@DougCalidas Oh you are. Think time for one, and even look into the definition of distilling and add some brains for critical thinking on what's happening. The best way when out innovated is to out innovate back. Copyright and Patents already protect genuine violations.
@SarahKHeck@mkratsios47 US folks have let go of their critical thinking skills. Distillation is like copying someone's homework answers; it doesn't expose the methods to reach results for any other inputs. Use of Kimi shows that it's innovations that have driven their results. Theft is a baseless slur.
@MTSlive Do you understand what distillation is and how it doesn't betray any logic on methods? Only fixed answers to specific questions. Only practical use is to max out benchmarks. Kimi is achieving good results to a wide range of cases; it's clear vendors are being out innovated.
@Michael7ucci Better to sanction politicians talking shit and learn to apply critical thinking. Trace back the conjecture sources and understand their incentives.
@0xileri That statement is intellectually dishonest. Distillation doesn't improve methods, only static answers to static questions with one specific instance of data.
As for being ahead, geographic location and ownership of non-US companies .NE US superiority in AI.
@ChrisRMcGuire@mkratsios47 Strong statement that's full of holes. It's an American obsession that anyone out innovating must be stealing from them. That doesn't hold up to any competent scrutiny.
@quxiaoyin OpenAI and Anthropic are trying to IPO and flooding the zone with PR that doesn't hold up to scrutiny. The ultimate answer is to out-innovate back. Alleging theft is a cheap shot.
@SecScottBessent Complete bollocks. US vendors have been out innovated, pure and simple. Distillation (copying homework answers) doesn't expose methods, only answers to fixed questions with one set of inputs. The ultimate antidote (and the American way) is to out-innovate back.
@UnderSecE What absolute bollocks. Distillation is like copying someones homework answers; it exposes answers to one specific question with one set of inputs. It doesn't betray the method to get answers for any other values.
Kimi results on a wide range of tasks they suggest out-innovated
@mkratsios47 This is bollocks. Distillation is like copying homework answers. You can copy but it doesn't give you methods usable for any other inputs. About the only useful application is Benchmaxxing - giving the answer to one single test with known, fixed inputs. Reality here is different
@pmddomingos Another tired post that someone else out innovating gets labelled as stealing. Lots of shit flowing about distillation too; copying homework answers doesn't expose methods for all other inputs. Best strategy is to out-innovate back.
@rdominguezibar More bollocks. The measure is tokens per item of work multiplied by the token cost - ie: cost to get a unit of work done - to a common standard of completion. Said Chairman can whistle dixi if 5x the tokens cost 1/35 the value each.
@ramez They'll grow by putting their retail pricing up - when the competitors are already at a much lower cost base. That may help their margin percentages, but not their revenue...
@pmddomingos This is bollocks. Distilling is like copying someones homework answers: only works if the inputs and questions are identical; it doesn't betray the method that works for all other inputs. Ok if you're doing fixed maxxing benchmarks; LLM Arena says this isn't what's happening