chaxor

36 ms·

421.
by chaxor 3y ago
Good question. I don't think there's any real answer to this at all. https://slatestarcodex.com/2020/03/10/for-then-against-high-...
422.
by chaxor 3y ago
But then it's processed sugars! https://slatestarcodex.com/2020/03/10/for-then-against-high-...
423.
by chaxor 3y ago
Yes, and in fact that's the best method available if you want good performance. I would suggest using a local open source model to do this however, to cut down on costs and make it far simpler to deal with than the unwieldy OpenAI syst
424.
by chaxor 3y ago
The wording of this would be extremely difficult though. Are local NER models part of this? Relation extraction? What about GPTs that only decode to DSLs? If the model only outputs DNA sequences is that an area that can be more illegal or
425.
by chaxor 3y ago
Perhaps, but taken with other works in the area, a better picture does emerge regarding this claim. Substantial works already showing reasoning capabilities in GPT-4, which show that these models do reason very well - near human performance
426.
by chaxor 3y ago
There are substantial works already showing reasoning capabilities in GPT-4, which show that these models do reason extremely well - near human performance for many causal reasoning tasks. (1) Additionally, there is a mathematical proof tha
427.
by chaxor 3y ago
Yet another reason to run your own local LLM. The new local LLMs are actually better than ChatGPT and approaching GPT-4 capabilities. Checkout Gaunaco - it's absolutely incredibly good, and quite easy to get running at home with a simp
428.
by chaxor 3y ago
Even when someone understands the architecture very well, there's still great utility in having a full graph representation of the full NN architecture. This could be for teaching students, or for doing analyses on the structure of th
429.
by chaxor 3y ago
https://miro.medium.com/v2/resize:fit:1100/format:webp/0*Y4G... Does that work?
430.
by chaxor 3y ago
You're right that Merkle trees are useful here, but adding proof of work to, e.g. run the code to produce some lfs data (database, NN weights, physical parameters, etc), which is checked by a checksum, could be a fantastic addition for
431.
by chaxor 3y ago
Do you mean a graph that contains all neurons from the network in one structure? Similar to these: https://gfycat.com/BonyTotalArthropods https://gfycat.com/BitesizedWeeBlacklemur That would be wonderful an
432.
by chaxor 3y ago
You definitely should not make claims like 'LLMs can't drive cars'. LLMs have already been shown several times to be able to navigate as agents in different world environments. Obviously, I'm not advocating this is a g
433.
by chaxor 3y ago
Listing the number of neurons in a brain has very little to do with these systems, so it's pretty meaningless. Also, the number of neurons in Wernicke's area and the PFC is quite a small fraction of the brain, making this even mor
434.
by chaxor 3y ago
I know this is a bit of a meme now in HN, and I agree that GPLv3 is certainly a much better license. However to push back a bit, I would say that at least the license allows researchers to do the important work that needs to be done. It se
435.
by chaxor 3y ago
Anyone that combats something based on the name alone isn't really worth listening to. There can be great use cases for Blockchain just like this, wherein the proof of work is less taxing, or optional. Of course, HN has a rabid respo
436.
by chaxor 3y ago
They will likely make Siri an LLM but not make a big deal about it. Just simply put it as a minor revision note, because that's effectively all it is.
437.
by chaxor 3y ago
I'll give a very simplistic answer to your question and skirt some detail. Essentially, Transformers can 'look at' everything within the context with perfect fidelity (only one 'hop' between tokens to 'get the
438.
by chaxor 3y ago
'shell scripts with worse syntax' lol I wish shell could emulate Alpine on a non-linux box. Shell script with worse syntax for config VM may be closer to a qemu cloud init file.
439.
by chaxor 3y ago
Cloud has a variable price (up to the whim of whatever they decide the price to be that day) so it's uncertain, but typically it is far" more expensive for this type of application. So when faced with 1) probably far more expensi
440.
by chaxor 3y ago
Man that's crazy. Think of all that extra time you'll have if timsort was used! An extra 2ms to go on that walk while your system is booting.
441.
by chaxor 3y ago
I see a few OSes on here, but does it work with BSD? I would imagine FreeBSD support would be desired for routers and such.
442.
by chaxor 3y ago
How could be anything other than new... It's Rust. It's not going to be from 1975 because it can't be.
443.
by chaxor 3y ago
There are plenty (perhaps far too many) tools for doing basically `curl` to OpenAI. Local LLM tools are needed however; and are much better for deploying systems on terabytes of data at fractions of the cost.
444.
by chaxor 3y ago
I believe the goal is to only restrict systems that have capabilities that do not yet exist (GPT- 4.5, or so). GPT-4, ChatGPT, or Wizard-Vicuna, etc are not really the concern to many experts regarding the type of safety legislation that is
445.
by chaxor 3y ago
You misspelled JavaScript
446.
by chaxor 3y ago
The models tend to degrade when trained to be safer. A GPT-4 talk on youtube by personnel from Microsoft has documented this phenomenon with the 'Tikz Unicorn' evolution shown in the GPT-4 technical paper. The model gets qualitati
447.
by chaxor 3y ago
It's pretty disgusting to see how many of these exist, yet so few are local-only with local LMs. It's exceedingly obvious that local LMs are desired and far better for this task, and only idiots would be interested in paying for s
448.
by chaxor 3y ago
Mistakes may be the next (extremely short lived) way we tell a bit from human, once 'correcting by plugin' becomes automatic in various pipelines. We get plenty of that, but we have the option to be lazy.
449.
by chaxor 3y ago
`rm -rf /` could be a set of 'bad words' too. Or `shred -uf /dev/sda`
450.
by chaxor 3y ago
Telemetry is incredibly useful for ensuring that people don't use the product. Many likely be doing a lot of switching away from Go intentionally, due to this.

More ›