xiphias2

16 ms·

271.
by xiphias2 2y ago
Maybe it was part of the reason that they accepted the PR. The fork would happen anyways if they don't allow any LLM. A bit like how Android came after iPhone with open source implementation.
272.
by xiphias2 2y ago
I guess people don't understand what it really means: government workers passing around naked selfies of their daughter that they are sending anyways to their boyfriends as part of the dating process.
273.
by xiphias2 2y ago
Sure, it would be just cool to not have to do that
274.
by xiphias2 2y ago
With HTTP streaming the browser shows that it's still loading data. Is there some mitigation for it after the initial loading?
275.
by xiphias2 2y ago
He didn't even tell whether ETH, the second highest value cryptocurrency is considered security or not, even though he was asked about it all the time.
276.
by xiphias2 2y ago
As you can see even the basic documentation is compromised at this point: https://unplannedobsolescence.com/blog/behavior-belongs-in-h... <button onclick="alert('I was clicked!')">Click me<
277.
by xiphias2 2y ago
You're right, sorry...I just tested Qwen models, not QwQ, I see QwQ only has 32B.
278.
by xiphias2 2y ago
Have you compared it with 8-bit QwQ-17B? In my evals 8 bit quantized smaller Qwen models were better, but again evaluating is hard.
279.
by xiphias2 2y ago
All you are writing is true, but I would still put the main responsibility to workaway. Writing that ,,you need to obtain the correct visa'' is just not enough for a big company thats main profile is arranging work.
280.
by xiphias2 2y ago
,,Some hosts may give a paid allowance to ensure they are offering at least the minimum wage in their country'' It's sad, but workaway is similarly to airbnb (hotel service) and uber (taxi service) uses ,,work'' an
281.
Ask HN: Lambda Labs 390x bill
2 points by xiphias2 2y ago|0 comments
282.
by xiphias2 2y ago
There seems to be some part of the LLM capabilities being lost by hacking some benchmarks. Claude 3.7 is a great example of a model clearly beating 3.5 in all benchmarks, but slowly destroying my code base by adding lots of extra lines or h
283.
by xiphias2 2y ago
Something even cooler would be a model trained for 4 or less (1.33) bit weights instead of quantized after pretraining. Math units are completely underutilized when I'm inferencing with batch size of 1, and post-training quantization u
284.
by xiphias2 2y ago
Putting generative AI inside tools without giving deep understanding of those tools to the AI generally made me more confused and frustrated than outside of it: for example Gemini forced itself on me on my SMS app, so I thought I ask it to
285.
by xiphias2 2y ago
2023
286.
by xiphias2 2y ago
The problem here is that the same can be said to spending trillions of dollars _after_ COVID was gone. It's quite common for goverments that cut spending aggressively getting hated (I have seen in other country where I was born and rai
287.
by xiphias2 2y ago
,,By my personal estimate currently GPT 4o DeepResearch is the best one. '' If the o3 based 3 month old strongest model is the best one, it's a proof that there were quite significant improvements in the last 2 years. I can&#
288.
by xiphias2 2y ago
Also reusing key/values for different queries can compress the KV cache, it can be an 1000x or 10000x improvement in bandwidth if the model is trained for it.
289.
by xiphias2 2y ago
I wish they would start producing graphs with quantized version performances as well. What matters is RAM/bandwidth vs performance, not number of parameters.
290.
by xiphias2 2y ago
Boycott is not the same a burning cars. The sales of the car I own are going down in general, but the only way it affects me is that I know the software bug won't be fixed. Tesla owners must be scared now to get attacked.
291.
by xiphias2 2y ago
I think you are right on all counts, but it doesn't make me depressed, it's just the way humanity moves forward. Since deepseek it's clear that regulary capture just doesn't work in this case, only pure competition. If U
292.
by xiphias2 2y ago
As a European coming to USA and seeing cybertrucks for the first time in real life I think it looks really fun, much more fun than on the pictures. Having lots of problems is expected though...I have an electric car in Europe and I knew the
293.
by xiphias2 2y ago
It was a longer process actually. Left was against the Vietnam war, right was full of war mongerers. While feminism and sexual freedom stayed on the left, military industry changed sides. I don't think the industry loved a new space st
294.
by xiphias2 2y ago
That's basically rg
295.
by xiphias2 2y ago
Motorola was bought for patents to defend Android, it was a clear win. Wiz is much harder to understand.
296.
by xiphias2 2y ago
There's no mention of Starlink, the main competitor (not fiber). Whenever I travel and see the high latencies in hotel, I'm waiting for being able to Starlink from my mobile phone. It requires huge antennas on the satellites, but
297.
by xiphias2 2y ago
I love my Porsche Taycan as well (even if its resale price is dropping like crazy and sales are down), but whenever I travel outside Europe/USA in LATAM, I only see 1 brand: BYD. The main differentiator why people will switch to EVs is
298.
by xiphias2 2y ago
While changing the core packages as a rewrite looks easy (,,just reimplement ls''), compatibility/stability may mean reproducing all the tiny but not safety critical bugs as well. There's enough data from the Android eco
299.
by xiphias2 2y ago
Llama.cpp already moves 1.5bit to the cores, they are converted in SRAM which has much higher bandwidth. FP4 only helps batched inferencing.
300.
by xiphias2 2y ago
Does FP4 support really matter if local inferencing is RAM bandwidth limited anyways? I think we should put more effort into compressing the models well to be able to use local GPUs better.

More ›