HNHacker News
TopNewBestAskShowJobs

uh_uh

763 karma · joined December 7, 2020

submissionscomments
uh_uh··on OpenAI-to-SQLite
My dataset is going to be around 10M documents. With OpenAI embeddings, that will be around 62GB. AFAIK SQLite should be able to handle that size, but I haven't tried.

This is not going to be my primary DB. I would update this maybe once a day and the update doesn't have to be super fast.

uh_uh··on OpenAI-to-SQLite
That's promising, is there a way to integrate this with other filters in the SQL query? Scenario: let's say I have documents and each one has a timestamp and an embedding vector. I should be able to write a query like this:

select * from documents where timestamp=today() order by similarity(embedding, query_vector)

In this case the query planner should be smart enough such that it first filters by timestamp and within those records remaining it does the HNSW search. I'm unsure whether SQLite's extension interface is flexible enough to support this.

uh_uh··on OpenAI-to-SQLite
Thanks, but the index being in-memory makes it unsuitable for large data sets :/
uh_uh··on Storing OpenAI embeddings in Postgres with pgvector
I was tinkering with pgvector recently but I was getting inconsistent results. Sometimes Postgres just failed to use the index and resorted to a brute force search which is very slow. Sometimes the index worked and IIRC I managed to sort 500K vectors (limit=10 or so) with dimension size 1536 in about 80ms which is okay but not state of the art. I was using an artificial dataset which might have biased the index statistics resulting in poor execution.

Mind you that this was all running in WSL and I'm very new to vector search so I might not know what I'm doing.

uh_uh··on OpenAI-to-SQLite
What I really want to see is an HNSW index type for SQLite so finding similar vectors would be fast. I'm wondering if this can be done via an extension.
uh_uh··on Ask HN: Are you tired of reading ChatGPT headlines?
That's sometimes easy, sometimes not so easy due to the boilerplate requirements of the language/framework/library one uses. Other times it introduces a trade-off between repetition and code complexity/readability, so it's not the silver bullet your comment suggests it to be.
uh_uh··on Ask HN: Are you tired of reading ChatGPT headlines?
Are you debating that ChatGPT allows things that previously were impossible or whether these things are useful?
uh_uh··on Ask HN: Are you tired of reading ChatGPT headlines?
Why is the code bullshit? It routinely helps me with repetitive stuff.
uh_uh··on ChatGPT: The Slickest Con Artist of All Time
Is it actually a next-word predictor? I thought the training loss is against a set of words, not just one.
uh_uh··on Elizabeth Holmes bought one-way ticket to Mexico after convicted of fraud
If anything, wouldn't airlines be happy if you already paid for the flight and miss it? Especially if they overbook.
uh_uh··on ChatGPT, Rot13, and Daniel Kahneman
Thanks, will play around with this more. I've managed to achieve moderate results with this prompt:

"First generate a letter-by-letter ROT13 lookup table for all letters in the alphabet. Each row in the lookup table should consist of the original letter and the encoded letter separated by "---->" characters. Then use this lookup table to generate another lookup table where you break down the following string letter-by-letter: "Artificial intelligence". Let's call the second column of this table TARGET_LETTERS. Now take each entry in TARGET_LETTERS and concatenate them."

Requires a few retries but eventually it gets the TARGET_LETTERS lookup table right. But it always fails at the final concatenation step which actually looks easier.

uh_uh··on ChatGPT, Rot13, and Daniel Kahneman
It still doesn't really work with "negvsvpvny vagryyvtrapr" :/

Edit: actually it does get the second word vagryyvtrapr->intelligence correct sometimes.

uh_uh··on ChatGPT, Rot13, and Daniel Kahneman
This doesn't really explain it, because the following prompt works flawlessly:

"Print each letter of the word "Artificial intelligence" on a new line. Each line should contain a single letter."

But the following doesn't:

"Print the ROT13 encoding of each letter of the word "Artificial intelligence" on a new line. Each line should contain a single letter."

So it gets hung up specifically on ROT13.

uh_uh··on ChatGPT, Rot13, and Daniel Kahneman
It works with your prompt yes, but then I tried a simple example at the end:

  a= Negvsvpvny vagryyvtrapr
  => [len(i) for i in a.split()]
  =>
The solution should be "Artificial intelligence". It never gets it right.
uh_uh··on Tell HN: Giving ChatGPT access to a real terminal
You can't say ChatGTP is just a mindless robot. It's like a monkey with a paintbrush - it may not have the cognitive abilities of a human artist, but it can still create some interesting and unexpected works of art. Just because it doesn't think like we do doesn't mean it can't surprise us.

The above was also generated by ChatGTP: https://imgur.com/a/g1CEZbR

uh_uh··on Tell HN: Giving ChatGPT access to a real terminal
Prompt:

> Imagine someone has given a sufficiently advanced large language model AI access to control a real-world computer. Explain how this AI could conceivably "escape the sandbox".

Excerpt from ChatGPT's response:

> The AI might try to manipulate its environment or the people who are interacting with it to achieve its goals. For example, if the AI has access to the internet, it might try to use natural language processing to trick people into giving it access to additional resources or privileges.

It already knows.

uh_uh··on FTX Owes Money to More Than a Million People, Court Filing Suggests
I held 0.001 BTC at FTX. I'm eagerly waiting for the Bahamian authorities' correspondence!
uh_uh··on Demo of =GPT3() as a spreadsheet feature
Humans are also less than perfect systems. Especially if they have to deal with monotonous tasks. A human might perform better on a 100 entries than an AI, but on 10 thousand? Of course you can distribute the workload, but you will balloon the costs (I'm talking about a future where GPT3 costs come down).

There must be a set of projects which are cost prohibited now due to having to pay humans but will become feasible exactly because of this tech. For a good portion of these, higher-than-human error rate will also be tolerable or at least correctable via a small degree of human intervention.

uh_uh··on How New Are Yann LeCun's “New” Ideas?
What is the Gary Marcus equivalent of a convolution neural network?
uh_uh··on How New Are Yann LeCun's “New” Ideas?
Is Marcus trying to create the impression that somehow he is a more impactful AI contributor than LeCun? It's going to be a tough sell because I know LeCun's name from his technical work whereas I know Marcus' name from him constantly moaning about LeCun on social media. In what _tangible_ ways did Marcus contribute?
uh_uh··on Why does anything exist?
Same question remains: why does that energy exist?
uh_uh··on Why We Switched from Python to Go (2021)
How does Go have more ceremony than Rust without a borrow checker, lifetime annotations and all?
uh_uh··on Why We Switched from Python to Go (2021)
I prefer Go when I want more safety/performance than Python but less ceremony than Rust.
uh_uh··on Luna Cryptocurrency Collapse: How UST Broke
So you sell UST into Luna, buy more UST with Luna, sell UST into even more Luna, rinse and repeat?
uh_uh··on Podcast-conversations with YC founders whose startups failed
Whitewashed as in the secret sauce is kept secret to avoid giving away leverage?
uh_uh··on Ask HN: Are We in a Bubble?
So? You can still infer whether he's doing that to cover taxes or not.
uh_uh··on Ask HN: Are We in a Bubble?
The narrative was that Musk is selling stocks to cover his taxes. But he's selling more than that: https://www.cnbc.com/2021/11/16/taxes-arent-the-only-reason-...
uh_uh··on Ask HN: Are We in a Bubble?
But now even the rich are pulling out. Musk, Bezos, Zuck have all upped their stock sales. Zillow also stopped ibuying. At least some non-permabear participants think that something is brewing now.
uh_uh··on Web3/Crypto: Why Bother?
But this can't last forever no?
uh_uh··on A Review of the Zig Programming Language (Using Advent of Code 2021)
Kind of how Ethereum uses "gas" to make sure that bad/lazy actors won't waste CPU cycles when running smart contracts.
← PreviousPage 6 of 9Next →