Yondon Fu

@yondon.eth

188 Following

332 Followers

Yondon Fu pfp

0 reply

0 recast

2 reactions

Yondon Fu pfp

In progress. Took a bit to get used to narration style, but I'm into it now. This from the blurb piqued my interest: "It is a hard-won utopia built on technologically-generated abundance, and also on complex and mandatory systems of labelling all public writing and speech."

0 reply

0 recast

1 reaction

Yondon Fu pfp

Huh didn't realize Chrome would automatically do this for resource hungry ads neat

0 reply

0 recast

1 reaction

Yondon Fu pfp

Seems plausible that protocols that prove hardware/software environments (via remote attestations) in open networks would result in the eventual restriction of hardware/software that is usable on the network. Always the case? https://warpcast.com/yondon.eth/0x417c3bfa

0 reply

0 recast

1 reaction

Yondon Fu pfp

Been reflecting on remote (hardware) attestations recently and the behavior incentivized by protocols that use them with the abandoned Web Environment Integrity proposal from last year as a case study. https://www.yondonfu.com/p/remote-attestations-and-dangerous

0 reply

0 recast

0 reaction

Yondon Fu pfp

Training log (gi): 35x1min (10s rest) for a birthday. Ok time for bed.

0 reply

0 recast

2 reactions

Yondon Fu pfp

Does macOS/iOS support attestations for secure enclave keys similar to what Android has with https://developer.android.com/privacy-and-security/security-key-attestation? I haven't seen any docs quite like the Android one yet, but maybe missing something

0 reply

0 recast

2 reactions

Yondon Fu pfp

AI generated samples and Made with AI labels. And how Drake's diss track ties into it all. https://www.yondonfu.com/p/ai-generated-samples-and-made-with?r=4rd4&utm_campaign=post&utm_medium=web

1 reply

0 recast

2 reactions

Yondon Fu pfp

An initial look at a content provenance standard using signed claims with assertions about media assets: https://warpcast.com/yondon.eth/0x6fc63b89

0 reply

0 recast

0 reaction

Yondon Fu pfp

New post on a content provenance technical standard: What C2PA does and does not solve. Plus questions from an initial dive into the 2.0 spec. https://www.yondonfu.com/p/what-c2pa-does-and-does-not-solve?r=4rd4&utm_campaign=post&utm_medium=web

0 reply

1 recast

2 reactions

Yondon Fu pfp

Neat ELO ranking arena for image models. Interestingly SDXL-Lightning, which is a distilled version of SDXL, ranks ahead of SDXL https://huggingface.co/spaces/TIGER-Lab/GenAI-Arena

0 reply

0 recast

0 reaction

Yondon Fu pfp

Playing with and mixing styles from LoRAs is fun. Readyweights (Herndon + Dryhurst) + Nausicaa (Ghibli)

2 replies

0 recast

1 reaction

Yondon Fu pfp

Me: *reaches for lapel to go for a lapel guard* Training partner: "Please no just not that" 😅

0 reply

0 recast

3 reactions

Yondon Fu pfp

Training log (gi): - Drilled passing DLR + transition to back take - Lost count of number of rolls but some good ones at the beginning - getting a bit better at staying calm and playing my game. Gave up a reverse arm bar later on though 😢 - Rolling with former wrestlers is tough 🫠

1 reply

0 recast

3 reactions

Yondon Fu pfp

Don't think anything clears the mind for me as much as rolling

1 reply

0 recast

6 reactions

Yondon Fu pfp

Training log (nogi) - Drilled kiss of the dragon to backtake - 6x5min rolls Why is it called kiss of the dragon? Nobody knows but it’s provacative

1 reply

0 recast

1 reaction

Yondon Fu pfp

Assuming that X open sourcing Grok means releasing open weights seems that this is a case of commoditizing your complements ala Meta with llama more than anything else given that Grok the product is where most of the value is vs. Grok the model?

1 reply

0 recast

1 reaction

Yondon Fu pfp

What are the strongest criticisms (if any) of the upcoming changes to congestion pricing? I'm aware of some of the pro arguments, but would like to learn more about the con arguments.

2 replies

1 recast

1 reaction

Yondon Fu pfp

Being able to train a 70b model on 2x consumer GPUs seems like a big deal and important milestone towards a diversity of actually user aligned personalized AI systems that can power tools for info/cyber defense. https://www.answer.ai/posts/2024-03-06-fsdp-qlora.html

1 reply

1 recast

4 reactions

Yondon Fu pfp

A way to think about finetuning v. prompt eng (+ any technique w prompting like RAG) that I've been playing with: Who takes on more of the burden of nudging the model to produce consistent good outputs for a domain - the model consumer at prompting time or the model trainer via examples at training time?

2 replies

0 recast

2 reactions