3 ms·

It would be interesting to play with this model directly (if it ever gets released...) One interesting thing is that foley is not real life, so depending on wh

It would be interesting to play with this model directly (if it ever gets released...) One interesting thing is that foley is not real life, so depending on where they get the dataset from (or how they "generated" it) they might be learning how things actually sound, or how foley artists make them sound. Its probably most noticeable in either car noises, or eagles: https://www.youtube.com/watch?v=jQI-ddEPTx4 https://www.youtube.com/watch?v=jQI-ddEPTx4 (which are often overdubbed with hawk noises)

echelon 2y ago
Adobe intern research almost never sees code / weights releases, sadly. It's an academic peg for the students and marketing for Adobe. Maybe future interns will start taking this into account and work somewhere they can reuse their research. We need a broader open research / open source conversation about ML models. Meta calling their approach "open" and what the OSI deems "open" are both flawed and/or misleading. China is really leading the "open" AI research game. They release so many goodies - papers, code, weights, training scripts, notebooks, and sometimes even training data. It's gotten to the point that whenever I see a Western publication, I groan. More recently their work seems to be done on H800s and export controlled GPUs. This is also why they're starting to experiment with wildly innovative architectures. Despite being resource constrained, they're really kicking ass.
- fngjdflmdflg 2y ago
  >Meta calling their approach "open" and what the OSI deems "open" are both flawed and/or misleading.[...] It's gotten to the point that whenever I see a Western publication, I groan. llama weights are open, and Qwen 1.0 was based on llama,[0] so clearly those papers are useful to some people, even if some people want to be angry at Meta for not being open enough for them. In fact I would say Qwen is less useful than llama as their technical reports are not nearly as detailed as Meta's. Also, Why groan just from seeing the origin of a project? Just read it to see if it is open or not. And there are also lots of popular Chinese models that are completely closed with no code or paper eg. Hailuo, Kling, etc. [0] https://arxiv.org/pdf/2309.16609 https://arxiv.org/pdf/2309.16609 p. 6

echelon 2y ago

Adobe intern research almost never sees code / weights releases, sadly. It's an academic peg for the students and marketing for Adobe. Maybe future interns will start taking this into account and work somewhere they can reuse their research. We need a broader open research / open source conversation about ML models. Meta calling their approach "open" and what the OSI deems "open" are both flawed and/or misleading. China is really leading the "open" AI research game. They release so many goodies - papers, code, weights, training scripts, notebooks, and sometimes even training data. It's gotten to the point that whenever I see a Western publication, I groan. More recently their work seems to be done on H800s and export controlled GPUs. This is also why they're starting to experiment with wildly innovative architectures. Despite being resource constrained, they're really kicking ass.

fngjdflmdflg 2y ago
>Meta calling their approach "open" and what the OSI deems "open" are both flawed and/or misleading.[...] It's gotten to the point that whenever I see a Western publication, I groan. llama weights are open, and Qwen 1.0 was based on llama,[0] so clearly those papers are useful to some people, even if some people want to be angry at Meta for not being open enough for them. In fact I would say Qwen is less useful than llama as their technical reports are not nearly as detailed as Meta's. Also, Why groan just from seeing the origin of a project? Just read it to see if it is open or not. And there are also lots of popular Chinese models that are completely closed with no code or paper eg. Hailuo, Kling, etc. [0] https://arxiv.org/pdf/2309.16609 https://arxiv.org/pdf/2309.16609 p. 6