karterk

Where is this, if I may ask?

> The moat of the frontier folks is just compute.

This is not really true. Google has all the compute but in many dimensions they lag behind GPT-5 class (catching up, but it has not been a given).

Amazon itself did try to train a model (so did Meta) and had limited success.

Satya saved Microsoft by doubling down on Azure and Cloud. Something that Balmer failed to do with Mobile.

There are many different classes of problems that are affected by tokenization. Some of them can be tackled by code.

Solving the strawberry problem will probably require a model that just works with bytes of text. There have been a few attempts at building this [1] but it just does not work as well as models that consume pre-tokenized strings.

[1]: https://arxiv.org/abs/2106.12672

It's interesting how all focus is now primarily on decoder-only next-token-prediction models. Encoders (BERT, encoder of T5) are still useful for generating embedding for tasks like retrieval or classification. While there is a lot of work on fine-tuning BERT and T5 for such tasks, it would be nice to see more research on better pre-training architectures for embedding use cases.

> Cows are treated very well in all the ranches I've been on.

Unfortunately the entire premise of your argument is based on a personal anecdote that's a gross generalization of vast number of cattle farms.

For JIT-ing you need to know the sizes upfront. There was an experimental branch for introducing jagged tensors, but as far as I know, it has been abandoned.

Can you give some examples of problematic C idioms in C++?

Every billion dollar idea is a million dollar idea.

Every million dollar idea is a 100K dollar idea.

There are always smaller segments within large segments that you can go after.

Yes, you can get a waiver if you apply for something called the "Letter of Undertaking" which allows you to not pay anything for exports. This was not available at the start.

No, they usually are very brazen about that, but this time they are generally clueless.

Back when GST was introduced, they asked all exporters of software services to pay the 18% GST and then claim a refund because exports are exempted. This was before they introduced a process to apply for a "NOC" (which must be renewed periodically I think). So, I duly paid GST for the first couple of months.

5 years later, I'm yet to receive the refunds. First, they said that they were authorized to only sanction refunds for exporters of physical goods and then they said that this is handled by the State government, who said that it was handled by the Central. Finally Central accepted that they are in-charge of the refund, but by then COVID happened and I could not follow up for 2 years. In May, when I tried to claim the refund again, they said that this refund pertains to 2016 and it's difficult to "reopen" the accounts for that year and that's handled by yet another department. So, they have written to them to ask for the process. No replies after two emails. Story continues...

Here's the 2022 edition: https://twitter.com/mrohanshah/status/1579829414916784128

What are some ways in which they are bad/broken?

Yes, we plan to do that, but we will start off with first supporting raw vector data type and search on that.

To offer a counter view: paper ballots were a huge source of fraud in India's elections until electronic voting machines were introduced. I don't wish to ever go back.

Have you looked at TinySeed? But they don't operate at a Series A scale in terms of cash deployed per investment. I suspect that the power law of returns in VC model requires really large outsized returns that cannot be achieved Without an all or nothing model.

Diversify initially to protect the downside but double down on your winner(s) to scale the upside!

There are many large arid, desert-like land available in almost every part of the world which can be put to use for Solar. I don't think cost of land is a huge issue.

Sure. At earlier jobs, I often got paged because of some bot running wild in crawling our web pages. Some pages are heavy and we don't expect them to be hit often, but once you crawl these pages indiscriminately (even if accidentally) that can bring a site down. There are also some pages whose underlying resources are billed in a pay-as-you-use model. Once again, heavy bot traffic ran up our bills.

Robots.txt allows the site owners to restrict such pages from being crawled by bots. Services that allow people to circumvent the restrictions are being rude to say the least. Many crawling services also use a farm of proxies that spoof their real identity with fake user agents to circumvent rate limiting etc. All of these "strategies" go far beyond basic automation and is quite shady in reality.

Do you follow robots.txt or do you allow your customers to bypass restrictions placed by the website? How do you feel about bypassing that? I know it's not illegal but it's certainly not ethical either.

> My app is officially dead.

Most markets are large enough to accommodate several businesses. There's really no reason to stop working on something just because you find that there is already someone solving that same problem. Who knows, they will shut it down at some point, or might head upmarket or pivot to a tangential area.

Any algorithm is going to get gamed sooner or later. The biggest problem I've with Google ranking is this:

1. Everyone now knows that Google favors long content pieces which covers the topic in depth (what, why, when, etc.)

2. So an army of content marketing firms are writing 2000 word posts for simple topics that can be covered in 200 words.

3. As the user gets lost in the 2000 word article, trying to find what they really need, Google treats this as a positive "lots of time spent on page" signal and rewards this behavior further.

The result is people trying to now write 3000 word articles to "one up" the other already long posts dominating the first few results.

My co-founder recently wrote about our deliberations on fund raising as an open source company here: https://typesense.org/blog/why-we-are-not-raising-funds/

Ultimately, there is no golden rule. One has to choose what's best for your customers and your stage of growth.

At Typesense [1] we face a similar problem with Slack. Slack has been great for us in terms of quickly discussing an issue but the free plan limits the searches to last 10K messages. We would be happy to pay for it but the current pricing model is just not catered towards large communities where users interact in periodic bursts. Most communities will not be able to afford it.

I've been tempted to index our Slack archive and make it searchable (afterall we are a search company) but I'm afraid that it's against Slack's ToS.

If someone from Slack is reading this, please fix this: I will be happy to pay Slack on a different plan which considers the use of Slack as a collaborative platform for an OSS community.

1: https://github.com/typesense/typesense

This is Kishore, co-founder of Typesense. We will be publishing an ARM64 build from the next release. And I hear you about the auto schema: working on that as well.

They don't respect robots.txt or have a way for websites to opt out. These might not be illegal but mighty rude.

Does Japan have a concept of suburbs? Are houses larger on the outskirts of Tokyo?

Do you have to do anything special to opt-in for the built-in batch inserts? Earlier, I was forced to use the buffer tables approach: how would one ditch that now?