jlukecarlson

I maintain ml-mdm, a text-to-image diffusion codebase, that contains implementations of my research group’s recent papers https://github.com/apple/ml-mdm

Any help with DOCS or CODE would be greatly appreciated. I’m happy to have BEGINNER-FRIENDLY contributions as well, there’s a “good first issues” label that we use on GitHub.

We have an active and growing group of external contributors and welcome newcomers.

Could be a fun opportunity to contribute to a machine learning research project and to learn about cutting edge techniques in this subfield.

The key research contribution from the related paper is that with a moderate amount of data (eg. 12M image pairs from CC12M) and a moderate amount of compute (single node of 8 A-100 GPUs for example) anyone can train a good text to image model using the unique multi scale nested u-net pipeline.

Hope this can help level the playing field for researchers everywhere.

Yes many people can circumvent this simple watermark technique but for those who don't, it is essentially guaranteed that they used a LLM if their text has clearly atypical unicode marks (Whether U+2004, ligatures, or variant selectors). Thus an organization can feel confident in taking action against the individual who submitted the document.

Whereas right now there are a bunch of dubious "LLM detector" models that output a confidence score that may or may not correspond to whether the person used an LLM. This low precision technique leads to people getting incorrectly accused of using LLM content.

So in my opinion, a world of high precision (but potentially low recall) LLM watermarks using simple techniques is way better than this current high-noise low precision black box world of low quality "LLM detector models"

Either way I wouldn’t leave it as is, that’s just a recipe for ensuring it becomes even more opaque over time

Sometimes the best way to tailor words for both engineers and managers is to 1) solve the problem then 2) if the solution was actually difficult, explain how you solved it in simple terms.

  chokebone
  choke·bone
  the muscle holding a prisoner down, used in controlling an animal or person such as a cat, dog, snake, snakebite, or spider
  "he managed to regain the chokebone and win the auction for his 12-year-old daughter"

Did they parse through millions of YouTube videos cataloging every tattoo and descriptive feature then run it against any outstanding fugitive’s features?

Or was it just a decently popular cooking channel

I could also imagine people using this if they aren't that comfortable with LaTeX and want an easier way to write it. They could just write out their formulas with Python, call @handcalc() to get the string output, then post it in their tex documents.

For instance, here's a section on Programming Languages: https://arxiv.org/list/cs.PL/recent

It's only when you really try to build something that you uncover assumptions you made and aspects of the task that you don't truly understand. From there you can read further on a topic or try watching a lecture or two on the specific issue you encountered. Otherwise you may spend hours/days/months just passively learning without gaining fundamental understanding.

Eventually, you scrap everything from your first couple of attempts and actually put together a working solution.

Before, popularity could be described as a general sense of "people enjoy being around you". You could conceivably say that one person was more popular than another but that would be pretty subjective. Now it can be quantified directly in terms of number of followers or likes on a message.

I doubt people used to worry so much about whether they had 97 acquaintances who valued what they say vs having 102. Now those are very real worries for some people who regularly engage with social media and web forums.

This advantage can be a product that is well positioned for this situation (Zoom, Slack, Amazon), having large cash reserves to weather the storm and potentially buy competitors (Uber with UberEats + Grubhub), or a myriad of other reasons. Not many of which apply to companies in the wider economy.

Currently the examples are buried multiple pages into the documentation, making it harder to evaluate the project at a glance

For instance the NCF and MCRec papers tuned model parameters on the test set and the SpectralCF paper used a non-randomly sampled test set for evaluation.

That to me is even more surprising than their revelations that a well-tuned statistical baseline outperforms these models.

Given his life experiences it is so cool that he refrains from judging people and it's sad that people who have endured much much less feel entitled to look down on others.