NitpickLawyer

6 ms·

91.
by NitpickLawyer 2mo ago
> But the real question is: Is it okay to steal By definition it cannot be stealing since you're paying for the tokens. It may be against their ToS, depending on what you end up doing with those tokens, but it cannot be stealing. If
92.
by NitpickLawyer 2mo ago
This links to the nvfp4 version of the model, so they only compare it to the bf16 in this model card. If you're looking for other similar model comparisons they are in the model card of the bf16 version here [1]. While it looks "b
93.
by NitpickLawyer 2mo ago
How are you serving them? This is most often caused by an incorrect template. (the thing that tells the inference engine how to parse the think/tool parts of the answer) For vllm the official recipe [1] (for another model in the same f
94.
by NitpickLawyer 2mo ago
It would be fascinating to take a snapshot of reviews (say on goodreads) now, and then one later after the watermarks are checked. I bet there will be a lot of people that "just hate llm generated fiction" that have given highly p
95.
by NitpickLawyer 2mo ago
> you'll get the best results with agents from the same tribe, others will trip over the mountain of idiosyncratic metaphors. That's actually not what people found in practice. There's some research from the folks making s
96.
by NitpickLawyer 2mo ago
> because it's anthropomorphizing a model. Is it though? There's a perfectly "technical" reason why this strategy should work, without any sort of anthropomorphising: Assume models are trained on vast amounts of data.
97.
by NitpickLawyer 2mo ago
Linking to a github repo for a binary release (no source code related to the agent that I could see) is a bit iffy IMO. You should clarify your intentions or link to something else. Might confuse folks.
98.
by NitpickLawyer 2mo ago
> What does that even mean? This article is the perfect exemplification of that Good Will Hunting quote: "So if I asked you about art you’d probably give me the skinny on every art book ever written. Michelangelo? You know a lot abo
99.
by NitpickLawyer 2mo ago
> My concern is what a misaligned model will do when they’re even more competent. I think the alignment talk is a red herring. It won't matter in the end, because there will be (if there aren't already) efforts to train offensi
100.
by NitpickLawyer 2mo ago
They do address some of these things in the final slides / "lessons learned" section of the defcon talk. Good security practices will continue to be good, but... and there are a lot of buts here. I disagree with your take tha
101.
by NitpickLawyer 2mo ago
It's not unexpected. Current model gains are mainly from RLing a pretrained model on lots and lots of scenarios. They have the models run scenarios, and RL on successful runs.
102.
by NitpickLawyer 2mo ago
First via ssrf -> groovy plugin and second via specially crafted package I believe.
103.
by NitpickLawyer 2mo ago
There's more details for the HF related incident in this [1] defcon talk. They said they're gonna have a proper post mortem at the end of the investigation (interesting to see if they publish full logs for the hf thing or not). Bu
104.
by NitpickLawyer 2mo ago
> detect extremely tiny gravitational waves via an array of hyper-accurate clocks. I wonder how tiny is feasible. Would it get to a level of accuracy where you could detect neutrinos? We know they have some (v tiny) mass, so it should in
105.
by NitpickLawyer 2mo ago
> I wonder if it will ever be released in that form now. Yeah, I had the same thought. The key thing for them was the price point at which they could deliver a ~30B model. I would buy one today if it was ~1000$ and could run whatever the
106.
by NitpickLawyer 2mo ago
Back in the kazaa and limewire days, you'd sometimes try to get a movie / episode from a series, wait hours / days for it to download, and when it was done you had a ~50/50 chance to actually watch what you wanted or an
107.
by NitpickLawyer 2mo ago
> lifting ideas from other engines and copy-washing them. You cannot copyright ideas. And that's a good thing. Remember the "one click checkout patent" saga? That is bad for everyone involved. We don't want that, as a
108.
by NitpickLawyer 2mo ago
1.2 is now on openrouter. You could try it there, they have hard limits per key.
109.
by NitpickLawyer 2mo ago
> already rules that output of an LLM can't be copyrighted Mind sharing such cases? I'm not aware of any so far. There's the one with images, but that's commonly miss-understood, that case was ruled on a technicality
110.
by NitpickLawyer 2mo ago
> trying to run it locally or with your own hardware is pointless. Serving local models has advantages other than price. If you work in restricted industries, or have a strong need to protect your IP, or if you just value privacy more th
111.
by NitpickLawyer 2mo ago
> nothing to do with socialism scaremongering What's "scaremongering" to you is "life" to GP. Besides missing their point, you're also saying "you held socialism wrong, we can make it work, if only if i
112.
by NitpickLawyer 2mo ago
If anything this shows how pangram is unreliable. Chances are, if someone is running an article generation thing, it'll be 100% generated. Writing 3 / 100 articles "by hand" doesn't make any sense.
113.
by NitpickLawyer 2mo ago
"Draw a pelican on a bicycle" is not a serious benchmark. "Draw an animation of this long ass scene from a movie, and only call me when everything works e2e" can be.
114.
by NitpickLawyer 2mo ago
Heh, yeah. At least we're currently a bit more prepared for a Seveneves scenario than we were say 15 years ago. Speaking of prediction powers, there's one that stands out and turned out to be quite prescient on a bunch of topics,
115.
by NitpickLawyer 2mo ago
> which is about as hard sci-fi as you can get. Interesting, I always thought that the modern interpretation is that hard sci-fi is "rooted in science" or close to it. The ansible (i.e. FTL communication) is, to our current kno
116.
by NitpickLawyer 2mo ago
> Is it the preferred way only because training is expensive? No, it's the preferred way because that's literally how you train it. Contrary to popular misconceptions, you don't "compile" data into weights. You
117.
by NitpickLawyer 2mo ago
Depending on how this data is being collected, a factor could be the rise of search / fetch / scrape capabilities, added in lots of places. Agentic this, google killer that, and so on. The easiest way to parse web content nowadays
118.
by NitpickLawyer 2mo ago
> Why did I need to know that? In a perfect world such a disclaimer would help anyone reading the output by setting expectations. We each have different experiences with this, and different levels of acceptance. Knowing that something is
119.
by NitpickLawyer 2mo ago
> We improve, we learn, we recalibrate our expectations based on what we've learned. That's not what people mean when they say "moving the goalposts". It means that people are adamant that something wasn't import
120.
by NitpickLawyer 2mo ago
(let's assume that)My 3dprinter is special. It has a bunch of values + an algorithm (i.e. a neural network) that takes input as tokens and outputs a printed object.

More ›