[{"data":1,"prerenderedAt":192},["ShallowReactive",2],{"project-38392260":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":14,"tier":18,"category":19,"subcategory":20,"format":21,"keywords":22,"reception":28,"sentiment":29,"stance_mix":30,"scored_comments":34,"aspects":35,"strip":37,"post_text":29,"what":39,"said":66,"maker_in_thread":126,"shape":131,"grid":134,"attempts":156,"same_week":157,"maker_shows":153,"maker_hidden":185,"reader_version":186,"card":187},"38392260","Open Australian Legal GPT2","The first open LLM for Australian law","title","Show HN: I built the first open LLM for Australian law","https:\u002F\u002Fumarbutler.com\u002Fhow-i-built-the-first-open-llm-for-australian-law\u002F","umarbutler.com","ubutler",1700744331,2023,40,9,1217,3109,"shown","ai","legal language model","other",[23,24,25,26,27],"australian law","llm","gpt2","legal ai","open model","quiet",null,{"positive":31,"neutral":32,"mixed":33,"critical":33},1,2,0,3,{"idea":36},{"positive":31,"critical":33},[33,34,33,38,34,33,33,33,33],5,{"problem":40,"audience":29,"pricing":29,"motivation":50,"time_to_build":29,"stack":54,"openness":63,"stage":64,"team_size":65},{"value":41,"source":42,"evidence":43},"Lack of an open language model with deep understanding of Australian law","maker",[44,47],{"id":45,"by":11,"text":46},"38392266","Hey HN, Last month, I had the honour of seeing my article on how I built the largest open database of Australian law…",{"id":48,"by":11,"text":49},"38399702","> 1) What’s the main use case for this? Originally, I intended to produce an LLM that could be used for answering questions related to Australian law, potentially via retrieval…",{"value":51,"source":42,"evidence":52},"Build on the author's Australian law database and push the legal AI field forward in Australia",[53],{"id":45,"by":11,"text":46},[55,59],{"value":56,"source":42,"evidence":57},"GPT2",[58],{"id":45,"by":11,"text":46},{"value":60,"source":42,"evidence":61},"Hugging Face",[62],{"id":45,"by":11,"text":46},"open_source","launched","solo",{"praise":67,"objections":76,"requests":84,"questions":96,"alternatives":109},[68],{"value":69,"source":70,"evidence":71},"Interest in legal AI for making Australian legislation searchable","commenter",[72],{"id":73,"by":74,"text":75},"38400925","gitgud","Is there a link to try it? This is literally the main application of ML that I’ve been awaiting for years. Making complex legislation and bureaucracy searchable and useful for…",[77],{"value":78,"source":42,"evidence":79},"Model is too small to beat GPT-4 on Australian law questions",[80,81],{"id":48,"by":11,"text":49},{"id":82,"by":11,"text":83},"38400970","See my response to \u002Fu\u002Fnextworddev (https:\u002F\u002Fnews.ycombinator.com\u002Fitem?id=38399702). In short, GPT-4 outperforms my LLM simply because my model is quite small (its smallest version…",[85,89],{"value":86,"source":70,"evidence":87},"A link to try the model",[88],{"id":73,"by":74,"text":75},{"value":90,"source":70,"evidence":91},"Comparison against GPT-4 with retrieval augmented generation",[92],{"id":93,"by":94,"text":95},"38400154","vermaat","How do you compare your own trained LLM versus using for example GPT4 + RAG (Vector DB + your Australian Law DB?",[97,104],{"value":98,"source":70,"evidence":99},"What is the main use case for the model",[100],{"id":101,"by":102,"text":103},"38399383","nextworddev","Thanks for sharing. Two questions: 1)What’s the main use case for this? 2)Was training your own LLM necessary? How did GPT-4 compare?",{"value":105,"source":70,"evidence":106},"Was training a custom LLM necessary versus GPT-4",[107,108],{"id":101,"by":102,"text":103},{"id":93,"by":94,"text":95},[110,115,119],{"value":111,"source":42,"evidence":112},"GPT-4",[113,114],{"id":48,"by":11,"text":49},{"id":82,"by":11,"text":83},{"value":116,"source":42,"evidence":117},"GPT 3.5 Turbo, Claude 2, Bard, Llama 2, Mistral",[118],{"id":48,"by":11,"text":49},{"value":120,"source":70,"evidence":121},"DevonThink",[122],{"id":123,"by":124,"text":125},"38393261","Obscurity4340","What do you think about DevonThink?",{"comments":34,"answered":32,"key":127},[128,129,130],{"id":45,"text":46},{"id":48,"text":49},{"id":82,"text":83},{"commenters":132,"max_depth":34,"maker_replies":34,"first_comment_mins":31,"lifespan_hours":133},7,30,[135,137,139,140,142,143,145,149,151],[45,33,33,136,11,31],"project\u002Fneutral\u002Fnone",[101,33,34,138,102,33],"project\u002Fneutral\u002Fidea",[48,33,33,138,11,31],[73,31,38,141,74,33],"project\u002Fpositive\u002Fidea",[93,32,34,138,94,33],[82,32,33,144,11,31],"project\u002Fneutral\u002Fperformance",[146,32,33,147,148,33],"38406874","commenter\u002Fneutral\u002Fidea","willy_k",[123,34,33,150,124,33],"other\u002Fneutral\u002Fnone",[152,153,33,154,155,33],"38406221",4,"other\u002Fneutral\u002Fidea","RecycledEle",[],[158,162,167,172,176,180],{"id":159,"name":160,"points":161,"category":19},"38432486","dalle.party",587,{"id":163,"name":164,"points":165,"category":166},"38434574","Trains.fyi",503,"infrastructure",{"id":168,"name":169,"points":170,"category":171},"38413660","fun-with-sockets",403,"creative",{"id":173,"name":174,"points":175,"category":19},"38453047","Dobb·E",394,{"id":177,"name":178,"points":179,"category":166},"38409164","IP Guide",363,{"id":181,"name":182,"points":183,"category":184},"38390714","React Spreadsheet 2",356,"dev_tools",false,"v2",{"id":188,"name":5,"line":6,"ai":185,"cat":19,"year":13,"points":14,"comments":15,"rec":28,"s":29,"top":14,"kw":189,"maker":11,"domain":10,"tier":18,"attempts":33,"strip":190,"url":9,"mix":191,"praise":69,"objection":78,"sub":20},38392260,[23,24,25,26,27],"030530000",[31,32,33,33],1793991078640]