Veedrac

49 ms·

571.
by Veedrac 6y ago
Fusing that would be very difficult, since it's hard to write that many registers with a single op, and I don't think they recommend it. This does mean signed overflow checking will be comparatively expensive. But thanks for the r
572.
by Veedrac 6y ago
If I thought 5 instruction fusions were necessary, I would not be a fan of fusion either.
573.
by Veedrac 6y ago
> having to fuse super complex (5 instruction or more) sequences Can you give an example of someone advocating for 5 instruction fusion? Normally it's limited to three.
574.
by Veedrac 6y ago
Mostly the concern around the lack of instructions in RISC-V revolves around a few well-known cases (eg. indexed loads) where the instructions to fuse are pretty canonical. There is always room for creativity, but that would be the same wit
575.
by Veedrac 6y ago
Opportunity cost in what sense? If the ISA is simple and regular, then the silicon costs should be miniscule, and the engineering costs not meaningfully larger than if those instructions were separate unfused instructions that had separate
576.
by Veedrac 6y ago
They have added them, it's just they're in bitmanip, which isn't finalized, nor is the extension mandatory.
577.
by Veedrac 6y ago
> Essentially that's part of the whole RISC idea - simple means faster - you can run your core clocks faster if the decode (and address calculations etc etc) are simpler - getting rid of lots of addressing modes was a big part of th
578.
by Veedrac 6y ago
I've never really been a fan of this take. It seems to rest a lot on the ideas that: 1) Fusion is hard. While it can be hard (fusing x86 will be), I do not see why it would be meaningfully difficult for RISC-V hardware (except on dev
579.
by Veedrac 6y ago
If you had a friend that wasn't able to suffer through some quirk of atypical DNA, would that person still have rights? Would it be unethical to enslave them? (Hint: yes)
580.
by Veedrac 6y ago
There's this really weird idea that you need high-bandwidth highly-multimodal sensory data to ‘truly’ understand things, but this makes no sense to me. 1. It's all just data in the end. The signals from your eye are not any more r
581.
by Veedrac 6y ago
> outside of consumer products Well FAANG all produce consumer products, so that wipes out a bazillion legitimate applications, but you've still got that Facebook and Google sell ads, which uses AI for targeting. Data centre cooling
582.
by Veedrac 6y ago
There are many BERT-based models that would have made for a good numeric comparison, had they tested on few-shot learning, but I'm not aware of any that have.
583.
by Veedrac 6y ago
The results reported there are what most people would call ‘semi-supervised learning’, not ‘few-shot’. The true few-shot results are in a few places in the paper, https://arxiv.org/abs/2009.07118 , labeled with ‘- dist’
584.
by Veedrac 6y ago
I had this discussion with Timo Schick, the first author, on Reddit. My final comment is copied below, with the relevant context. Note that GPT-3's approach to *GLUE involved no training on the task, just a good choice of prompt, where
585.
by Veedrac 6y ago
Here's a simpler, but more technical, attempt at an explanation. --- while (true) { Treat your initial 64 bit random value as a fixed point fraction in the range from 0 to just under 1. u64 fract = random64(); Con
586.
by Veedrac 6y ago
No it isn't. The encoding is conditional on all previously generated characters in the context window.
587.
by Veedrac 6y ago
I agree that they'd be underground.
588.
by Veedrac 6y ago
This argument disproves fossils. E: To be pedantic, it only disproves most fossils.
589.
by Veedrac 6y ago
The issue is that the network has to learn from an n-step transition, which is hard to do when minimally parameterized.
590.
by Veedrac 6y ago
> In reality it's $45-50k car. Is something wrong with the $38k version?
591.
by Veedrac 6y ago
Density on the leading node (CPU & GPU graphs): https://docs.google.com/spreadsheets/d/1NNOqbJfcISFyMd0EsSrh... I've also got an NVIDIA performance graph, since GPUs scale well with transistors: https:&#
592.
by Veedrac 6y ago
The key most people miss about the anthropic principle, and that seems to be the confusion here, is that when you're filtering out false hypotheses (those inconsistent with the observed facts), all remaining consistent explanations inc
593.
by Veedrac 6y ago
Better overly granular than overly coarse.
594.
by Veedrac 6y ago
document.getElementsByTagName('video')[0].playbackRate = 3
595.
by Veedrac 6y ago
Thanks.
596.
by Veedrac 6y ago
Eg. what alternative?
597.
by Veedrac 6y ago
The article doesn't give examples, but the paper does. Here are a few examples of question overlap, for context. Test Question: who plays max voice in a goofy movie Train Question: who does max voice in a goofy movie Ans
598.
by Veedrac 6y ago
OpenAI Five lost just 42 out of 7,257 matches. It's true that it has exploitable weaknesses, but it did not simply fall apart.
599.
by Veedrac 6y ago
A computation can represent causality without being made of causality-neurons.
600.
by Veedrac 6y ago
GPT-3's failure at larger addition sizes is almost fully due to BPE, which is incredibly pathological (392 is a ‘digit’, 393 is not; GPT-3 is also never told about the BPE scheme). When using commas, GPT-3 does OK at larger sizes. Not

More ›