ijk

20 ms·

121.
by ijk 1y ago
> If they had bought each book themselves would it be fair use? So this is only about the piracy? The earlier ruling covered exactly that question: - Anthropic downloaded many books (from LibGen and elsewhere). This piracy is what the cu
122.
by ijk 1y ago
Given that at this point the games that have been delisted include IGF award winners and art games that have been shown in museums, I think we're pretty far past pointing at any individual games as a reason to justify this.
123.
by ijk 1y ago
I want that in an editor. It's also a good way to check if your writing is too predictable or cliche. The perplexity calculation isn't difficult; just need to incorporate it into the editor interface.
124.
by ijk 1y ago
> AI is shorthand for “we have no clue how to solve this problem, but AI can do it for us.” I find this ironic, given that it is extremely difficult to have an AI solve a problem that you cannot, yourself, solve in some capacity.
125.
by ijk 1y ago
> 2. Remembering they are supposed to write tests and keep ALL of them green (just like our human juniors...) I think the core principle that everyone is forgetting is that your evaluation metric must be kept separate from your optimiz
126.
by ijk 1y ago
In the USA, some things that come to mind: * Legal changes. Web platforms have section 230 protection to host user content. For payment processing this might be something like the proposed Credit Card Competition Act (CCCA) that requires ba
127.
by ijk 1y ago
It sounds like you're proposing doing this operation on the tokens in the reasoning. While it would be interesting to know if allowing it to choose arbitrary tokens, the biggest issue is that there's quite a bit of evidence that t
128.
Group Behind Steam Censorship Policies Have Powerful Allies(vice.com)
21 points by ijk 1y ago|1 comments
129.
by ijk 1y ago
For Steam, it looks like Collective Shout, NCOSE, and Exodus Cry are behind this particular push. [1] [1] https://www.vice.com/en/article/group-behind-steam-censorshi...
130.
by ijk 1y ago
> Targeting them with what? > What could possibly hold enough leverage that Visa would jeopardize their sweet gig as an ideology-neutral, essential piece of American infrastructure siphoning 1-2% off of every dollar of consumer spendi
131.
by ijk 1y ago
In aggregate? Signs point to yes. For the general purpose SFT base models. We see some evidence even with RNNs vs Transformers. You're essentially finding a function that models language. Use the same optimization function, get a simil
132.
by ijk 1y ago
Other payment processors, mostly. So other credit card companies (e.g. JCB [1]), government run payment services like Pix in Brazil [2], theoretically crypto, etc. [1] as a random example: https://archive.kyivpost.com/techno
133.
by ijk 1y ago
One factor is the ongoing campaigns from number of moral crusading groups who lobby them to cut off payment processing for things they don't approve of. NCOSE has been working for decades on the project, and targeting credit card compa
134.
by ijk 1y ago
Public control over AI models is a distinct thing from everyone having access to an AI server (not that national AI would need a 1:1 ratio of servers to people, either). It's pretty obvious that the play right now is to lock down the A
135.
by ijk 1y ago
RealTalk has some interesting features that I wish there was a more complete writeup that explained it in detail. Like, you can write a script that talks to functionality that may or may not exist yet. Programming by moving pieces of paper
136.
by ijk 1y ago
The light level isn't an issue in practice: when I visited the actual installation during the day, the building was brightly lit with natural light and the projections were easily visible, to the point that I didn't think about it
137.
by ijk 1y ago
> Doing better than nothing is a really low hanging fruit. As long as you don't do damage - you do good. That second sentence is the dangerous one, no? It's very easy to do damage in a clinical therapy situation, and a lot of t
138.
by ijk 1y ago
Unlike current videogames, LLMs by default flips the relationship with the limits: the agent is completely open-ended by default, but the player has to play along to a certain extent. This makes the experience a lot closer to a tabletop gam
139.
by ijk 1y ago
What I want to know is if the flood of scraping everyone has been complaining about is coming from people trying to scrape for training or bots doing RAG search. I get that everyone wants data, but presumably the big players already scraped
140.
by ijk 1y ago
Are there any examples of that? All the documentation I saw seemed to be about building an MCP server, with very little about connecting an existing inference infrastructure to local functions.
141.
by ijk 1y ago
I wish the documentation was clearer on that point; I went looking through their site and didn't see any examples that weren't oversimplified REST API calls. I imagine they might have updated it since then, or I missed something.
142.
by ijk 1y ago
Near as I can tell it's supposed to make calling other people's tools easier. But I don't want to spin up an entire server to invoke a calculator. So far it seems to make building my own local tools harder, unless there&#
143.
by ijk 1y ago
In theory, couldn't you distill a non-infringing model from an infringing one? Just prompt it for continuations and give it a whack every time the output matches something in your dataset of copyrighted works. You'd need the copyr
144.
by ijk 1y ago
And even on open LLMs, GPU instability can cause non-determinism. For performance reasons, determinism is seldom guaranteed in LLMs in general.
145.
by ijk 1y ago
The math tokenization research is probably closest. GPT-2 tokenization was a demonstratable problem: https://www.beren.io/2023-02-04-Integer-tokenization-is-insa... (Prior HN discussion: https://news.ycombinator.
146.
by ijk 1y ago
You can view the tokenization for yourself: https://huggingface.co/spaces/Xenova/the-tokenizer-playgroun... [496, 675, 15717] is the GPT-4 representation of the tokens. In order to determine which letters the toke
147.
by ijk 1y ago
Well, which is easier: Count the number of Rs in this sequence: [496, 675, 15717] Count the number of 18s in this sequence: 19 20 18 1 23 2 5 18 18 25
148.
by ijk 1y ago
DSPy, notably, includes functionality for finetuning models. [1] [1] https://dspy.ai/tutorials/games/
149.
by ijk 1y ago
This might be obvious, but just to state it explicitly for everyone: you can freeze the weights of the existing layers if you want to train the new layers but want to leave the existing layers untouched.
150.
by ijk 1y ago
For LLM inference parallel GPUs is mostly fine (you take some performance hit but llama.cpp doesn't care what cards you use and other stuff handles 4 symmetric GPUs just fine). You get more problems when you're doing anything trai

More ›