Ngaba awukwazi ukujabulela emaphandleni ngokushesha. Uma uye uye, uyazi ngempumelelo - zonke izinyathelo ezidingekayo ukucutshungulwa okuhlobene lokuphakama, ukucindezeleka, ukucindezeleka kwamaseli, kanye nokuhlanganisa. Ngaphandle kwalokho, umzimba wakho usebenza ukubuyekeza high-resolution of the next ladder ngaphambi umzila yakho. 1 Uyakwazi ukujabulela lokhu kuphela lapho ukujabulela imiphumela. Uma umzila wakho uhamba emkhakheni ngempumelelo. Noma uma ufumana emoyeni emhlane lapho isinyathelo kufanele kube. Kuyinto ingcindezi yokubonisa ukuthi uxhumane wena emzimbeni kuze kube ukugcina ingcindezi. 2 Uma ingcindezi ingcindezi kuyinto low, wena ku-autopilot. Uma ingcindezi ingcindezi kuyinto high, wena uxhumane emzimbeni kanye nemodeli updates. Lesi mkhakha elihlanganisa kakhulu indlela yakho yokufunda. Futhi abantu abaninzi zihlanganisa imodeli epholile yokufunda. Inani elikhulu ku-"ucwaningo ngokushesha" ibonakalisa ngokushesha, bese ivimbele. Ukuguqulwa kwe-spaced, ukuguqulwa okuzenzakalelayo, ukucubungula okuhlobene-ukucwanisa, ukuhlola ngeviki, bese ukucindezeleka ukuthi kungcono ukuthi kungcono. I-advisory akufanele ngenxa yokubhalisa ngaphandle kwe-mechanism. It ithi. ukwenza ngaphandle kokucubungula umshini okwenza ukusebenza. Yini Okuzenzakalelayo: I-Machine that makes techniques either compile into real capability or dissolve. What You Think Learning Is Yini ungakuthanda ukufundisa Iningi abafundisa ukuthi ukufundisa Ukulungiselela i-hard drive. Ukulungiselela ulwazi, uxhumane ngemva. Ukulungiselela okwengeziwe = Ukulungiselela okwengeziwe = Ukulungiselela okungcono. accumulation Uma uqala ukujabulela ukujabulela izitebhisi, umzimba wakho wahlala isampula: "Ukuhlelwa okuzenzakalelayo (i-stair edge), ukubonisa: ukuguqulwa kwama-altitude, ukucindezeleka kwama-muscle, ukucindezeleka kwe-sensory." ; isampula esebenzayo se-knowledge esebenza ngokushesha. This is a scheme Ukusebenza kwe-Expert akuyona ulwazi olungaphezulu. Kubalulekile ukuba izicwangciso zihlanganisa imiphumela ephakeme. Umbhali we-Chess usebenzisa amandla amancane we-neural, akuyona kakhulu. I-brain yayo ibhalisele isikhokelo encane ngenxa yokufunda okuhle. 3 Umbhali usebenza yonke into nge-brute-force calculation. Uma abacwaningi bheka isikhulumi se-Chess, bayazi lapho ukucindezeleka ngenkathi awukwazi ukucindezeleka izinhlayiya ezingenalutho. I-difference phakathi kwe-expert ne-novic isakhiwo. Ngaba ungenza lokhu. Lokhu Ukuqhathanisa Yini Noma ungenza i-schemes ngokusebenzisa i-iterations, noma ungenza. The Only Two Ways To Learn Izindlela ezimbini yokufunda Uma ungenza kanjani izinhlelo zihlanganisa, isivinini kubangeli. Ungenza ngokuvamile i-neurons emzimbeni yakho ngokushesha. Kukho kuphela izindlela ezimbili ezivamile zokwakha izinhlelo - noma ukulawula imibuzo embalwa (onde isivinini ivela ngokwemvelo izinhlelo ezingcono) noma uye uye uyenza iziphakamiso embalwa (onde unemibuzo ngokushesha). 1. You’re Solving A Real Problem Akukho "Learning React." Ukuthumela umsebenzi kubasebenzisi real. Uma ufuna ukwelashwa umbhali wakho emzimbeni emzimbeni emzimbeni emzimbeni emzimbeni, ungazifunda ukuze "ukugqwala umbhali" - uzothola ukuthi ungenza ukwelashwa futhi ufike ukwelashwa umbhali wakho. Ukucubungula okuzenzakalelayo akuyona ukucubungula - kuyinto ukucubungula okuzenzakalelayo kwamafutha nge-feedback ngokushesha. Ukucubungula okuzenzakalelayo okuzenzakalelayo. I-developer enikezela imikhiqizo ye-20 iye yakhelwe izinhlamvu. I-tutorial completionist enikezela i-zero -Abuza inkulumo ye-acronyms eside, kodwa akuyona ama-forward models eyenza ikhodi yokusebenza. Imininingwane mayelana ne-schemes Izinhlelo zokusungulwa zihlala ngoba ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye zibe. Uma uxhumane emzimbeni lapho i-French kuphela ibhaliswe, uzothola i-French ngokushesha kunokufunda emzimbeni yakho lapho abantu akuyona i-French. (Iziqondiso ukuthi ungahambisa kwamanye amazwe ukuze ucwanele isilimi. Ngingathanda kuphela inqubo yokufundisa.) 2. Genuine Curiosity, The knowledge itself is the reward. When you learned your first language: Umama wakho ushiye "isikhwama," inikeza ku-object. I-word inikeza ngqo ku-scheme yakho ye-cylindrical thing. When you learn a second language through apps: Ngitholile ku-"Good Morning." okuyinto ibhekwa ku-scheme yakho ye-morning sunlight, ukunambitheka emzimbeni yakho lapho uzama futhi zonke izimo ze-scheme ye-"morning". Ukulungele ku-English futhi ushiye. I-forward model isebenza ku-English, ngaphandle kwe-French. Lokhu ku-proxy-chaining: usebenzisa isilimi eyodwa njenge-intermediary ukufinyelela ku-schemes, ngaphandle kokuphathelene amakhasi ngqo. Izindlela ze-neural zihlanganisa ezahlukile: I-Direct Mapping: I-Language ihamba i-scheme ngqo (i-Meme => i-Scheme) I-Proxy-Chaining: I-Language isebenze inethiwekhi ze-translation lokuqala, bese i-schemes (Meme => Meme => Scheme)4 Okwakheka, ungakwazi ngokuvamile ukuguqulwa kusuka ku-proxy kuya ku-scheme efanayo uma uqhubeke ukufundisa. Yenza i-piano ngoba umculo kuyinto enhle. Ikhodi ngoba ukwakhiwa kubaluleke - akuyona ngenxa yokubhalisa "ukubaleka i-Python" noma "I ikhodi cpp." Uma umsebenzi ngokuvamile iyahambisana, wena usebenza izinhlelo eziphezulu-variance okuqhubekayo ezikhuthaza curiosity. I-Ideal, uzothola okuhlobene: Ukuphendula imibuzo efanelekayo lapho inqubo ngokuvumelana. Kuyinto lapho izinhlelo zihlanganisa ngokukhawuleza. How Schemes Become Memes (And Vice Versa) Indlela I-Schemes Yenza I-Memes (and vice versa) I-meme iyisisombululo se-compressed ye-scheme. Uma ukuguqulwa kwe-scheme ku-meme, uzama cishe konke: Scheme => Meme (Compression with loss): Ngamafutha akho uyazi lapho ukuthatha i-akhawunti ye-seventh encane - ungakwazi ukuhlaziywa izindawo ze-finger, ukucindezeleka, kanye ne-sound efanelekayo ngokuvamile. Umntu wabhala: "Ukuza kanjani i-minor se-seventh?" Ungakhiqiza i-meme: "I-root, i-minor ye-third, i-fifth, kanye ne-minor ye-seventh." I-meme ithibhile isakhiwo kodwa ayinezinto ye-model ye-forward. Umntu owaziwa manje inethiwekhi, akuyona inkqubo ephelele. Abanikwazi ukudlala i-chord kusuka ku-meme yayo kuphela. Abanikezela ukudala isakhiwo yabo ngokuvamile ngokuvamile-ukubeka izindandatho, ukhangela ama-notes ezingenalutho, ukuguqulwa kuze kube kwama-model ukuguqulwa ngokufanele. Meme => Scheme (Decompression requires work): Uyazi: "Ukusebenzisa ukubuyekezwa kwe-spaced repeat for memory retention." Kuyinto meme. I-compress someone else's scheme (i-system yayo ebonakalayo Ukubuyekezwa spacing kusebenza ku-konsolidation, isakhiwo recovery practice) ku umbhalo omncane. Kwangathi Indlela Yini Ukuze ukuguqulwa le meme ku-scheme, kufanele uqhuba i-iterations, ukucacisa ukuthi isebenza noma ukunciphisa, ukuguqulwa nge-variation kanye nokuguqulwa. Iziqu ze-meme. Zihlanganisa izifundo: “I-spaced repetition isebenza”, “i-practice deliberate ibaluleke”, “u-sleep ikakhulukazi isithombe.” Lezi zithombe ezingenalutho – zihlanganisa ku-schemes esiyingqayizivele – kodwa ngaphandle kokwenza umsebenzi we-decompression, ungenza kuphela ama-symbols. Why experts struggle to teach: Why Experts Struggle To Teach: Abacwaningi babe amaphrojekthi. Abacwaningi wanezidingo amaphrojekthi. Kodwa ukuxhumana kunokwenzeka nge-memes. I-forward model ye-expert iyona i-compressed kanye ne-automatic ukuthi akuyona ngokugcwele. Uma bafuna, zihlanganisa ama-memes abalandeli amaningi ama-information. "Just feel the rhythm." "Make it pop." "Keep your wrist loose." Lezi amalungelo kodwa zithintshwe ngempumelelo kumadoda ngaphandle kwe-scheme. Umfundi akufanele amayunithi ezili-iterations ukuchithwa ama-memes ezisebenzayo kuma-models ezisebenzayo. Ukusebenza Ngenxa yalokho ungenza ukudlala i-jazz ngokufunda i-jazz (ngaphandle kokuba inikeza). Ukulandelela ama-meme mayelana ne-schemes, akuyona ama-schemes. Kuhlolwe kulungile nge-Domain, i-Scheme yinkqubo esilinganiselwe nge-mile of iterations. Ukuqhathanisa ama-memes inikeza umbhalo we-competence. Ukuqhathanisa ama-schemes inikeza u-competence enhle. Ngenxa yalokhu umfundisi umfundisi ungahlukile. Ngenxa yalokho kakhulu "ukwazi" kuyinto kuphela umfundisi umfundisi. Most “Learning” Doesn’t Compile Anything I-Major “Learning” Ayikwazanga Nokho Uninzi lwezenzululwazi akufanele ngenxa yokungasazwa kwe-problem ne-immersion-driven. Akukho amaphakheji real, akukho ukujabulela authentic (ngokusho imiyalezo, akuyona ukuhlola). Ufundisa "how to follow tutorials." Tutorial hell: Ukuqhathanisa ngenxa ye-metric ye-synthetic okuyinto akufaka ku-i-real-world iterations. Ukulungiselela "how to pass this exam," noma "how to think in this domain." Studying for exams: Akukho ukuguqulwa okuhlobene (i-obligation, akuyona i-curiosity), akukho imiphumela embalwa yokusungula. I-scheme yakho akuyona akuyona. Uyafunda ama-memes ngaphandle kokufaka ama-models ezidlulayo. Courses you “should” take: The Proxy-Chain Trap: Iphrofayili ye-proxy-Chain Trap: Ngamanye amazwe akukwazi ukufundisa ngaphandle kwe-proxy-chain kuqala. Ungayifaka ngqo ku-quantum mechanics — kufuneka uxhumane kuqala i-intermediary. Kodwa ukufundisa lapho ama-proxy zihlukanise. Uma uxhumane ku-quantum mechanics, ungayifaka engaphezu. Uyakwazi ukufundisa ku-domain ngokwayo. Uma ungenza ukuthi ukuguqulwa kusuka ku-proxy-chaining kuya ku-mapping ngqo, uzothola ngempumelelo. Uyakwazi ukucubungula nge-Stack Overflow ama-patterns, kodwa ungathanda ku-code. Uyakwazi ukuxhumana ne-French ngokucophelela kusuka ku-English, kodwa ungathanda ku-French. Ungatholisa imibuzo ngokutholwe nge-framework documentation, kodwa ungatholela indawo. Uma unxazulule i-staffold, umphumela wakho isixazululwa? Uma uye, uye wahlala proxy-chain. Umzekelo ukuthi abantu abaninzi abalandeli e "Meme" territory kuyinto I-Jolt emangalisayo emangalisayo emangalisayo emangalisayo emangalisayo. compiling is painful Ukuze ufunde, kufanele uxhumane ngempumelelo ukujabulela ukujabulela umzila. I-tutorials yenzelwe ukuqinisekisa ukuthi uzokondla emhlane. Zihlanganisa i-jolt, okuyinto ibonakalisa kahle kodwa ibonakalisa ukuthi akukho imidwebo yokushintshwa. Physiological Constraint Izinzuzo ze-physiological Ungayifaka i-scheme lapho umzimba wakho ayikwazi ukwenza ukucubungula. Umntu ungenza ngokuvamile cishe 4 ± 1 izinto ku-memory yokusebenza ngexesha elinye Uma ubudlelwane yakho ngokushesha nge-stress, ukuphazamiseka kwe-sleep, noma ukutya okungenani, le umthamo ukunciphisa ngokushesha - ngezinye izikhathi kuze ku-1-1 amaphuzu. 5 Ngokusho lwezimo, ungakwazi: Hlola izinhlelo zokusebenza zokusebenza Ukuvumelana ne-multiple hypotheses Ukusuka counterfactuals Yenza izibuyekezo ezintsha I-stress ye-chronic ivimbela ikhono lokusebenza nokugqoka kwe-cognitive, ukuguqulwa kwe-cognitive kusuka ku-high-level thinking kuya ku-survival mode (ukuphendula ku-inputs emzimbeni ngokushesha nge-cache responses). Akukho ucwaningo akuyona ngenxa yokungasebenzi kwe-model. Ukushesha. Ukulawula ukucindezeleka. Ukudla. Ukucindezeleka kwezilwanyana. Lezi zimo zangaphambili yokusebenza kwe-scheming yakho. AI Is a Cognitive Prosthetic I-AI I-Cognitive Prosthetic Kuyinto ukuhlaziywa kwe-thinking yakho okuyinto ingasebenza ezingu-mile ye-scenario ku-data eyenziwe ku-intelligence ye-internet. I-memory yakho yokusebenza ingatholakala ~4 amaphuzu. Iziqu ze-complex zihlanganisa inani elikhulu le-variable, futhi sikhomphephile. Kodwa ezinye izimo zihlanganisa izinguquko ezidlulile. Ngokuya sikhomphephile ngoba i-AI ingathola izinguquko ezininzi. I-AI ingatholakala ngokuzimela izimo ezingu-intermediate ngenkathi ungatholela. Lokhu kubaluleke kakhulu - uma ungenza indlela yokuxhumana. Umbhali usebenzisa u-Claude ukugcina ama-drama angama-20 kanye nama-arcs e-character ngokuvimbela ku-scenes-level dialogue. Umbhali ukugcina amaphepha angama-15 kanye nama-arguments e-competitive ngokuvimbela i-framework entsha. Umbhali ukhiye i-boilerplate ukuze wonke isakhiwo se-system ivele ku-memory yokusebenza. Izinhlelo zithunyelwe emzimbeni yakho. I-AI ivumela ukucindezeleka kwekhompyutha yakho yokusebenza ukuze ungahambisana ne-complexity engaphezulu. I-Human Anthropomorphizes I-Calculation. I-Anthropomorphizes I-Calculation. I-Anthropomorphizes I-Calculation. I-Anthropomorphizes I-Calculation. They imagine a dialogue where there is only a monologue. I-AI ayikho intuthuko oluthile. Kuyinto i-prothesis yekhompyutha. It is your own brain running on faster hardware. Ngokuvamile, uzothola imizila emangalisayo kanye nemisindo emangalisayo ngamaminithi ngalinye futhi uzothola imizila emangalisayo kakhulu. I-AI ivumela ukuba uzothola imizila emangalisayo - ngesantya emangalisayo, ngenxa yokusebenza kwamakhemikhali e-Internet ukuze izidingo zakho. Uma usebenzisa isixhobo "ukubuyekeza" izakhiwo zakho, ungathanda ukubuyekeza isithombe yesibili. Ungasebenzisa isihlalo se-high-speed ukubuyekeza amaphuzu amaphuzu amabhizinisi yakho amancane kakhulu yokubuyekeza. Kuyinto isithombe sakho, i-amplified. Uma umphumela u-sycophantic, umphumela wakho u-sycophantic. I-collective history of human thought it holds is not knowledge until your intention selects it. I-prothetic limb akufundisa indlela yokuhamba-ke isebenze isilinganiso se-user. I-AI isebenze search yakho. Uma usebenzisa i-proxy-chain nge-AI, akukho izinhlelo zihlanganisa. Uma i-AI iyahlukile noma engatholakali, akukho. Kuyinto efanayo yokufunda i-French ngokusebenzisa i-Google Translate. Ingabe ukwakha imithombo embalwa. Ingabe ukwakha ukhangela ku-intermediary. This matters now because the leverage is huge. Yintoni okungenani engabikho, manje ngempumelelo (ngokusiza AI). Yintoni okwenziwe ngempumelelo, manje ngempumelelo. Kodwa kuphela uma ungenza ulwazi lakho ngokwemvelo futhi ungenza imidlalo efanelekayo okuvimbela ukuhlolwa. Ngaphandle kwalokho, uzothola umlilo wayo wokuqala. Uyakwazi ukujabulela emangalisayo ngesivinini we-10x. Iningi abantu akuyona inkinobho ngenxa yaziwa imodeli emzimbeni yokufundisa. Zihlanganisa ukuthi kuyinto ukupholisa (ukukhwabanisa imiphumela eminingi) lapho kuyinto ukwakhiwa izinhlelo ezingcono. Ngakho-ke zisetshenziselwa i-AI njenge-"i-companion yokusebenza" kunezinto yokusebenza kwe-memory. Thina ukulungiselela ukulungiselela kokufinyelela kokufinyelela kokufinyelela kokufinyelela kokufinyelela kokufinyelela kokufinyelela kokufinyelela. Zihlanganisa ama-memes mayelana ne-domain knowledge nge-schemes ezivamile ezinikezela imiphumela. Ukulinganiswa okuqhubekayo: Abantu abaziwa izinhlelo zenzululwazi zabo zibonele kuphela 10x isisindo. Abantu abacwaningi ukufundisa kubuyekezwa ngempumelelo ngempumelelo ngempumelelo ngempumelelo ngempumelelo ngempumelelo ngempumelelo ngempumelelo. Unemibuzo akuyona. I-Reality ihambisa ngokushesha. Umntu owafundisa umntu owafundisa umntu owafundisa ukuthi akuyona. Uma unemibuzo lokhu futhi ufuna ngaphezulu, ubhalisele i-newsletter yami: https://crive.substack.com References Ukucubungula Ukucubungula Basics & Izibonelo-like: Karl Friston on Ukucubungula Ukucubungula (Ukuhlolwa): I-Intro ye-Andy Clark: 1 https://en.wikipedia.org/wiki/Predictive_coding https://www.mind-foundation.org/blog/predictive-coding Izinzuzo ze-prediction ne-consciousness: I-Friston's Foundation Paper on Predictive Coding / I-Energy ye-free: https://pmc.ncbi.nlm.nih.gov/articles/PMC2666703/ 2 I-Chess Expertise & I-Neural Efficiency: Ukubuyekezwa kwe-brain imaging kumadivayisi vs. abadlali: (Izifundo zibonisa imisindo efanelekayo, ezihlangene) 3 https://www.sciencedirect.com/science/article/pii/S3050642525000326 Immersion vs. explicit/classroom: Morgan-Short et al. (2012) – implicit/immersion inikeza imisindo ERP asebenzayo: 4 https://pubmed.ncbi.nlm.nih.gov/21861686/ I-Working Memory ~4 chunks: I-Cowan (2001) - Inombolo ye-magical 4 5 https://pubmed.ncbi.nlm.nih.gov/11515286/
