lhl pfp

lhl

@lhl

58 Following
14 Followers
lhl pfp
lhl
@lhl
From a market perspective, I actually think Bitcoin will be strong in the short term w/ ETF news, macro, etc, but as "P2P electronic cash," IMO that died when the small-blockers won in 2017. Lightning security dev throwing in the towel: https://lists.linuxfoundation.org/pipermail/bitcoin-dev/2023-October/022032.html
0 reply
0 recast
1 reaction
lhl pfp
lhl
@lhl
A few months ago StyleTTS2 was announced, and the demos sounded great: https://styletts2.github.io/ - some code finally dropped in the repo, so here are some notes on inferencing (and a link to comparison w/ TTS VITS and TTS XTTS output) for those interested: https://llm-tracker.info/books/howto-guides/page/styletts-2
0 reply
0 recast
0 reaction
lhl pfp
lhl
@lhl
If you can script Python, IMO WhisperX to gpt-4-16k. https://llm-tracker.info/books/evals/page/transcription-test
0 reply
0 recast
0 reaction
lhl pfp
lhl
@lhl
Most of my year w/ AI has actually been objectively somewhat unproductive since I've been deep in AI research/yakshaving instead of using it to get projects done, but the past couple days I've been having fun with ChatGPT CI playing with hilbert curves: https://fediverse.randomfoo.net/notice/AZMjVLoHwi5BqAwia8
0 reply
0 recast
3 reactions
lhl pfp
lhl
@lhl
Yeah, it doesn't have everything automatically indexed - uses a local quadrant vector db I believe. Revisiting a big codebase I haven't touched in a while and using Bloop, Cody, Mentat, and Aider and will see what works (maybe Cursor, but I dont' want to switch IDEs)...
1 reply
0 recast
0 reaction
lhl pfp
lhl
@lhl
Yesterday, something incredibly cool was released. The first (afaik) open LoRA MoE proof of concept - it lets you run 6 llama2-7B experts on a single 24GB GPU. Here's my setup notes: https://llm-tracker.info/books/howto-guides/page/airoboros-lmoe
0 reply
0 recast
2 reactions
lhl pfp
lhl
@lhl
You could give Bloop.ai a try, I believe it actively indexes and let's you interrogate any codebase (and presumably docs)
1 reply
0 recast
0 reaction
lhl pfp
lhl
@lhl
Here's a list of resource I've gathered: https://llm-tracker.info/books/llms/page/learning-resources - for LLMs first LoRA MoE project just got released which is pretty exciting to me. The video stuff (AnimatedDiff, CoDef, etc) is pretty exciting too IMO
0 reply
0 recast
2 reactions
lhl pfp
lhl
@lhl
There are less than 2 days left for participating in the KZG ceremony: https://ceremony.ethereum.org/ - for fun, I asked ChatGPT-4 Code Interpreter to plot what the chances were for getting in under the deadline https://chat.openai.com/share/80fa7bc0-e099-4577-bad9-d026e7994660
0 reply
0 recast
0 reaction
lhl pfp
lhl
@lhl
I finally got around to testing out the Keystone Pro I ordered and got deep into a rabbit hole of SLIP-39, BIP-39 roundtripping, SSKR etc. I ended up just making a new SLIP-39 seed for testing, but one thing I missed was Multi PIN support. Doing some research, will compare more vs Lattice and Ledger soon.
0 reply
0 recast
0 reaction
lhl pfp
lhl
@lhl
Presumably, you could have the DKMS w/ an updating shared secret via a double ratcheted gossip protocol. Membership would be revoked based on verification of a valid state? I haven't quite thought through it all and already getting pretty complex, so curious if smarter people have already given it some thought...
0 reply
0 recast
0 reaction
lhl pfp
lhl
@lhl
Recently I've been thinking about how one might go about enforcing privacy/deletions on a distributed social network. You would need to have posts encrypted, and a distributed key management system that would allow revocations - anyone know of any work being done on this?
1 reply
0 recast
0 reaction
lhl pfp
lhl
@lhl
Maybe of interest for those looking to run their own LLMs, I've compiled the most comprehensive list of 1B+ parameter models I've seen online, as well as evals, some fine tunes and datasets: https://docs.google.com/spreadsheets/d/1kT4or6b0Fedd-W_jMwYpb63e1ZR3aePczz3zlbJW-Y4/edit?usp=sharing
0 reply
0 recast
0 reaction
lhl pfp
lhl
@lhl
I've been pretty heads down doing deep dive on AI/LLMs this past month and keeping some notes. I'm sure it'll change over time but here's my current take on AI Safety xrisk: https://mostlyobvious.org/?link=/Reference%2FSoftware%2FGenerative%20AI%2FSafety
0 reply
0 recast
0 reaction
lhl pfp
lhl
@lhl
Here's a Tokyo food guide I made for a friend last fall https://lhl.notion.site/Tokyo-Food-Recommendations-0d3dd2d2be87403e83292bb4e640d667 and some general Japan travel tips I keep (focusing on transit, internet, etc): https://mostlyobvious.org/?link=%2FReference%2FTravel%2FJapan%20Notes
0 reply
0 recast
0 reaction
lhl pfp
lhl
@lhl
Spending most of my time recently with generative AI - training models in SD, playing around w/ LLMs, etc. I think I'm mostly caught up (although new products and breakthroughs keep flooding in), but good reminder that despite regulatory headwinds, crypto marches on: https://twitter.com/avsa/status/1638921979188191232
0 reply
0 recast
0 reaction
lhl pfp
lhl
@lhl
Authenticating my Farcaster account on @perl. [0x447ff8316749a3e1956dcea4198e2abe1cfe2a394047364667da03796559c39c6abe3f20077d0df33abed0ad452a09a552a056564d010773e42db3d6ff0198dc1c]
0 reply
0 recast
0 reaction
lhl pfp
lhl
@lhl
Btw brand new and enjoying the Farcaster community/vibe a lot but just fyi I’m mainly on Linux and Android these days (happy to test out builds). Also while I’m happy to treat fc as my cryptotwitter “list” I think actual List/topic support might be something I’d be keen on.
0 reply
0 recast
0 reaction
lhl pfp
lhl
@lhl
Maybe take a look at this and see if it does what you’re looking for: https://github.com/joeycastillo/twitter-archive-site
0 reply
0 recast
0 reaction
lhl pfp
lhl
@lhl
tap tap. Let’s see how this compares to lens and mastodon…
0 reply
0 recast
0 reaction