[{"data":1,"prerenderedAt":166},["ShallowReactive",2],{"project-39770118":3},{"id":4,"name":5,"line":6,"line_source":7,"title":8,"url":9,"domain":10,"maker":11,"posted_at":12,"year":13,"points":14,"comments":15,"rank_in_year":16,"shown_in_year":17,"top":18,"tier":19,"category":20,"subcategory":21,"format":22,"keywords":23,"reception":29,"sentiment":30,"stance_mix":31,"scored_comments":35,"aspects":36,"strip":38,"post_text":41,"what":42,"said":57,"maker_in_thread":99,"shape":108,"grid":110,"attempts":129,"same_week":130,"maker_shows":15,"maker_hidden":159,"reader_version":160,"card":161},"39770118","Anything To JSON","a language model for structured extraction","title","Show HN: Anything To JSON – a language model for structured extraction","https:\u002F\u002Fanythingtojson.com\u002F","anythingtojson.com","edunteman",1710958289,2024,18,8,1990,3515,57,"shown","ai","structured extraction model","web_app",[24,25,26,27,28],"structured extraction","json extraction","language model","constrained generation","structured output","quiet",null,{"positive":32,"neutral":33,"mixed":34,"critical":33},2,1,0,4,{"idea":37},{"positive":33,"critical":33},[39,34,40,34,33,34,39,34],5,3,"Hey HN! Erik here from banana.dev\n\nWe’ve trained a small(ish) language model on structured extraction, and today we’re launching a playground for it at https:\u002F\u002Fanythingtojson.com. Give it a try!\n\nThis model continues our work on structured generation, following last week’s launch of Fructose[1], a python client for strongly-typed LLM responses.\n\nThere seem to be two distinct halves of the problem intended to be solved by Fructose and structured generation:\n\n1. the reasoning ability of the model, such as performing chain of thought, creative acts, and natural language tasks. In a way, the “busi",{"problem":43,"audience":30,"pricing":30,"motivation":49,"time_to_build":30,"stack":53,"openness":54,"stage":55,"team_size":56},{"value":44,"source":45,"evidence":46},"Getting reliable, schema-correct JSON from text without breaking receiving code","maker",[47],{"id":48},"post",{"value":50,"source":45,"evidence":51},"Handle the JSON-structuring part of LLM tasks so receiving code doesn't break",[52],{"id":48},[],"unknown","launched","company",{"praise":58,"objections":67,"requests":81,"questions":89,"alternatives":94},[59],{"value":60,"source":61,"evidence":62},"Chaining a small model for JSON structure saves the big model's attention for reasoning","commenter",[63],{"id":64,"by":65,"text":66},"39770818","yeldarb","Nice, makes sense to chain models so you don't waste the attention of the big smart model on the grunt work of JSON structure. I bet there are some good post-processing…",[68,75],{"value":69,"source":61,"evidence":70},"Small model cannot match large closed-source models",[71],{"id":72,"by":73,"text":74},"39770615","Oras","How does it differentiate from using json output with OpenAI? I think it can be useful for some cases, but you’ll always have the limitation of the model size to compete against…",{"value":76,"source":45,"evidence":77},"Hallucinates values for missing fields, such as inventing a name",[78],{"id":79,"by":11,"text":80},"39770866","interesting idea! yeah next big challenge for us is hallucination in null fields (IE: if you ask for a \"name\" from text that doesn't have a name you usually get \"John\") so need…",[82],{"value":83,"source":61,"evidence":84},"Support for image inputs",[85],{"id":86,"by":87,"text":88},"39771231","1997","Awesome, agree on the nulls! Any plans to add support for image inputs?",[90],{"value":91,"source":61,"evidence":92},"How it differs from OpenAI JSON mode",[93],{"id":72,"by":73,"text":74},[95],{"value":96,"source":61,"evidence":97},"OpenAI JSON mode",[98],{"id":72,"by":73,"text":74},{"comments":35,"answered":32,"key":100},[101,102,105],{"id":79,"text":80},{"id":103,"text":104},"39771715","we have no plans for image or other modalities. I'd like to keep it just text-to-text so it can be as sharp of a tool as possible",{"id":106,"text":107},"39770752","openai json mode will ensure JSON, but not strictly the right JSON schema to match whatever object you're expecting to return. Keys could be missing, fields could be improperly…",{"commenters":39,"max_depth":35,"maker_replies":35,"first_comment_mins":109,"lifespan_hours":32},30,[111,113,115,117,119,121,122,126],[64,34,39,112,65,34],"project\u002Fpositive\u002Fidea",[79,34,34,114,11,33],"project\u002Fneutral\u002Fperformance",[86,34,40,116,87,34],"project\u002Fneutral\u002Fidea",[103,34,34,118,11,33],"maker\u002Fneutral\u002Fidea",[72,33,33,120,73,34],"project\u002Fcritical\u002Fidea",[106,33,34,116,11,33],[123,32,39,124,125,34],"39770423","maker\u002Fpositive\u002Fnone","edwinwee",[127,32,34,128,11,33],"39770441","commenter\u002Fneutral\u002Fnone",[],[131,135,140,145,150,154],{"id":132,"name":133,"points":134,"category":20},"39706004","Skyvern",422,{"id":136,"name":137,"points":138,"category":139},"39759325","jnv",373,"dev_tools",{"id":141,"name":142,"points":143,"category":144},"39756325","Godspeed",328,"productivity",{"id":146,"name":147,"points":148,"category":149},"39714053","algebraic-nnhw",310,"hardware",{"id":151,"name":152,"points":153,"category":139},"39770908","GritQL",280,{"id":155,"name":156,"points":157,"category":158},"39755858","Causal 2.0",261,"web_saas",false,"v2",{"id":162,"name":5,"line":6,"ai":159,"cat":20,"year":13,"points":14,"comments":15,"rec":29,"s":30,"top":18,"kw":163,"maker":11,"domain":10,"tier":19,"attempts":34,"strip":164,"url":9,"mix":165,"praise":60,"objection":69,"sub":21},39770118,[24,25,26,27,28],"50301050",[32,33,34,33],1793991087079]