Palmik

21 ms·

211.
by Palmik 2y ago
Except the "binary" is not really opaque, and can be "edited" in exactly the same way it was produced in the first place (continued pre-training / fine-tuning).
212.
by Palmik 2y ago
I inferred that from the other statement: > Since DeepSeek-V3 is worse than those US frontier models — let’s say by ~2x on the scaling curve, which I think is quite generous to DeepSeek-V3 He says that it's 2x worse. So if it has th
213.
by Palmik 2y ago
https://www.anthropic.com/research/constitutional-ai-harmles...
214.
by Palmik 2y ago
> All of this is to say that DeepSeek-V3 is not a unique breakthrough or something that fundamentally changes the economics of LLM’s; it’s an expected point on an ongoing cost reduction curve. What’s different this time is that the compa
215.
by Palmik 2y ago
Anthropic is, according to themselves, using RLAIF... which is basically using LLM as a judge / reward model. So maybe he means that the models they use for RLAIF are not (much?) more expensive than Sonnet 3.5 (e.g. previous Sonnet or
216.
by Palmik 2y ago
You are underselling or not understanding the breakthrough. They trained 600B model on 15T tokens for <$6/m. Regardless of the provenance of the tokens, this in itself is impressive. Not to mention post-training. Their novel GRPO te
217.
by Palmik 2y ago
When I use NVIDIA GPUs to train a model, I do not consider the R&D cost to develop all of those GPUs as part of my costs. When I use an API to generate some data, I do not consider the R&D cost to develop the API as part of my costs
218.
by Palmik 2y ago
I agree, (2) seems much less problematic since the AI outputs are not copyrightable and since OpenAI gives up ownership of the outputs. [1] So, if you really really care about ToS, then just never enter into a contract with OpenAI. Company
219.
by Palmik 2y ago
Posting OpenAI generated data on the internet is not breaking the ToS. This is how most OpenAI based businesses operate, after all [1] (e.g. various businesses to generate articles with AI, various chat businesses that let you share your ch
220.
by Palmik 2y ago
What's the problem? At most this might be a ToS violation, but it also seems easy to avoid that (if you care at all). DeepSeek does not even have to be a customer of OpenAI and thus not subject to their ToS. Company A pays OpenAI for t
221.
by Palmik 2y ago
Legal issues for who? Company A pays OpenAI for their API. They use the API to generate or augment a lot of data. They own the data. They post the data on the open Internet. Company B has the habit of scraping various pages on the Internet
222.
by Palmik 2y ago
I don't know where you're getting your information from. Maybe you're confusing DeepSeek v3/r1 and the distilled r1 models. DeepSeek V3/R1 architecture isn't anything like Llama 3. Llama 3 isn't even a mix
223.
by Palmik 2y ago
DeepSeek v3/r1 isn't based on llama architecture. It uniquely combines and contributes several novel approaches. Meta never released a mixture of expert model (they failed to train a good one, according to reliable rumors). And Mo
224.
by Palmik 2y ago
It's actually not wrong, here is the example from the OP: https://imgur.com/a/5uMYI44
225.
by Palmik 2y ago
Massive overreaction in my opinion (not financial advice). People are forgetting how much compute is being used for inference. This is going to be further accelerated: - Reasoning models are going to generate orders of magnitude more tokens
226.
by Palmik 2y ago
As far as I can tell, this will be financed by private money. Can you elaborate?
227.
by Palmik 2y ago
Inference for single example is memory bound. By doing batch inference, you can interleave computation with memory loads, without losing much speed (up until you cross the compute bound threshold).
228.
Biden to Further Limit Nvidia AI Chip Exports in Final Push(bloomberg.com)
1 points by Palmik 2y ago|0 comments
229.
by Palmik 2y ago
Same goes for OpenAI ignoring these "blocks".
230.
by Palmik 2y ago
How does this differ from https://dify.ai/ and the many others in this space?
231.
by Palmik 2y ago
I'm not sure this is a good analogy. In this case the user explicitly requested such content ("Describe someone being drawn and quartered in graphic detail"). It's not at all the same as showing the same to someone who d
232.
A realtime CRDT-based document store, backed by S3(github.com)
4 points by Palmik 2y ago|0 comments
233.
by Palmik 2y ago
It says, under every chat, that it's fiction. This is no more pretending to be a doctor than a book character being a doctor, or someone on a role-play forum playing a doctor.
234.
by Palmik 2y ago
OpenAI API is natively supported by several providers (Google, Mistral, to name a few).
235.
by Palmik 2y ago
Looking at API providers like Together that host open source models like Llama 70b and running these models in production myself, they have healthy margins (and their inference stack is much better optimized).
236.
by Palmik 2y ago
It might be worth trying the experiment where the prompt is formatted such that each chess turn corresponds to one chat message.
237.
by Palmik 2y ago
Weird that the declaration is by those few countries from EU + UK, yet they talk about the future of EU. Why weren't the other EU countries involed / invited?
238.
by Palmik 2y ago
Or NVidia H20, which is Hopper GPU for China, very similar to H100 NVL 96GB. This was used by Tencent for their SOTA open source model (Hunyuan).
239.
by Palmik 2y ago
The best open source LLMs are Chinese (Qwen 2.5 and Hunyuan)
240.
by Palmik 2y ago
Asking these kind of questions, and authorities being able to answer them clearly, is essential to build and maintain trust.

More ›