HNHacker News
TopNewBestAskShowJobs

trolan

82 karma · joined June 7, 2021

submissionscomments
trolan··on Custom instructions for ChatGPT
obtained their new system prompt:

>Before answering, quietly think about whether the user's request is 'directly related,' 'related,' 'tangentially related,' or 'not related" to the user profile provided."

This is secret sauce? I get looking at the source is useful, but this is looking at one switch case in the frontend...

trolan··on Custom instructions for ChatGPT
Can anyone tell me a reason why either 'hacking' a prompt, leaking it or trying to keep your prompts hidden has any kind of value?

All I see is you found a way to get it to talk back to you when it was told not to, which a toddler does as well for the same value.

I can't imagine any, or any meaningful amount, of the secret sauce being in the words in the prompt.

trolan··on Something in space has been lighting up every 20 minutes since 1988
The article covers that. The only known versions do not fit the physics, time scale or amount of energy observed.
trolan··on Llama 2
A properly instructed hippo would stay out of the deep end
trolan··on Discord is not documentation
Help me with my education. I'm just a lowly 4th year SWE major and we haven't covered definitions of servers yet.

Webster's says: > a computer in a network that is used to provide services (such as access to files or shared peripherals or the routing of email) to other computers in the network

And if a computer is: > a programmable usually electronic device that can store, retrieve, and process data

How is it not a server? It is a computer, it is used to provide services, it's programmable and electronic, routes content... Do you have to assume each discord instance is running in its own container for you to consider it a server? It serves things.

trolan··on Discord is not documentation
> a computer or computer program which manages access to a centralized resource or service in a network.

I think by definition you're wrong. By layman's terms you're wrong too. A server serves things, either content, chat text or your meal. Discord servers serve a collection of text channels located on a centralized resource.

Now, for the real lesson: take yourself and everyone else less seriously.

trolan··on NotebookLM: An AI Notebook
I think a Language Model being used to summarize text with specific direction is one of the primary use cases. I don't really see too many gimmicks here, besides the obvious Google issues
trolan··on PdfGptIndexer: Indexing and searching PDF text data using GPT-2 and FAISS
Yes, when new technology comes you may need to upgrade. OpenAI aren't hero's but they are covering the cost to move people from old to new embedding models
trolan··on PdfGptIndexer: Indexing and searching PDF text data using GPT-2 and FAISS
Ah, I see what the top comment was implying. I was a bit short sighted on that side. Yes, you'd be tied to OpenAI for any new queries you need to generate. There could be some ways to offload that (vector of a vector) but it is a cloud dependency. I'd argue not a cost dependency based on how cheap these are.
trolan··on PdfGptIndexer: Indexing and searching PDF text data using GPT-2 and FAISS
No, I have anecdotal evidence using it. How does it compare in your usage of OpenAI and competitors?
trolan··on PdfGptIndexer: Indexing and searching PDF text data using GPT-2 and FAISS
The only OpenAI 'crap' being used here is to generate the embeddings. Right now, OpenAI has some of the best and cheapest embeddings possible, especially for personal projects.

Once the vectors are created tho, you're completely off the cloud if you so choose.

You can always swap out the embedding generator too, because LangChain abstracts that for your exact gripes.

Everything else is already using huggingface here and can be swapped out for any other model besides GPT2 which supports the prompts.

trolan··on Langchain Is Pointless
I actually found a lot of value when I left the LangChain ecosystem and started using jinja templates. The syntax from home assistant moved over and I can just pass dictionaries to render a prompt now.
trolan··on GPT-4 API General Availability
I signed up under my .edu to use the $18 credit for a school project and the phone # was all it took to know I was the same person.
trolan··on My Kind of REPL
Have you found any added benefit in your approach using mutation testing?

Really interesting stuff I'm going to play with personally and in school I think. Thank you.

trolan··on With plugins, GPT-4 posts GitHub issue without being instructed to
It's no more victim blaming than if someone ignored every safety regulation and got themselves hurt or killed. You are being told these things are beta, can produce incorrect results and then in every Convo OP is feeding an option for 'Do things with GitHub' context to the model.

Of course it's going to bias to using GitHub especially after he already used one plugin

trolan··on With plugins, GPT-4 posts GitHub issue without being instructed to
With their latest functions update to the API it seems the logic is simply passing methods to call the API via JSON to a trained model and it predicts which function is best to call for the info, then OpenAI does the call for ChatGPT, then feeds the resulting Json back to ChatGPT which uses it to give an answer.
trolan··on gpt-author: Generate an original fantasy novel using GPT-4 and Stable Diffusion
Behind the Bastards did a great 2 parter on the dangers of these books and being marketed to children. Basically, because they lack substance and their storytelling is not consistent, they argued there could be a lack of development if children get these books from overworked parents trying to keep a controller or screen out of their hands.
trolan··on The Kia Boys will steal your car for clout
This reminds me of 'Chicago Punch' theft of GM vehicles in the 90s with Saginaw steering columns. My father owned an impound lot and there were dozens of these a month coming through, so much so we had a few screwdrivers stuck in the fence as keys whenever we needed to pull one up. I imagine if we started putting those kind of exploits on AIM and got attention like TikTok does we would have seen these kinds of numbers 3 decades ago.
trolan··on San Diego fireworks glitch (2012) [video]
Big Bay Boom! Every year they have 4 barges in the bay synchronized to music, so you can watch from all over the bay. Only show I've seen better is in Norfolk.
trolan··on Ask HN: Who is hiring? (July 2023)
Would this be a position a software engineering student could navigate in US timezones? Maybe part time during the school year/full time summer and beyond graduation?

Relevant project: https://github.com/Trolann/ParkingSucks/

trolan··on Quart is an async Python web microframework
I had a project which was making API calls that could take 90 seconds to generate a response from external services. My flask API was obviously blocked and when I started letting other users on I fixed the bottleneck by updating (I think) 3-5 lines of code and adding an asyncio loop.

From my limited use it does seem like a drop in replacement for 'async flask'

trolan··on Vector support in PostgreSQL services to power AI-enabled applications
> - It seems like these embeddings are learned during the training phase. Presumably by backpropagating to the embedding during each epoch?

As far as I understand it, your first part is accurate.

> - Given that they are learned, doesn't that mean that they are completely context specific to a given trained model? Ie, they can't readily be shared on their own?

Yes, but you only have to use that context model in order to generate new embeddings to add or search the existing store. You can then use whatever is found wherever you want. Think of the vector store as a searchable database, where the vector is unique to the exact text embedding (like a hash function I believe, but I think without the possibility of collisions). You can search and compare vectors, then use the vector to determine what text was embedded (using simple relational queries or storing the text in the same row as the vector).

It means you can store chunks of information as embeddings in a vector database, search for similar content to a query and get chunks of text back related to what you searched.

> - What does "similar" mean here? Is there some emerging practice on how to determine how close is close "enough" between multiple vectors for the purpose of similarity searches? Is this too determined by the model weights somehow?

Cosine similarity search is the most used recently because it's what OpenAI recommends. There's also dot product. They are doing calculations to find the closest other vector(s) across the dimensions of the embedding vector. A 2D plane with a vector on it is 2 dimensions. Some models embed 700-800 different dimensions. OpenAI uses 1536 dimensions.

Imagine comparing a dozen 2D vectors to each other to find the three with the closest angle to a chosen other one. Somewhat easy to move them and overlay, and if they are pointed in the same direction they are likely related. Doing so in 1536 dimensions can't be imagined, but there is still an angle between the vectors with a dot product. It's this mathematical similarity which is used.

> I hope I'm not missing something fundamental with these questions. Or maybe I hope I am missing something and someone points out my errors. That's good too. :)

The use of the Postgres vector store is more from the use after an embedding model has been trained and there is a use for the data being embedded, like helping customers shop for similar items based on their search, and returning descriptions; or feeding a user question in to find context to feed to an LLM for few shot prompting.

← PreviousPage 2 of 2