Hey there ッ
Moon
Joined September 2021
- Tweets358
- Following544
- Followers42
- Likes2.6K
Jacob retweeted
(1/7)📢 New paper! Speech AI is getting better at hearing and speaking many languages. But what if the input mixes two languages, or is in one the model never learned? It guesses one language it knows and imposes that language's sounds on everything.
We find why, and fix it 🧵
PocketTTS now speaks Hebrew 🎉
LoRA fine-tune on IPA phonemes, supports any espeak language.
100% in-browser instant streaming, no GPU.
Higher quality than Kokoro, faster than Piper.
Built on @kyutai_labs
Try it: huggingface.co/spaces/thewh1…
Makefiles and npm scripts break the first time someone clones on Windows, so I built chore: a task runner with its own portable shell baked in.
Builtins for download, extract, archive, copy, sha256. chore check flags non-portable commands.
github.com/getchore/chore 🧵
chore check parses the chorefile without running anything and flags every command that won't exist on someone else's machine — with the builtin to swap in.
Works for coding agents too: they write curl | tar by reflex, chore check catches it, they fix it themselves from the hint.
I wanted this for years, so I finally built it.
Like ngrok, but peer-to-peer.
punch 3000
Share a local app through a plain browser URL. No server in the middle, no account, no config. The link connects straight to your machine over WebRTC, end-to-end encrypted.
punchpage.pages.dev
Open source: github.com/thewh1teagle/punc…
Jacob retweeted
Took us a minute, but stacked PRs are now on GitHub 🥞
github.blog/changelog/2026-0…
Jacob retweeted
We also worked with academics at ETH Zurich, Tel Aviv University, and the University of Haifa to build CryptanalysisBench, a benchmark for studying LLMs’ cryptanalysis abilities.
arxiv.org/abs/2607.18538
Jacob retweeted
Kimi K3 open weights drop today.
For 11 days the best open source model in the world has been stuck behind one provider.
Good enough to beat Fable 5 in design. Too slow to actually build with.
That ends today.
2.8 trillion parameters, free on Hugging Face. Every fast inference provider gets a copy.
The moment someone serves K3 at real speed, it becomes the best value in vibe coding.
I will be benchmarking every K3 endpoint that goes live this week.
Jacob retweeted
For my first post, I’m sharing a letter @NVIDIA signed on why open models matter.
AI will transform every industry, power every company, and be built by every country.
Open models strengthen safety and cybersecurity, accelerate innovation and diffusion, and enable sovereignty.
The world needs both frontier closed models and frontier open models.
images.nvidia.com/pdf/Open-W…
I connected Alexa + ChatGPT Live + Fable 5. It genuinely feels like talking to Einstein in your living room.
Another win for open source!
Just got accepted into Anthropic's Claude for Open Source program. Thanks @AnthropicAI for supporting open source maintainers ❤️
This led to a follow-up project with a bigger idea: ReNikud. What if Hebrew pronunciation could be learned directly from audio, without manual nikud labels? We run Hebrew ASR + IPA ASR on 1,700 hours of speech, keep only transcripts that agree, and train on 1.5M pseudo-labeled sentences with everyday spoken Hebrew including slang, loanwords, and colloquial forms. 🧵
Everything is open: paper, code, models, and ILSpeech, a studio-quality Hebrew TTS dataset with IPA annotations.
📄 Paper: arxiv.org/abs/2506.12311
🌐 Project: phonikud.github.io
🔊 Follow-up (ReNikud): renikud.github.io
🤗 Data: huggingface.co/Phonikud
Huge thanks to @maxmelichov, @oyd11, and @MorrisAlper 🙏 More soon :)
#interspeech2026
