good job of explaining the GPT, you can’t really tell even if you pay $20. The only way is to try out the GPT and hope you hit the right features. Chat is a pretty bad interface to discover features.
I end up creating multiple threads “frontend”, “backend”, etc. but a GPT for the whole project is a great idea!
How do you keep the GPT updated so that it knows about the final decision made for a specific problem. Like if api schema changes or the db is moved from SQLite to Postgres.
Agreed, but Gen AI already has a lot of use cases, the challenges are with “taming” the technology - Rare hallucinations, RAG (or equivalent) figured out, more deterministic.
Assuming some of these challenges are solved, integrations will follow.
Imagine the web before CSS, JS, AJAX. That’s where we are with Gen AI.
~13B models should work well with plenty of room for other applications. Lately, I’ve heard good things about Solar10B, but new models come in a dozen by day, so it might have already been changed.
I have usually seen people recommend to chunk by sentences or paragraphs or some fixed length of characters. IMO, all these are suggested because they are easy to write code for, but in reality, length of a meaningful chunk depends entirely on the data. The way we chunk an FAQ document vs a PRD is different.
Based on this assumption, I have a couple of questions:
1. Is chunking the most significant factor in RAG quality?
2. If there are no limitations, would humans that are experts in that dataset, be the best people to create chunks?
> this would only serve the niche of "not-registered for GST and using cash"
Wouldn’t this segment have an incentive to not digitize their records as they’re evading tax?
This was the case several years ago, not sure if still relevant anymore. People evading tax were only interested in non-cloud solutions for their records.
Congratulations on the launch! I’d love to see something like this succeed. However, there are some challenges that you might have to overcome. Here are some random thoughts:
Ingest Client Data - You will have to find customers who ingest dynamic data schemas. The example in the video shows more of a standard schema, which can be mapped once (using Lume or otherwise). No need to add an overhead or extra cost to run that data pipeline.
Normalize Data - One of the challenges will be to establish quality metrics for these mappings. Based on the demo, the quality score is supposed to be 100% all the time, but that’s far from the truth. Real data is messy. Validations will catch a lot of issues, but there will still be cases where incorrect mappings slip through. Ability to provide metrics around this will be very helpful in adoption.
Response Time - I’m not sure if you’re using OAI or your own models in the background. Even small latency in the pipeline for 100s of millions of records adds up to hours/days delay.
I wonder why they launched it today, when it was supposed to be a week-off for all OAI employees and amidst whatever else is going on?
One possibility is to capture Thanksgiving gatherings as a growth hack where people demo this to their families/friends and increase app downloads for OAI.