Posts by corbin@defcon.social
(DIR) Post #ARsEpvHmo7p07DQ9Pk by corbin@defcon.social
2023-01-21T16:40:42Z
2 likes, 0 repeats
@soatok They were so anti-furry that they momentarily forgot how to be racist.
(DIR) Post #AShpIdXmU9ce2Xb4YC by corbin@defcon.social
2023-02-15T14:06:57Z
0 likes, 0 repeats
@simon @glyph Or perhaps we're in an infinite library, and we only just now realized how many possible books there are.I keep thinking of reduction, rendering, distillation. You can reduce a log to wood pulp -- don't drink that! But you can do more nasty stuff to it and eventually produce books and furniture.I wonder what man-made horror will provide the glue in my analogy...
(DIR) Post #ASoEePr7im8plDsyTA by corbin@defcon.social
2023-02-18T15:36:40Z
0 likes, 0 repeats
@lanodan @lispi314 @ariadne This is a slight misstatement of what would be legal in a world (or at least a USA) without copyright.I've helped reverse-engineer a few systems, from GPU drivers to video-game servers. If copyright didn't exist, then in every case, we could have skipped the cleanroom steps and folks could have ignored any taint when writing open documentation.
(DIR) Post #AT1Ht3NQPCGdIw928m by corbin@defcon.social
2023-02-24T23:26:45Z
0 likes, 0 repeats
@simon I was playing with Huggingface's question-answering pipeline this week. I fed it my documentation for my house, and it was able to answer basic questions in under a second, on my laptop.That said, I think that what might be more interesting for you is Petals, KoboldAI Horde, or other distributed approaches. On a phone, edge computing is at a premium; even if a model fits, it might be cheaper to run elsewhere.
(DIR) Post #ATWAkVbM6DEsgPeaae by corbin@defcon.social
2023-03-11T21:02:41Z
0 likes, 0 repeats
@simon I think that your reaction is reasonable, but e.g. https://huggingface.co/bigscience/bloom is already licensed appropriately. Perhaps the real innovation here is the use of 4-bit weights; I don't think anybody anticipated that a large model would still be cogent at reduced fidelity.
(DIR) Post #ATWQ4E5OQhv7SoZ0HQ by corbin@defcon.social
2023-03-11T23:54:38Z
0 likes, 0 repeats
@simon Yep. I have a small harness (<100 lines of Python using Huggingface's Transformers) and I can choose from about half a dozen working models, including BLOOM. My current platform is a last-gen Intel CPU with about 16GiB RAM free.There's also Petals, for the specific case of BLOOM, but I prefer my privacy. I think that Petals buy-in starts at 10GiB RAM.
(DIR) Post #AU3kbvHj0zgDX2qbBY by corbin@defcon.social
2023-03-28T01:50:18Z
0 likes, 0 repeats
@simon @glyph @fancysandwiches @geoffreylitt First, notice that we don't want to just minimize environmental damage; we want to structurally prevent it.Here's my thoughts: what if we had programming environments which automatically simplify, refactor, prove, and prune code? What if we had build systems which automatically remove differences between packages, forget module boundaries, and permanently optimize compositions? We could destructure model-generated code and absorb it.I'm not thinking of e.g. Unison or Dark. I'm thinking of languages with genuine eta-equivalence, with structural types instead of nominal types, and IDEs designed to eat and digest code.It goes without saying that copyright can't be invited to the party.
(DIR) Post #AUQEOiZ3v9h8e36vlw by corbin@defcon.social
2023-04-07T22:07:34Z
0 likes, 0 repeats
@simon How about "ChatGPT is a scam" or "ChatGPT is falsely advertised"? The issue here is that OpenAI is selling this product; it's like a scene from The Jungle, and regulation is incoming in several jurisdictions.But, what should we do if OpenAI were to act ethically and pull ChatGPT? Well, we would still need to explain why large models are prone to confabulation!
(DIR) Post #AUiMKxtLB4CuXo1zGq by corbin@defcon.social
2023-04-16T16:01:19Z
0 likes, 0 repeats
@simon I don't understand why this is desirable. As you yourself point out, the amount of data that has to be streamed and cached by the Web browser is unreasonable.
(DIR) Post #AUiMqy6NvlV2SnOQ3E by corbin@defcon.social
2023-04-16T16:06:42Z
0 likes, 0 repeats
@simon I can trivially prove that my local LLaMA harness won't make any network calls. Doing the same for a browser is a massive headache.Sandboxes are an anti-pattern; they are what we use for untamed software. However, LLMs are brand-new and trivial to tame, so no sandboxes are required.
(DIR) Post #AUiRSbfoATDFitAGOG by corbin@defcon.social
2023-04-16T16:58:49Z
0 likes, 0 repeats
@simon Generate code in a language which denotes pure total functions. Then automatic execution can't do anything worse than waste a few moments of CPU time or a few GiB/min of RAM, and automatic analysis of code is relatively straightforward.People are mostly generating Python and ECMAScript. EMCAScript technically can be tamed, but Python can't.If you want to generate untrusted code and inspect it, then you need to avoid Turing-completeness. We've known this for like a century.
(DIR) Post #AUiePibtDfCvlxHime by corbin@defcon.social
2023-04-16T19:24:08Z
0 likes, 0 repeats
@simon I guess I ought to write a blog post explaining what taming is. There's an old E document, at least: http://www.erights.org/elib/legacy/taming.htmlYes, WebAssembly is tamed. Yes, emulators written in tamed languages are freely tamed. No, Python's native type theory is not tamed simply by running in a managed runtime; for example, CPython is not tamed, although PyPy has object spaces which are somewhat tame.
(DIR) Post #AVfTF2P68SM3NAvmhU by corbin@defcon.social
2023-05-11T04:29:46Z
1 likes, 1 repeats
@ariadne Not only is asking for donations not against their code of conduct, but the administrators of sfba.social have an Open Collective page for receiving community donations. Not just illogical, but hypocritical.