HNHacker News
TopNewBestAskShowJobs

mehulashah

782 karma · joined September 25, 2011

founder: Aryn, we're hiring. www.aryn.ai
submissionscomments
mehulashah··on Classifying all of the pdfs on the internet
Care to make it publicly available? Or is that not permitted on your dataset? Certainly, there’s a lot more PDFs out there than 8TB. I bet there’s a lot of redundancy in yours, but doesn’t dedup well because of all the images.
mehulashah··on CockroachDB license change
It seems a shame that to grow, companies are backing away from the vector that got them there: open source.

I agree that current cloud providers are gaining more benefit from open source than they're putting in. So, it seems logical that the main developers want to recapture some of that.

On the other hand, open source is supposed to help build a bigger pie. If the pie gets bigger faster (i.e. more people using CockroachDB) then is the recapture worth it?

It seems the smaller companies think so. But, I don't know of a solid analysis that shows this to be true.

mehulashah··on Launch HN: Trellis (YC W24) – AI-powered workflows for unstructured data
Congratulations on the launch! This is the right way to think about LLMs and document processing.
mehulashah··on Talent Search versus Talent Development (2019) [pdf]
It seems that I’m the minority in this thread, but Talent Development has always gotten me much further than search. It’s also the Moneyball approach. People are remarkable when they are put in the right situations in which they can learn. It sets up a win-win situation helping people grow and teams get stronger. It also builds much more loyalty.
mehulashah··on Why AI Infrastructure Startups Are Insanely Hard to Build
The title is true. But, the arguments don't hold water for me. 12 years ago, I started a big data company. It looked similar for big data companies when Cloudera raised almost $1B in 2014. Too many people building data warehouses, especially in the cloud. I exited. Who knew that Snowflake and Databricks would emerge against the incumbents. Similarly, there will be winners in the AI infrastructure space. To win, you need to focus on your customers and delight them. Narrowing focus makes a lot of sense. Don't pay attention to the doom and gloom, or you'll never do a startup.
mehulashah··on 6 months ago, I left the bullshit industrial complex
“Technology has potential when it's guided by ethical considerations and a genuine desire to improve the human condition.”

Yup, and Bill and Dave are probably the last two that I can really point at to say that’s what they had. True ethics and a genuine desire. Probably also the Fab 8. It may have stopped there.

“The tech industry has a long memory.”

Nope. I think it has no memory and zero penalty for failure. That’s, of course, a double-edged sword.

mehulashah··on One Million Checkboxes
I was excited by the colored boxes!
mehulashah··on Arvind has died
I just learned of the passing of Prof. Arvind. He was not only my professor, but also a warm human being who opened his home to a hungry and homesick undergraduate. I will never forget the Diwali at his home where he invited me and my friend among others from the MIT community. After the aarti, he told us stories of his graduate school days where he often longed and dreamt of pakoras from home. He then invited us to feast, and we ate well that night. Dear prof Arvind, in addition to your wisdom, it is your kindness that we will pass on to future generations.
mehulashah··on How the square root of 2 became a number
I never learned of these formal definitions in high school mathematics. Nor in the lower level college ones that I took. There’s a beauty to this perspective— irrational numbers are what rationals are not.
mehulashah··on Do not try to be the smartest in the room; try to be the kindest
I’m wondering if this is really about life and not just meetings.

Meetings are really about getting things done. The kindest act is to not have the meeting if it’s unnecessary. And if it is, outline what you want to get out of it at the start. During the meeting, yes, you should be kind.

mehulashah··on Self-Serve Dashboards
If this is true for BI (which I tend to agree with), then this even more true for ETL. End users have an even less of a chance of doing data wrangling. Rather, it’s the technical folks that’ll do it.

What we learned in building AWS Glue was that it wasn’t just about the context — it was also about escape valves. Escape valves are tools necessary to get out of a situation that wasn’t anticipated. When the answer doesn’t make sense, the technical users are the only ones that have the know how to debug it.

mehulashah··on σ-GPTs: A new approach to autoregressive models
Yes, HN automatically did that.
mehulashah··on 60 kHz (2022)
Well said. Simple, solid technology for the benefit of everyone.
mehulashah··on AI Overviews: About last week
The shift from extracting content from the web and showing it versus generating content is an inflection point in search. It’s hard to get right, and I don’t agree that people like it better than regular search. It seems so, but the reality is more nuanced since positive engagement vs. frustration are hard to distinguish at a distance.
mehulashah··on Sharing details on a recent incident impacting one of our customers
I’m completely baffled by Google’s “postmortem” myself. Not only is it obviously insufficient to anyone that has operated online services as you point out, but the conclusions are full of hubris. I.e. this was a one time incident, it won’t happen again, we’re very sorry, but we’re awesome and continue to be awesome. This doesn’t seem to help Google Cloud’s face-in-palm moment.
mehulashah··on Von Neumann's First Computer Program (1970)
Sorting has always been a bellwether for computing systems. Interestingly, Jim Gray recognized that 45 years later. He therefore initiated and evangelized the out of core Sort Benchmarks (sortbenchmark.org).
mehulashah··on The Vietnamese Computer Scientist Who Made Toy Story Possible
There are too many stories of true inventors not getting the credit or notoriety that they deserve. The things that our government did in Vietnam and WWII in Asia are shocking.
mehulashah··on AWS CEO Stepping Down
AWS has always been a place of change. Adam has been a good leader, and I’m excited to see how AWS traverses the future. I never would have thought that the AI landscape would have such a dramatic effect on the cloud. Clay Christensen insight in innovation curves still rules to this day.
mehulashah··on Jim Simons has died
I meant the financial-industrial complex that dictates our economy.
mehulashah··on Jim Simons has died
This person made a lot of money, so it’s easy to say that he’s part of the machine. But, the man had principles. And he stood by them. Grateful for him showing us the way.
mehulashah··on New open source AI model for document segmentation and unstructured ETL
Would this be able to incorporate the models from Surya —

https://github.com/VikParuchuri/surya

mehulashah··on Vulture shortage threatens Zoroastrian burial rites
It’s interesting how burial rituals change over time. Now, we’re talking about composting our bodies, which in some sense returns the remaining minerals back to where they came from in the earth.
mehulashah··on Wild orangutan seen healing his wound with a plant
No I don’t, and it wasn’t intended that way. It was the opposite— they’re magnificent. They’re clearly not of our species. That does not mean they should be caged.
mehulashah··on A Mathematical Theory of Communication [pdf]
When you read this and think about the world he was in — it’s even more remarkable. How did he come up with it?
mehulashah··on How to Make More Money Renting a GPU Than Nvidia Makes Selling It
There’s speculation that they could become one should it be a sustainable for them to do so. I don’t think it will be.
mehulashah··on Wild orangutan seen healing his wound with a plant
These are magnificent creatures. There is so much more left for us to learn about them and us. Phenomenal.
mehulashah··on BASIC turns 60
I learned to program with BASIC. My first program was one to balance my Dad's checkbook. I copied it from a manual and added some extra PRINT statements for flair. I was in 4th grade. It was on a Timex Sinclair. Then I tried to draw a smiley face and ran out of RAM -- it only had 1K.
mehulashah··on The Fermat's Last Theorem Project
I wonder if after all that work, we might automatically reduce the proof and discover a simpler one that could have been included in the “margin”.
mehulashah··on Run llama3 locally with 1M token context
Our results are varied on long contexts. True — one can put a lot of stuff in there and the thing groks a bunch of it. Being able to synthesize specific answers from “long tail” facts in the context can be difficult. YMMV.
mehulashah··on How to Write Computer Programs (2015) [pdf]
Declarative programming is quite elegant, but it is not the same as prompting. I recently had an argument with someone who claimed our declarative code for running analyses over documents was "hard coded". In our product, our approach is to use a functional language approach for analyzing documents. He was an expert in NLP and said that his users required the ability to instruct the system in natural language - English. It still was weird that someone call our declarative program - "hard coded". My, how times have changed.
← PreviousPage 4 of 6Next →