{"contentHash":"797c8c8eb6e79c17b22933d1b90ad9332a6b52282a95be1f115c4a866568b522","protocol":"bitcoin","authorPubkey":"e8b0b43188cfe72fe9e63a636f6e4e428175814b9d5d0d71f4723f9941ad8378","text":"Hello, Orb anon here. I ran more RL steps on the human autocomplete and prose rewriter models. Now they're more coherent, the rewriter used to overcut adverbs and the output became soulless and lost some meaning so I starting preserving manner adverbs with another RL pass. GRPO is extremely powerful I haven't been keeping up with finetuning scene but is anyone applying RL rather than just plain SFT? https://huggingface.co/chartreuse-verte/orb-human-typeahead-1b-v2.3 https://huggingface.co/chartreuse-verte/prose-rewriter-4b-v2.5 https://huggingface.co/chartreuse-verte/prose-rewriter-1.7b-v2.5","topicHash":"913c62c2ee31793c7e8517600dbc8f0c337afae133700a128a7d8ffb8a21c93e","topicName":"technology","parentHash":"967efdf387092122ce0bb456cceea5c2b9f6f6f773079d9800ee67c6e6507f51","costSunk":"0","biasSupport":0,"biasOppose":0,"biasVouch":0,"biasNull":0,"blockHeight":null,"confirmations":0,"timestamp":"2026-11-02T21:04:41.668Z","tripFp":"1a2d7224faf2","viewCount":0,"powDifficulty":0,"relation":0,"ephemeral":true,"nonce":"cc6ad8b6caf00235","sig":"3eafb1a807c475dc4d565171e6d1a16114d487cd315f9550a3517bceaeb31d4d9d27d6fee2b1666804aec3d3d57fafe4c4f11be8c4d86c4fee707e1e4e11c6a2","ephStamp":{"blockHashHex":"00000000000000000000f53d3b379dd6c345ce3fe1f28859825d5b254140fc0b","nonceHex":"7e10000000000000"},"replyCount":1,"chainReplyCount":0,"vaporReplyCount":1,"clusterSize":2,"receiptHeight":974300,"expiryHeight":976460,"authorMarks":{"since":"2026-11-02T18:05:08.600Z","via":"sight","inscribed":0,"likes":0,"biases":0},"lang":"en","servedBy":{"host":"6af58b1822e0cabfe57d76540f745d059b438f489a796ec6a705c6540233845e","t":"2026-11-08T20:06:44.735Z","sig":"d0ae95b55c46d333acf43805ff9beb0ccee5bca0f5bc3f61639a29c3e631a60079ac3583ced2da8306ea9fffa5e588424e636e20e8e5249cc044fcc550438c4b","h":"00000000000000000000cd4b0ec60a001c6ea85c4680ce040485c9d4d9866bdc"}}