jychang

25 ms·

301.
by jychang 1y ago
Correct, in that models know nothing about themselves other than what they are told. Deepseek R1 will tell you that it's created by OpenAI. GPT-4o mini is supposed to be ~8b params from estimates.
302.
by jychang 2y ago
That's pretty terrible. I'm getting 18tok/sec Gemma 3 27b QAT on a M1 Max 32gb macbook.
303.
by jychang 2y ago
... He's referring to "Christ is risen". That's way more broad (conceptually) and very in-character for the Pope, compared to some transient current events.
304.
by jychang 2y ago
Try mradermacher/amoral-gemma3-27B-v2-qat-GGUF
305.
by jychang 2y ago
Those are the same quant, but this is a good example of why you shouldn't use ollama. Either directly use llama.cpp, or use something like LM Studio if you want something with a GUI/easier user experience. The Gemma 3 17b QAT GGUF
306.
by jychang 2y ago
MLX is slower than GGUFs on Macs. On my M1 Max macbook pro, the GGUF version bartowski/google_gemma-3-27b-it-qat-GGUF is 15.6gb and runs at 17tok/sec, whereas mlx-community/gemma-3-27b-it-qat-4bit is 16.8gb and runs at 15tok&
307.
by jychang 2y ago
I don't think ollama is quantizing the embeddings table, which is still full FP16. If you're using MLX, that means you're on a mac, in which case ollama actually isn't your best option. Either directly use llama.cpp if y
308.
by jychang 2y ago
> financial firm That means they have direct fiber connections to the Tier 1/Tier 2 guys. The big ones have direct fiber connections to the NYSE. If they're not rolling their own connection, they're using BT Radianz, IPC S
309.
by jychang 2y ago
> I’m old enough to know that even in the 80s Mr. Rogers was in front of congress asking the government not to cut funding because some conservative congressmen were opposed to some of the content. Exactly. Congress funding something ten
310.
by jychang 2y ago
This comment reads like simple.wikipedia.org
311.
by jychang 2y ago
How old are you? Are you familiar with broadcast TV that is supported by "viewers like you"? And historically TV advertising was banned until the FCC allowed it. And you do realize that news used to be paid? You'd be surprise
312.
by jychang 2y ago
American websites implement GDPR even though that's an EU law. Websites that are used across geopolitical boundaries will invariably follow US law. There will certainly be a few exceptions, but if the law is written like the GDPR, then
313.
by jychang 2y ago
Yeah? The government defines what is murder, defines what is tax evasion, and defines tons of other stuff already? Some states already have laws against billboards?
314.
by jychang 2y ago
Good thing they’re trying to move off x86, then
315.
by jychang 2y ago
To be fair, pacing as a big flaw of LLMs has been a constant complaint from writers for a long time. There were popular writeups about this from the Deepseek-R1 era: https://www.tumblr.com/nostalgebraist/778041178124926
316.
by jychang 2y ago
You're just describing calling a customer service phone line in India.
317.
by jychang 2y ago
Yeah, openweb-ui is the best frontend for API queries. Everything seems to work well. I've tried LibreChat before, but the app is terrible at generating titles for chats instead of leaving it as "New Chat". Also it lacks a wo
318.
by jychang 2y ago
LLMs clearly do have a world model though. They represent those ideas at higher level features in the feedforward layer. The lower level layers are neurons that describe words, syntax, and local structures in the text, while the upper level
319.
by jychang 2y ago
... You can't see if a dick pic is a dick pic unless you open it?
320.
by jychang 2y ago
I don't think most users use those integrations, they're mostly just a feature bullet point.
321.
by jychang 2y ago
Would you do it if it was cheaper per kwh? Say you have an EV where any battery you get from the swap station is better than 90% degradation, and you pay $0.40 per kwh for the electricity in the battery- but you get the option to take a bat
322.
by jychang 2y ago
Eh "when everyone is special nobody is anymore" is strongly not true when it comes to nukes and MAD. This is obvious.
323.
by jychang 2y ago
I don't get how Apple patched this. What does the patch change to do differently? I don't see a way of fixing this without shutting off the current Airtag network.
324.
by jychang 2y ago
Definitely not. They don't distill their original models. 4o is a much more distilled and cheaper version of 4. I assume 4.5o would be a distilled and cheaper version of 4.5. It'd be weird to release a distilled version without ev
325.
by jychang 2y ago
I love Scott, but I think OP's meaning is not about verifying discoveries, but rather having the tools necessary to create discoveries in the first place. You can't invent an electric motor without electricity. Alternatively, thin
326.
by jychang 2y ago
It's 2025, regular english is machine readble
327.
by jychang 2y ago
Yeah, I was thinking "what does the Wine guys have to do with this?" This project is definitely going to get C&D'd.
328.
by jychang 2y ago
Who cares anymore in 2025? Maybe in 1999, but now in about 1 year we'll have agents that can manage subscriptions automatically. Actually, I'm pretty sure OpenAI Operator can already do that, but I don't pay $200 for Pro so I
329.
by jychang 2y ago
Rename it to “United Kingdom of England and Scotland” and party like it’s 1707. Problem solved.
330.
by jychang 2y ago
I don't buy this, simply because if the Chinese government were to back an effort, it wouldn't be Deepseek. Alibaba has Qwen. Baidu, Huawei, Tencent, etc all have their own AI models. The Chinese government would most likely push

More ›