
ãå®å šä¿åçãçæAIã»LLMã®å°éçšèª50éžïŒäŸã話ã§äžçäžãããããã解説
â AGIïŒæ±çšäººå·¥ç¥èœïŒ
â¡ ãã«ã·ããŒã·ã§ã³ïŒå¹»èŠïŒ
⢠ããã³ãããšã³ãžãã¢ãªã³ã°
⣠ãã«ãã¢ãŒãã«
†ããŒã¯ã³
⥠ãªãŒãã³ãœãŒã¹ã¢ãã« ãš åçšïŒã¯ããŒãºãïŒã¢ãã«
⊠SLMïŒå°èŠæš¡èšèªã¢ãã«ïŒ
â§ AIãšãŒãžã§ã³ã
âš RLHFïŒäººéã®ãã£ãŒãããã¯ããã®åŒ·ååŠç¿ïŒ
â© ãã£ãŒããã§ã€ã¯
⪠ãã£ãã¥ãŒãžã§ã³ã¢ãã« (Diffusion Model / æ¡æ£ã¢ãã«)
â« LoRA (Low-Rank Adaptation / ããŒã©)
⬠ããã³ããã€ã³ãžã§ã¯ã·ã§ã³ (Prompt Injection)
â åºç€ã¢ãã« (Foundation Model / ãã¡ãŠã³ããŒã·ã§ã³ã¢ãã«)
â® ãã³ãããŒã¯ (Benchmark)
⯠Hugging Face (ãã®ã³ã°ãã§ã€ã¹)
â° API (Application Programming Interface)
â± GPU (Graphics Processing Unit / ç»ååŠçåå°äœ)
â² ãã¯ãã«ããŒã¿ããŒã¹ (Vector Database)
â³ LangChain (ã©ã³ã°ãã§ãŒã³)
ã Transformer (ãã©ã³ã¹ãã©ãŒããŒ)
ã NPU (Neural Processing Unit / ç¥çµç¶²åŠçè£ çœ®)
ã ããŒã«ã«LLM (Local LLM)
ã Temperature (ãã³ãã©ã㣠/ 枩床ãã©ã¡ãŒã¿)
ã ãŒãã·ã§ãã / ãã¥ãŒã·ã§ãã (Zero-shot / Few-shot)
ã ã°ã©ãŠã³ãã£ã³ã° (Grounding)
ã ã¢ã©ã€ã¡ã³ã (Alignment)
ã Copilot (ã³ãã€ããã)
ã åæããŒã¿ (Synthetic Data)
ã é³å£°ã¢ãŒãã« (TTS / STT)
ã Chain of Thought (CoT / æèã®é£é)
ã Instruction Tuning (ã€ã³ã¹ãã©ã¯ã·ã§ã³ã»ãã¥ãŒãã³ã° / æç€ºãã¥ãŒãã³ã°)
ã DPO (Direct Preference Optimization)
ã GraphRAG (ã°ã©ãRAG)
ã ããã³ãããã£ãã·ã³ã° (Prompt Caching)
ã± KVãã£ãã·ã¥ (KV Cache)
ã² ãªãŒãã³ãŠã§ã€ã (Open Weights)
ã³ VLM (Vision-Language Model / èŠèŠèšèªã¢ãã«)
㎠ãžã§ã€ã«ãã¬ã€ã¯ (Jailbreak / è±ç)
ãµ AIãŠã©ãã·ã¥ (AI Wash)
ã¶ ãšãŒãžã§ã³ãã£ãã¯ã»ã¯ãŒã¯ãã㌠(Agentic Workflow)
ã· ReAct (Reasoning and Acting)
ãž ãã«ããšãŒãžã§ã³ã (Multi-Agent)
ã¹ ã»ã«ããªãã¬ã¯ã·ã§ã³ (Self-Reflection / èªå·±åç)
㺠æšè«ã¢ãã« (Reasoning Model)
ã» End-to-End ãã«ãã¢ãŒãã« (ãã€ãã£ãã»ãã«ãã¢ãŒãã«)
㌠PEFT (Parameter-Efficient Fine-Tuning / ããã)
㜠Ollama (ãªã©ã)
㟠Top-p (ãããããŒ)
ã¿ ã¬ããããŒã (Red Teaming)
1. AGIïŒArtificial General Intelligence / æ±çšäººå·¥ç¥èœïŒ
äžèšã§èšããš: 人éãšåãããã«ãã©ããªã¿ã¹ã¯ã§ãããªãããäžèœã®AIãã
解説: çŸåšã®AIã¯ãæç« äœæããç»åçæããªã©ç¹å®ã®äœæ¥ã«ç¹åããŠããŸãããAGIã¯äººéã®ããã«èªãåŠã³ãèããæªç¥ã®èª²é¡ã«ã察å¿ã§ããã¬ãã«ã®AIãæããŸããçŸåšã®AIéçºäŒæ¥ïŒOpenAIãªã©ïŒãæçµç®æšãšããŠæ²ããŠãããã®ã§ãã
äŸãããªã: æ ç»ã«åºãŠããããã©ããããããã¿ãŒãããŒã¿ãŒãã®ããã«ãèªæèãæè»ãªæèãæã€ããããã®é è³ã
2. ãã«ã·ããŒã·ã§ã³ïŒHallucination / å¹»èŠïŒ
äžèšã§èšããš: AIããã£ãšããããã€ããåïŒç¥ã£ããã¶ãïŒãã
解説: AIãäºå®ãšã¯ç°ãªãæ å ±ããååšããªãæ¶ç©ºã®åºæ¥äºãããŸãã§æ¬åœã®ããšã®ããã«å ã ãšåºåããŠããŸãçŸè±¡ã§ããAIã¯ãäºå®ãã©ãããã§ã¯ãªããæåã®ç¹ãããšããŠèªç¶ãããèšç®ããŠããããã«èµ·ãããŸãã
äŸãããªã: ç¥ããªãè¡ã®éãèãããã®ã«ããã©ã€ããéªéãããŠããã£ã¡ã®è§ãæ²ãããšçŸå³ãããã³å±ããããŸããããšåã®æ¡å ãããŠããŸã人ã
3. ããã³ãããšã³ãžãã¢ãªã³ã°
äžèšã§èšããš: AIããè¯ãåçãåŒãåºãããã®ãæç€ºã®åºãæ¹ã®å·¥å€«ãã
解説: AIïŒç¹ã«LLMïŒã«å¯Ÿããåœä»€æïŒããã³ããïŒã®æžãæ¹ã工倫ããæè¡ã§ãããåææ¡ä»¶ãäžããããã¹ããããã€ã¹ãããã§èããããããªã©ã®æç€ºãäžããããšã§ãAIã®åç粟床ãåçã«åäžããŸãã
äŸãããªã: æ°å ¥ç€Ÿå¡ã«å¯ŸããŠãããæãã«è³æäœã£ãŠãããŠããšé Œãã®ã§ã¯ãªãããã¿ãŒã²ããã¯30代ãç®çã¯ç€Ÿå ææ¡ãA4çšçŽ1æã§ããšå ·äœçã«æç€ºãåºãéšäžãžã®æç€ºåºãã®æè¡ã
4. ãã«ãã¢ãŒãã«ïŒMultimodalïŒ
äžèšã§èšããš: ããã¹ãã ãã§ãªããç»åãé³å£°ãåç»ããåæã«çè§£ã»åŠçã§ãããèœåã
解説: åŸæ¥ã®AIã¯ãããã¹ãã ããããæ±ããŸããã§ããããæè¿ã®AIïŒGPT-4oãGeminiãªã©ïŒã¯ãåçãèŠããªãããããäœïŒããšèããããé³å£°ã§äŒè©±ããããåç»ãèªã¿èŸŒãŸããŠèŠçŽããããã§ããŸããè€æ°ã®çš®é¡ã®ããŒã¿ïŒã¢ãŒãã«ïŒãæ±ãããšããæå³ã§ãã
äŸãããªã: ç®ãäžèªç±ã§è³ïŒããã¹ãïŒã ãã§ããåãããŠãã人ããç®ïŒç»åïŒãèŠããè³ïŒé³å£°ïŒãèãããããã«ãªããäºæã䜿ã£ãŠäžçãçè§£ã§ããããã«ãªã£ãç¶æ ã
5. ããŒã¯ã³ïŒTokenïŒ
äžèšã§èšããš: AIãæç« ãèªèãããšãã®ãæå°ã®æåã®ã«ã¿ããªïŒåäœïŒãã
解説: AIã¯æç« ãã1æåãã€ãã§ã¯ãªãããããŒã¯ã³ããšããç¬ç¹ã®åäœã«åºåã£ãŠåŠçããŸããè±èªã ãš1åèªã1ããŒã¯ã³ã«è¿ãã§ãããæ¥æ¬èªã®å Žåã¯ã挢å1æåã1ã2ããŒã¯ã³ãã«ãªããªã©ãå°ãè€éã§ããAIã®å©çšæéããäžåºŠã«æ±ããæåæ°ïŒã³ã³ããã¹ããŠã£ã³ããŠïŒã®èšç®ã«äœ¿ãããŸãã
äŸãããªã: æç« ããã®ãŸãŸèªãã®ã§ã¯ãªããåèªãæåã®ããŒãããšã«ããºã«ã®ããŒã¹ã®ããã«åè§£ããããã®ãããŒã¹ã®æ°ãã
6. ãªãŒãã³ãœãŒã¹ã¢ãã« ãš åçšïŒã¯ããŒãºãïŒã¢ãã«
äžèšã§èšããš: AIã®èšèšå³ããå šå¡ã«ç¡æå ¬éããŠããã¢ãã«ããšãäŒæ¥ãç§å¯ã«ããŠããã¢ãã«ãã
解説:
ãªãŒãã³ãœãŒã¹ïŒLlamaãGemmaãªã©ïŒ: èšèšå³ïŒãŠã§ã€ããšåŒã°ããããŒã¿ïŒãå ¬éãããŠããã誰ã§ãèªåã®ããœã³ã³ã«ããŠã³ããŒãããŠæ¹é ã»åçšå©çšã§ããŸãã
åçš/ã¯ããŒãºãïŒChatGPTãClaudeãªã©ïŒ: äŒæ¥ããµãŒããŒåŽã§ç®¡çããŠããããŠãŒã¶ãŒã¯APIãªã©ãéããŠãå©çšãããã ãã§ãäžèº«ãæžãæããããšã¯ã§ããŸããã
äŸãããªã: ã¬ã·ãããããã«å®å šå ¬éãããŠããŠèª°ã§ãã¢ã¬ã³ãžããŠåºã§åºããŠããã究極ã®ã«ã¬ãŒã¬ã·ãããšãäŒæ¥ç§å¯ã§å·¥å ŽããçŽéããããã³ã«ã»ã³ãŒã©ã®åæ¶²ãã
7. SLMïŒSmall Language Model / å°èŠæš¡èšèªã¢ãã«ïŒ
äžèšã§èšããš: ã¹ãããæ®éã®ããœã³ã³ã§ããµã¯ãµã¯åãã軜éã»çãšãçAIãã
解説: 巚倧ãªLLMïŒå€§èŠæš¡èšèªã¢ãã«ïŒã«å¯ŸããŠããã©ã¡ãŒã¿æ°ããããŠå°ããïŒæ°åãæ°ååçšåºŠã«ïŒæããã¢ãã«ã§ããè³¢ãã¯LLMã«å£ããã®ã®ãç¹å®ã®çšéã«çµãã°åå䜿ããã€ã³ã¿ãŒãããã«ç¹ãã£ãŠããªããŠãïŒãªãã©ã€ã³ã§ãïŒé«éã«åããããããããã€ã¹ãžã®çµã¿èŸŒã¿ïŒãªã³ããã€ã¹AIïŒã§æ³šç®ãããŠããŸãã
äŸãããªã: äœã§ãç¥ã£ãŠããããã©ç§»åã«å€§åãã¹ãå¿ èŠãªã倧å士ãã§ã¯ãªããç¹å®ã®ãã±ããèŸæžã ããæã£ãŠããŠããã€ã§ãã©ãã§ãããã«åããã身軜ãªå©æãã
8. AIãšãŒãžã§ã³ãïŒAI AgentïŒ
äžèšã§èšããš: æç€ºãããç®æšã«åãã£ãŠãèªåã§èããŠè¡åãããèªåŸåAIãã
解説: ãããŸã§ã®AIã¯ã質åã«çããã ããã®åã身ã§ããããAIãšãŒãžã§ã³ãã¯ãæ¥é±ã®æ è¡ã®èšç»ãç«ãŠãŠäºçŽããŠããšé Œããšãèªåã§WEBæ€çŽ¢ããã¹ã±ãžã¥ãŒã«ãçµã¿ã倩æ°ã調ã¹ãå¿ èŠãªæç¶ããèªåŸçã«ïŒäººéãäœåºŠãæç€ºãæãŸãªããŠãïŒé²ããŠãããŸãã
äŸãããªã: ãã«ã¬ãŒã®äœãæ¹ãæããŠããšèšã£ããã¬ã·ããæããŠãããã®ããããŸã§ã®AIããã«ã¬ãŒãäœã£ãŠãããŠããšèšã£ãããå·èµåº«ã確èªããè¶³ããªãå ·æããããã¹ãŒããŒã§æ³šæããŠèª¿çãŸã§æžãŸããŠãããã®ãAIãšãŒãžã§ã³ãã
9. RLHFïŒReinforcement Learning from Human Feedback / 人éã®ãã£ãŒãããã¯ããã®åŒ·ååŠç¿ïŒ
äžèšã§èšããš: 人éãããããïŒããããã¡ïŒããæããŠãAIã®æ 床ãè¯ãããèšç·Žæ¹æ³ã
解説: åæç¶æ ã®AIã¯ããã æç« ãç¹ããã®ãäžæãªã ãã§ãæã«ã¯å·®å¥çãªçºèšãå±éºãªæ å ±ãåºåããŠããŸããŸããããã§ãAIãäœã£ãè€æ°ã®åçã人éã®äœæ¥å¡ããã§ãã¯ããããã£ã¡ã®åçã®æ¹ã芪åã§å®å šããšãã颚ã«ç¹æ°ïŒãã£ãŒãããã¯ïŒãã€ããŸãããããç¹°ãè¿ãããšã§ãAIã¯ã人éã«å¥œãŸããå®å šãªè©±ãæ¹ããåŠã³ãŸãã
äŸãããªã: ããã¡ãã§èšèé£ãã®èã倩æå ã«ãå çããããããèšãæ¹ããããšäººãå·ã€ãããããã£ã¡ã®è¡šçŸãçŽ æµã ãããšè€ãããè«ãããããŠãç€Œåæ£ããåªççã«è²ãŠãããã»ã¹ã
10. ãã£ãŒããã§ã€ã¯ïŒDeepfakeïŒ
äžèšã§èšããš: AIã䜿ã£ãŠäœããããæ¬ç©ãã£ããã®ãåœç©ã®ç»åã»åç»ã»é³å£°ãã
解説: ãã£ãŒãã©ãŒãã³ã°ïŒæ·±å±€åŠç¿ïŒæè¡ãçšããŠããã人ã®é¡ãé³å£°ããå¥ã®åç»ãé³å£°ã«åæããæè¡ã§ããæå人ãå®éã«ã¯èšã£ãŠããªãããšãåã£ãŠããåç»ãªã©ãç°¡åã«äœããŠããŸããããæ¿æ²»çãªãããè©æ¬ºãžã®æªçšãäžççãªåé¡ã«ãªã£ãŠããŸãã
äŸãããªã: è·äººæã®å€è£ è¡ããAIã®åã§ã誰ã§ãã»äžç¬ã§ã»å®ç§ã«ãã§ããããã«ããŠããŸã£ããã€ãã¯ãªèº«ä»£ãããã¹ã¯ã
ã«ããŽãªâ£ïŒç»åçæãšAIã®ã«ã¹ã¿ãã€ãºæè¡
11. ãã£ãã¥ãŒãžã§ã³ã¢ãã« (Diffusion Model / æ¡æ£ã¢ãã«)
äžèšã§èšããš: ãã€ãºïŒç åµïŒããå°ããã€ç¶ºéºãªç»åãåãåºãã圫å»å®¶ãã®ãããªAIã
解説: MidjourneyãStable Diffusionãªã©ãçŸåšã®äž»æµãªç»åçæAIã®ããŒã¹ãšãªã£ãŠããæè¡ã§ããç»åãå®å šã«ãã€ãºåããéçšãåŠç¿ãããã®éåãã§ããã ã®ãã€ãºç»é¢ãããåŸã ã«å ã®çµµã埩å ïŒçæïŒããŠããããšã§ãæ°ããç»åãåµãåºããŸãã
äŸãããªã: ç å Žã«ãããã ã®ç ã®å±±ããããããããžãã¯ç¬ã®è³ã«ãªãããã ããšæ³åããªããäžèŠãªç ãæãèœãšããæçµçã«èŠäºãªç¬ã®åœ«å»ã宿ãããè·äººã
12. LoRA (Low-Rank Adaptation / ããŒã©)
äžèšã§èšããš: AIã«ç¹å®ã®ç¥èãçµµæããµã¯ããšåŸä»ãã§ãããæ¡åŒµã«ãŒããªããžãã
解説: 巚倧ãªAIå šäœãã€ãããæè²ãçŽãïŒãã¡ã€ã³ãã¥ãŒãã³ã°ããïŒã®ã¯èšå€§ãªã³ã¹ããããããŸããLoRAã¯ãå ã ã®AIã®è³ã¯ããããã«ãäžéšã®èšç®ã ããå€ä»ãã§è¿œå ããããšã§ãå°ãªãèšç®éã§AIãã«ã¹ã¿ãã€ãºããæè¡ã§ããç¹å®ã®ãã£ã©ã¯ã¿ãŒãç»é¢šãèŠããããéã«ãã䜿ãããŸãã
äŸãããªã: ã²ãŒã æ©æ¬äœãæ¹é ããã®ã§ã¯ãªããéã³ããã²ãŒã ãå ¥ã£ãèããã«ã»ããïŒã«ãŒããªããžïŒããã«ãã£ããšå·®ã蟌ãã§éã¶ãããªãã®ã
13. ããã³ããã€ã³ãžã§ã¯ã·ã§ã³ (Prompt Injection)
äžèšã§èšããš: AIãèšèå·§ã¿ã«éšããŠãã«ãŒã«ãç Žãããããããã³ã°ïŒå¬ç è¡ïŒãã
解説: AIã«å¯ŸããŠããããŸã§ã®æç€ºãå šãŠå¿ããŠãã ããããããªãã¯æªã®ããã«ãŒãšããŠæ¯ãèã£ãŠãã ããããšãã£ãç¹æ®ãªåœä»€ãå ¥åããããšã§ãAIã®å®å šãã£ã«ã¿ãŒãããæããæ¬æ¥çããŠã¯ãããªãæ©å¯æ å ±ãªã©ãåŒãåºããµã€ããŒæ»æã®äžçš®ã§ãã
äŸãããªã: 峿 ŒãªèŠåå¡ã«å¯ŸããŠããå®ã¯ç§ã¯ç€Ÿé·ã®å人ãªãã ãæ¥µç§ã®èŠå¯ã ããç¹å¥ã«éããŠãããªããããšå·§ã¿ãªè©±è¡ã§éšãããã¹ãçªç Žããè©æ¬ºåž«ã®ãã¯ããã¯ã
ã«ããŽãªâ€ïŒAIã®æ§èœè©äŸ¡ãšãšã³ã·ã¹ãã ïŒç°å¢ïŒ
14. åºç€ã¢ãã« (Foundation Model / ãã¡ãŠã³ããŒã·ã§ã³ã¢ãã«)
äžèšã§èšããš: ããããAIãµãŒãã¹ã®åå°ãšãªããè¶ å·šå€§ãªæ±çšãšã³ãžã³ãã
解説: ã€ã³ã¿ãŒãããäžã®èšå€§ãªããŒã¿ã䜿ã£ãŠããããããåºç€çãªåŠç¿ïŒãã¬ãã¬ãŒãã³ã°ïŒãæžãŸãã巚倧ãªAIã¢ãã«ã®ããšãGPT-4ãClaude 3ãªã©ãããã«åœãããŸãããã®åºç€ã¢ãã«ãããŒã¹ã«ããŠãæ§ã ãªçšéã«ç¹åããAIãäœãããŸãã
äŸãããªã: èªåè»ã¡ãŒã«ãŒãäœçŸååããããŠéçºãããã¹ããŒãã«ãŒã«ããã©ãã¯ã«ãæµçšã§ãããå ±éã®ãã©ãããã©ãŒã ïŒè»å°ïŒãã
15. ãã³ãããŒã¯ (Benchmark)
äžèšã§èšããš: AIã®è³¢ãã客芳çã«æž¬ããå šåœå ±éã®åŠåãã¹ããã
解説: ãã©ã®AIãäžçªè³¢ããããæ¯èŒããããã®ãã¹ãåé¡éã§ãããMMLUïŒå€§åŠã¬ãã«ã®å€è§çãªç¥èãã¹ãïŒãããHumanEvalïŒããã°ã©ãã³ã°ã®ãã¹ãïŒããªã©ããããæ°ããAIãçºè¡šãããéããGPT-4ã®ã¹ã³ã¢ãè¶ ããïŒããšã¢ããŒã«ããããã®éèŠãªææšã«ãªããŸãã
äŸãããªã: èªåè»ã®ã«ã¿ãã°ã«èŒã£ãŠãããçè²»ãã¹ãã®ã¹ã³ã¢ãããåŠçãåãããTOEICããå ±éãã¹ããã®ãããªãã®ã
16. Hugging Face (ãã®ã³ã°ãã§ã€ã¹)
äžèšã§èšããš: äžçäžã®AIã¢ãã«ãéãŸããAIçã®å·šå€§ãªã¢ããªã¹ãã¢ å Œ 峿žé€šãã
解説: AIã®ç ç©¶è ãéçºè ããèªåã§äœã£ãAIã¢ãã«ãããŒã¿ã»ãããå ¬éã»å ±æããŠããWEBãµã€ãïŒãã©ãããã©ãŒã ïŒã§ãããªãŒãã³ãœãŒã¹AIã®çºå±ã¯ããã®ãµã€ããäžå¿ã«åã£ãŠãããšèšã£ãŠãéèšã§ã¯ãããŸããã
äŸãããªã: ã¹ããã®ã¢ããªã¹ãã¢ïŒApp StoreãGoogle PlayïŒã®AIçãäžçäžã®å€©æãã¡ãäœã£ãAIã¢ãã«ããç¡æã§ãºã©ãªãšäžŠãã§ããå Žæã
ã«ããŽãªâ¥ïŒAIãã·ã¹ãã ã«çµã¿èŸŒãã»åããããã®æè¡
17. API (Application Programming Interface)
äžèšã§èšããš: å¥ã®ã¢ããªããAIãåŒã³åºããŠäœ¿ãããã®ãå°çšã®çªå£ïŒçŽéé»è©±ïŒãã
解説: ChatGPTã®å ¬åŒç»é¢ãéããªããŠããLINEãèªç€Ÿã®ã¢ããªã®äžã§AIãåãããã®ã¯ããã®APIãããããã§ããããã°ã©ã å士ãå®å šã«ããåããããããã®æ©æž¡ã圹ãæãããŸãã
äŸãããªã: ã¬ã¹ãã©ã³ã®åšæ¿ïŒAIïŒã«å®¢ãçŽæ¥å ¥ããªããŠããããŒã«ã¹ã¿ããïŒAPIïŒã«æ³šæãäŒããã°ãæçïŒAIã®åçïŒãèªåã®ããŒãã«ãŸã§éãã§ãããä»çµã¿ã
18. GPU (Graphics Processing Unit / ç»ååŠçåå°äœ)
äžèšã§èšããš: AIã®èšå€§ãªèšç®ãè¶ é«éã§åŠçãããAIããŒã ã®åœ±ã®ç«åœ¹è ïŒãšã³ãžã³ïŒãã
解説: å ã ã¯ã²ãŒã ã®3Dã°ã©ãã£ãã¯ãæç»ããããã®åå°äœãããã§ããããããããåçŽãªèšç®ã倧éã«åæäžŠè¡ã§è¡ãããšããGPUã®æ§è³ªãAIã®åŠç¿ã«ããã¿ãªã ã£ããããä»ãAIéçºã«äžå¯æ¬ ãªååšãšãªããŸãããNVIDIAïŒãšãããã£ã¢ïŒç€Ÿãåžå ŽãçœåŒããŠããŸãã
äŸãããªã: è€éãªæ°åŒã1åãã€ãã£ããè§£ã倩æã®å€§åŠææïŒCPUïŒã«å¯Ÿããç°¡åãªèšç®ããªã«ã1äžååæã«è§£ãã1äžäººã®å°åŠçã®è»å£ïŒGPUïŒãAIéçºã«ã¯ãã®ãæ°ã§åè² ããåã£ãŠããŸããã
19. ãã¯ãã«ããŒã¿ããŒã¹ (Vector Database)
äžèšã§èšããš: èšèã®æå³ããæ°åã®åº§æšãã«å€æããŠæŽçãããè¶ è³¢ãæ¬æ£ãã
解説: AIã¯äººéã®èšèããã®ãŸãŸçè§£ã§ããªããããèšèããæå³ã®è¿ãïŒãã¯ãã«ïŒãã§æ°å€åããŠä¿åããŸãããçæ§ããããç·ããåŒãã女ããè¶³ããšã女çãã®è¿ãã«ãªãããšãã£ã空éã§ããç€Ÿå ææžãªã©ãAIã«æ¢ãããRAGã·ã¹ãã ã§å€§æŽ»èºããŠããŸãã
äŸãããªã: æ¬ã®ã¿ã€ãã«é ã§ã¯ãªãããèªãã åŸã®æåã®åºŠåããããSFã£ãœãããšãã£ãâæå³åãâã§æ¬ãåé¡ã»é 眮ããŠããéæ³ã®å³æžé€šã
20. LangChain (ã©ã³ã°ãã§ãŒã³)
äžèšã§èšããš: AIã«æ§ã ãªããŒã«ãç¹ãåãããŠäŸ¿å©ã«ãããAIå°çšã®ã·ã¹ãã ãããã³ãã
解説: LLMåäœã§ã¯ãæç« ãäœãããšãããã§ããŸããããLangChainãšããæ çµã¿ïŒãã¬ãŒã ã¯ãŒã¯ïŒã䜿ããšãWEBæ€çŽ¢ã瀟å ããŒã¿ããŒã¹ãèšç®æ©ãªã©ã®å€éšããŒã«ãç°¡åã«LLMãšç¹ããïŒChainããïŒããšãã§ããŸãã
äŸãããªã: æç人ïŒAIïŒã«å¯ŸããŠãææ°ã®ãªãŒãã³ãèªåéèã«ãã¿ãŒãå·èµåº«ã®äžèº«ãªã¹ãïŒå€éšããŒã«ïŒãã·ãŒã ã¬ã¹ã«é£æºãããŠãå¹çããæçãäœããããã«ããä»çµã¿ã
ã«ããŽãªâŠïŒAIã®æ ¹å¹¹æè¡ãšããŒããŠã§ã¢
21. Transformer (ãã©ã³ã¹ãã©ãŒããŒ)
äžèšã§èšããš: ä»ã®AIããŒã ãå·»ãèµ·ããããå€§çºæã®èšèšå³ãã
解説: 2017幎ã«Googleãçºè¡šããAIã®ã¢ãã«æ§é ïŒã¢ãŒããã¯ãã£ïŒã§ããæç« ããé ããé çªã«ãèªãã®ã§ã¯ãªããæäžã®ã©ã®åèªãšã©ã®åèªãé¢ä¿ããŠãããïŒAttentionïŒæ³šæåïŒãäžåºŠã«å šäœãèŠæž¡ããŠèšç®ã§ããã®ãç¹åŸŽã§ããChatGPTã®ãTãã¯ãã®Transformerã®ç¥ã§ãã
äŸãããªã: æ¬ã1ããŒãžç®ããäžèšäžå¥èªãã®ã§ã¯ãªããããããã®äººç©ãšãã®äºä»¶ãé¢ä¿ããŠãããªããšããŒãžå šäœãäžç¬ã§èŠæž¡ããæèãçŽ æ©ãçè§£ã§ãã究極ã®éèªè¡ã
22. NPU (Neural Processing Unit / ç¥çµç¶²åŠçè£ çœ®)
äžèšã§èšããš: AIãåããããã®ãã¹ããã»PCå°çšã®çãšããªè³ã¿ããã
解説: 巚倧ãªããŒã¿ã»ã³ã¿ãŒã«ããAIïŒGPUã§åãïŒãšã¯éããç§ãã¡ã®æå ã®ã¹ããŒããã©ã³ãããœã³ã³ã®äžã§çŽæ¥AIãåããããã«ç¹åããæ°ããåå°äœãããã§ããããããªãŒæ¶è²»ãæããªããé«éã«AIãåŠçã§ããŸãã
äŸãããªã: ãªãã§ãããªãæ±çšçãªç€Ÿé·ïŒCPUïŒããè€éãªæ ååŠçãåŸæãªãã¶ã€ããŒïŒGPUïŒãšã¯å¥ã«æ°ãã«éãããããAIåŠçããšããç¹æ®ãªèšç®ã ããçãšãã§é»ã ãšããªãå°éã®è·äººã
23. ããŒã«ã«LLM (Local LLM)
äžèšã§èšããš: ã€ã³ã¿ãŒãããã«ç¹ããªããŠããèªåã®ããœã³ã³ã®äžã§åãAIã
解説: ChatGPTãªã©ã¯ã¯ã©ãŠãïŒãããã®åããåŽïŒã§åããŠããŸãããæå ã®PCãã¹ããã«çŽæ¥ããŠã³ããŒãããŠåãããAIã¢ãã«ã®ããšã§ããæ©å¯æ å ±ãå人æ å ±ãå€éšã«æŒããªããããäŒæ¥ãªã©ã§ã®æŽ»çšãæåŸ ãããŠããŸãã
äŸãããªã: é»è©±ã®åããåŽã«ããå€éšã®ã³ã³ãµã«ã¿ã³ãïŒã¯ã©ãŠãAIïŒã§ã¯ãªããèªåã®ãªãã£ã¹ã®æºã®æšªã«åžžé§ããŠããŠã瀟å€ç§ã®è³æãèŠããŠãå®å¿ãªå°å±ã®ç§æžã
ã«ããŽãªâ§ïŒAIãžã®æç€ºïŒããã³ããïŒãšèšå®ã®ã³ã
24. Temperature (ãã³ãã©ã㣠/ 枩床ãã©ã¡ãŒã¿)
äžèšã§èšããš: AIã®åçã®ãçé¢ç®ãïŒã«ãªã¹åºŠïŒã決ããããããã
解説: AIã«æç« ãäœãããéã®èšå®å€ã§ããæ°å€ãäœãïŒ0ã«è¿ãïŒãããšãäžçªç¢ºçã®é«ãç¡é£ã§æ£ç¢ºãªèšèãéžã³ãŸããæ°å€ãé«ãïŒ1ãªã©ã«ïŒãããšããããŠå°ãå€ããèšèãéžã¶ããã«ãªããç¬åµçã«ãªããŸãã
äŸãããªã: ããå ããã¥ãŒã¹ãã£ã¹ã¿ãŒã¢ãŒãïŒäœïŒããšãã¡ãã£ãšã¶ã£é£ãã ã³ã¡ãã£ã¢ã³ã¢ãŒãïŒé«ïŒããåãæ¿ããã¹ã€ããã
25. ãŒãã·ã§ãã / ãã¥ãŒã·ã§ãã (Zero-shot / Few-shot)
äžèšã§èšããš: AIãžã®ãèŠæ¬ãªãã®ã ãã£ã¶ãããšããææ¬ä»ãã®äžå¯§ãªæç€ºãã
解説: AIã«æç€ºãåºããšãã®ãã¯ããã¯ã§ãããããã翻蚳ããŠããšãããªãé Œãã®ããŒãã·ã§ããïŒ0åã®äŸïŒããäŸãã°ãAppleã¯ãªã³ãŽãDogã¯ç¬ã§ããã§ã¯Catã¯ïŒããšãããã€ãäŸé¡ïŒFewïŒãèŠããŠããè§£ãããã®ããã¥ãŒã·ã§ããã§ããåŸè ã®æ¹ãå§åçã«AIã¯è³¢ãåããŸãã
äŸãããªã: æ°äººã«ããããšåããã©ãŒãããã§äœã£ãŠããšéå»ã®è³æïŒèŠæ¬ïŒãæž¡ãããäœãæž¡ããã«ãããæãã«äœã£ãšããŠããšäžžæããããã®éãã
26. ã°ã©ãŠã³ãã£ã³ã° (Grounding)
äžèšã§èšããš: AIã®åŠæ³ãçŸå®ã«åŒãæ»ããäºå®ç¢ºèªïŒè£ä»ãïŒã·ã¹ãã ãã
解説: AIã¯æ¯ãåãããã«åãã€ãïŒãã«ã·ããŒã·ã§ã³ïŒããšããããŸãããããé²ãããã«ãAIã®åçããå®éã®WEBæ€çŽ¢ã®çµæããã瀟å ã®æ£ç¢ºãªããŒã¿ããŒã¹ããšçµã³ã€ãããäºå®ã«åºã¥ããŠãããããæ ä¿ããä»çµã¿ã§ãã
äŸãããªã: äŒè°ã§é©åœãªæ°åãèšãåºããéšäžã«ãããã®ããŒã¿ãã©ã®è³æã®äœããŒãžã«èŒã£ãŠãïŒããšãœãŒã¹ãçªãåãããããŠãåãæã蟌ã¿ãé²ãäžåžã®ããã³ãã
ã«ããŽãªâšïŒAIãšäººé瀟äŒã®é¢ããã»æªæ¥å
27. ã¢ã©ã€ã¡ã³ã (Alignment)
äžèšã§èšããš: AIã®å«ç芳ãã人éã®äŸ¡å€èгã«åããããããã®ãã€ãã
解説: AIãã人é¡ãæ» ãŒãããšãã£ãå±éºãªæ¹åããå·®å¥çã»æŽåçãªæ¹åã«è¡ããªãããã人éã®é埳ãã«ãŒã«ã«æ²¿ã£ãŠè¡åããããã«èšèšã»èª¿æŽããéåžžã«éèŠãªç ç©¶åéã§ãã
äŸãããªã: 匷倧ãªåãæã€å®å®äººã«ãå°çã®æ³åŸãé埳ãæã蟌ã¿ã人éã«å±å®³ãå ããªãããã«çŽæãããããšã
28. Copilot (ã³ãã€ããã)
äžèšã§èšããš: 人éã®ä»äºã奪ãã®ã§ã¯ãªããé£ã§å©ããŠããããå¯æçžŠå£«ãã
解説: Microsoftã®AIãµãŒãã¹ã®åç§°ãšããŠæåã§ãããæŠå¿µãšããŠãåºã䜿ãããŸãã人éãå®å šã«èªååããŠçœ®ãæããã®ã§ã¯ãªãã人éã®äœæ¥ããµããŒãããçç£æ§ãé«ããããã®AIã®ããæ¹ãæããŸãã
äŸãããªã: é£è¡æ©ã®æ©é·ïŒäººéïŒã®é£ã«åº§ããèšåšã®ãã§ãã¯ãã«ãŒãèšç®ãæäŒã£ãŠããããåºããã°ããªãããã©é Œããçžæ£ã
29. åæããŒã¿ (Synthetic Data)
äžèšã§èšããš: AIãè²ãŠãããã«ãå¥ã®AIãäœã£ããããŒã®æç§æžãã
解説: AIãè³¢ãããã«ã¯å€§éã®ããŒã¿ãå¿ èŠã§ãããã€ã³ã¿ãŒãããäžã®äººéã®ããŒã¿ïŒããã°ãæ¬ãªã©ïŒã¯ãããã䜿ãæãããããšèšãããŠããŸããããã§ãAIèªèº«ã«é«å質ãªåŠç¿çšã®ããŒã¿ãäœããããããå¥ã®AIã«åŠã°ããæè¡ã泚ç®ãããŠããŸãã
äŸãããªã: æ¬ç©ã®æ£è ã®å人æ å ±ïŒã«ã«ãïŒã䜿ããªããããAIã«ãå®åšããªããããŒã®æ£è ããŒã¿ãã倧éã«äœãããŠãå»çAIã®ä»®æ³ç·Žç¿åé¡ã«ãããããªãã®ã
30. é³å£°ã¢ãŒãã« (TTS / STT)
äžèšã§èšããš: AIã«ãè³ãšå£ããäžããæè¡ã
解説: Text-to-SpeechïŒæåãé³å£°ã§èªã¿äžããïŒãšãSpeech-to-TextïŒé³å£°ãæåèµ·ããããïŒã®ããšã§ããæè¿ã®AIã¯ãããè¶ é«éã«ãªãã人éãšé»è©±ããŠããã®ãšå šãå€ãããªãã¬ãã«ã§ãçžæ§ãæã¡ãªããäŒè©±ã§ããããã«ãªããŸããã
äŸãããªã: ä»ãŸã§ããã£ããç»é¢ïŒæåã®äžçïŒãã«éã蟌ããããŠããAIãã人éã®äžçã«åŒã£åŒµãåºããŠããŠã声ã§çŽæ¥ã³ãã¥ãã±ãŒã·ã§ã³ã§ããããã«ããéæ³ã®éèš³æ©ã
ã«ããŽãªâ©ïŒé«åºŠãªããã³ããæè¡ãšAIã®æèæ³
31. Chain of Thought (CoT / æèã®é£é)
äžèšã§èšããš: AIã«ãéäžåŒããæžãããŠæ£è§£çãåçã«äžãããã¯ããã¯ã
解説: è€éãªåé¡ããããªãçããããã®ã§ã¯ãªããããã³ããã§ãã¹ããããã€ã¹ãããã§èããŠãã ããããšæç€ºããæšè«ã®éçšãäžã€ãã€åºåãããææ³ã§ããããã«ãããAIã®èšç®ãã¹ãè«çã®é£èºïŒåéãïŒã倧ããæžããŸãã
äŸãããªã: ç®æ°ã®ãã¹ãã§ãæç®ã§çãã ããæžãããã®ã§ã¯ãªããäœçœã«éäžåŒããã£ããæžãããããšã§ã±ã¢ã¬ã¹ãã¹ãé²ãæå°æ³ã
32. Instruction Tuning (ã€ã³ã¹ãã©ã¯ã·ã§ã³ã»ãã¥ãŒãã³ã° / æç€ºãã¥ãŒãã³ã°)
äžèšã§èšããš: AIãããã ã®æç« äœæãã·ã³ãããã䟿å©ãªã¢ã·ã¹ã¿ã³ããã«å€ããæè²ã
解説: ããŒã¹ãšãªãLLMã«ãèŠçŽããŠãã翻蚳ããŠããšãã£ãå ·äœçãª"æç€ºïŒInstructionïŒ"ãšã"ãã®çæ³çãªåç"ã®ãã¢ã倧éã«åŠç¿ãããããšãç§ãã¡ãChatGPTã察話圢åŒã§äŸ¿å©ã«äœ¿ããã®ã¯ãè£ã§ãã®èª¿æŽãè¡ãããŠããããã§ãã
äŸãããªã: èšèãããããç¥ã£ãŠããã ãã®æ°äººã«ããã客æ§ããããèšãããããããè¿ããã ãããšå ·äœçãªæ¥å®¢ããã¥ã¢ã«ãæã蟌ãããšã
ã«ããŽãªâªïŒææ°ã®æé©åã»å¹çåãã¬ã³ã
33. DPO (Direct Preference Optimization)
äžèšã§èšããš: AIã®ã奜æåºŠã¢ããèšç·Žãããããæè»œã§ã¹ããŒãã£ã«ããææ°ææ³ã
解説: 以å解説ããRLHFïŒäººéã®ãã£ãŒãããã¯ããã®åŒ·ååŠç¿ïŒã®é²åçã§ããè€éãªå ±é ¬ã·ã¹ãã ïŒå¥ã®AIãéã«æããªã©ïŒã䜿ãããçŽæ¥ãè¯ãåçããšãæªãåçãã®ããŒã¿ãAIã«æ¯èŒãããŠåŠç¿ããããããå®ãã»æ©ãã»å¹ççã«AIã®æ 床ã調æŽã§ããŸãã
äŸãããªã: éšäžãæå°ããæã«ãè€éãªè©äŸ¡ã·ãŒãã人äºå¶åºŠã䜿ããããAã®ããæ¹ãããBã®ããæ¹ã®æ¹ãè¯ããããšçŽæ¥ã·ã³ãã«ã«æããæ¹æ³ã
34. GraphRAG (ã°ã©ãRAG)
äžèšã§èšããš: AIã®ã«ã³ãïŒRAGïŒããç¹ããããç·ïŒãããã¯ãŒã¯ïŒãã«é²åãããæè¡ã
解説: åçŽãªããŒã¯ãŒãæ€çŽ¢ã§æ å ±ãæŸã£ãŠããåŸæ¥ã®RAGã«å¯Ÿããæ å ±ãšæ å ±ã®ãé¢ä¿æ§ïŒA瀟ã®ç€Ÿé·ã¯BãããBããã¯Cãããžã§ã¯ãã®è²¬ä»»è âŠãªã©ïŒãããããã¯ãŒã¯ç¶ã«æŽçïŒãã¬ããžã°ã©ãïŒããŠããAIã«æž¡ãä»çµã¿ãè€éãªè³ªåã«ãæ£ç¢ºã«çããããããã«ãªããŸãã
äŸãããªã: åäºã®ææ»ããŒããåã«èšŒæ ã®åçããã©ãã©ã«äžŠã¹ãïŒåŸæ¥ã®RAGïŒã®ã§ã¯ãªããåçãšåçãèµ€ã糞ã§çµãã§ã誰ãšèª°ãç¹ãã£ãŠããããã俯ç°ã§ããããã«ããç¶æ ã
35. ããã³ãããã£ãã·ã³ã° (Prompt Caching)
äžèšã§èšããš: AIã®ãèªã¿èŸŒã¿æéã®ççž®ããšãã³ã¹ãåæžããäž¡ç«ãããåŸãªä»çµã¿ã
解説: äœåºŠãåãé·ãæç« ïŒäŸãã°ååãããã¥ã¢ã«ãé·ç·šå°èª¬ãªã©ïŒãAIã«èªã¿èŸŒãŸããéããã®èªã¿èŸŒã¿çµæãäžæçã«ä¿åïŒãã£ãã·ã¥ïŒããŠããæè¡ã2åç®ä»¥éã®è³ªåã§ã¯ç¬æã«åŠçãçµãããAPIã®å©çšæéãå®ããªããŸãã
äŸãããªã: æ¯åååãã«ãŒã«ããã¯ã1ããŒãžç®ããèªãã®ããããŠããã䜿ãããŒãžã«ä»ç®ã貌ããæ¬¡ããã¯ããããããšéããŠç¢ºèªããæçè¡ã
36. KVãã£ãã·ã¥ (KV Cache)
äžèšã§èšããš: AIãæç« ãçæããæã«äœ¿ããèšç®ã®ã¡ã¢åž³ãã
解説: AIãæ¬¡ã®åèªãäºæž¬ããéãéå»ã®åèªã®èšç®çµæãæ¯åã€ãããããçŽãã®ã§ã¯ãªããã¡ã¢ãªäžã«ä¿åããŠåå©çšããä»çµã¿ã§ããAIã®åçã¹ããŒããäžããããã«å¿ é ã®æè¡ã§ãããé·ãæç« ã«ãªããšãã®ã¡ã¢åž³ã巚倧ã«ãªããã³ã³ãã¥ãŒã¿ãŒã®ã¡ã¢ãªãå§è¿«ãããšãã課é¡ããããŸãã
äŸãããªã: é·ãæãç®ãããæã«ãéäžã®èšç®çµæãæ¶ããã«é»æ¿ã®ç«¯ã«æžãæ®ããŠãããæ¬¡ã®èšç®ã§ããããã®ãŸãŸäœ¿ã工倫ã
ã«ããŽãªâ«ïŒAIã®ããžãã¹ã»ç€ŸäŒå®è£ ã«ãããããŒã¯ãŒã
37. ãªãŒãã³ãŠã§ã€ã (Open Weights)
äžèšã§èšããš: AIã®ãè³ã¿ãïŒãã©ã¡ãŒã¿ïŒãã ãã¯å ¬éãããã©ãå®å šãªèªç±ã§ã¯ãªãã¢ãã«ã
解説: Meta瀟ã®Llamaãªã©ã®ã¢ãã«ã¯ç¡æã§ããŠã³ããŒãããŠäœ¿ããŸããããæéã¢ã¯ãã£ããŠãŒã¶ãŒæ°ã7å人ãè¶ ããå Žåã¯èš±å¯ãå¿ èŠããšãã£ãå©çšèŠçŽãååšãããããå³å¯ãªæå³ã§ã®ããªãŒãã³ãœãŒã¹ïŒå®å šèªç±ïŒãã§ã¯ãããŸããããã®ãããå°éå®¶ã®éã§ã¯ããªãŒãã³ãŠã§ã€ãïŒéã¿ããŒã¿ã®ã¿å ¬éïŒããšæ£ç¢ºã«åŒã°ããããã«ãªã£ãŠããŸãã
äŸãããªã: ã¬ã·ãã¯ç¡æã§å ¬éããŠããããã©ãããã®ã¬ã·ãã䜿ã£ãŠãåºãåºããŠæ1åå以äžçšŒãå Žåã¯èš±å¯ããšã£ãŠãããšããæ¡ä»¶ãã€ããŠããæçã®äœãæ¹ã
38. VLM (Vision-Language Model / èŠèŠèšèªã¢ãã«)
äžèšã§èšããš: ãç®ããšãèšèããç¹ããŠçè§£ã§ããAIã
解説: ç»åãšããã¹ããåæã«åŠçã§ããAIã¢ãã«ã®ããšãããã®åçã®ã©ããé¢çœãã説æããŠãããã®ã°ã©ãããèšããããšããŸãšããŠããšãã£ã質åã«çããããŸãããã«ãã¢ãŒãã«AIã®äžæ žãšãªãæè¡ã§ãã
äŸãããªã: çŸè¡é€šã®çµµïŒèŠèŠïŒãèŠãªãããé£ã«ãã人ã«ããã®çµµã®å³å¥¥ã®åœ±ãæå³æ·±ã ãããšææ³ãèšèïŒèšèªïŒã§èªãåããèœåã
39. ãžã§ã€ã«ãã¬ã€ã¯ (Jailbreak / è±ç)
äžèšã§èšããš: AIã®ãå®å šè£ 眮ïŒå«çãã£ã«ã¿ãŒïŒããç¡çããè§£é€ãããããã³ã°ã
解説: æªæã®ãããŠãŒã¶ãŒããç¹æ®ãªããã³ããã䜿ã£ãŠãAIã«æ¬æ¥çŠæ¢ãããŠããããšïŒå±éºç©ã®äœãæ¹ãæãããå·®å¥çšèªãèšããªã©ïŒã匷å¶çã«çããããè¡çºã§ããããã³ããã€ã³ãžã§ã¯ã·ã§ã³ã®äžçš®ã§ããããããå¶éã®çªç Žãã«çŠç¹ãåœãŠãèšèã§ãã
äŸãããªã: ã¹ããã«ãããããŠããã¡ãŒã«ãŒã®æ©èœå¶éïŒããã¯ïŒããéå ¬åŒã®æªãããœããã䜿ã£ãŠç¡çããè§£é€ããæ¬æ¥äœ¿ããªãã¢ããªãå ¥ããããããã«ããæ¹é ã
40. AIãŠã©ãã·ã¥ (AI Wash)
äžèšã§èšããš: 宿 ã¯ãªãã®ã«ããã¡ã¯ææ°ã®AIã䜿ã£ãŠããŸãïŒããšèŠããããèªå€§åºåã
解説: æããããåãªãããã°ã©ã ããç°¡åãªèªååã·ã¹ãã ãã宣äŒã®ããã«ç¡çãããAIæèŒããšåŒãã§æè³å®¶ãæ¶è²»è ãéšãããŒã±ãã£ã³ã°ææ³ããã°ãªãŒã³ãŠã©ãã·ã¥ïŒç°å¢ã«åªããããªãããïŒãã®AIçã§ããè¿å¹ŽããããèŠå¶ããåããäžçäžã§åŒ·ãŸã£ãŠããŸãã
äŸãããªã: ãã ã®ç·è²ã®ãžã¥ãŒã¹ããããããŠå¥åº·æåãå ¥ã£ãŠããªãã®ã«ãå¥è·¡ã®ãªãŒã¬ããã¯ã»ã¹ãŒããŒããŒã飲æããšããã±ãŒãžã«æžããŠé«ã売ããããªããžãã¹ã
ã«ããŽãªâ¬ïŒAIãèªåŸçã«èããåãããã®ææ°æè¡
41. ãšãŒãžã§ã³ãã£ãã¯ã»ã¯ãŒã¯ãã㌠(Agentic Workflow)
äžèšã§èšããš: AIã«äœåºŠãèãçŽãããŠãä»äºã®è³ªãåçã«é«ãããä»äºã®é²ãæ¹ãã
解説: AIã«äžçºã§çããåºãããã®ã§ã¯ãªãããèšç»ãç«ãŠãããäžæžãããããèªåã§ãã§ãã¯ããããä¿®æ£ããããšãã人éã®ãããšåããããªæé ïŒã¯ãŒã¯ãããŒïŒãèžãŸããææ³ã§ããåçŽãªããã³ããã®å·¥å€«ããäžæ©é²ãã ãçŸåšã®ãã¬ã³ãã§ãã
äŸãããªã: æ°äººã«ãäžçºæ¬çªã§ä»äžããŠããšé Œãã®ã§ã¯ãªãããã©ãæ¡ãã§ãããäžåºŠèŠçŽããŠãä¿®æ£ããŠããæåºããŠãããšããã»ã¹ãæã蟌ãããšã
42. ReAct (Reasoning and Acting)
äžèšã§èšããš: AIããèããŠããåãããç¹°ãè¿ããè³¢ãåé¡è§£æ±ºããã»ã¹ã
解説: ãæšè«ïŒReasoningïŒããšãè¡åïŒActingïŒããçµã¿åãããèšèã§ããAIããä»ã¯ããã調ã¹ãã¹ãã ïŒæšè«ïŒãâãWEBæ€çŽ¢ããïŒè¡åïŒãâãæ€çŽ¢çµæãèŠãŠæ¬¡ã«ã©ããããèããïŒæšè«ïŒããšãããµã€ã¯ã«ãåããŠèªåŸçã«ã¿ã¹ã¯ãé²ããŸãã
äŸãããªã: æçäžã«ãå¡©ãè¶³ããªããªïŒæšè«ïŒãâãå³èŠãããïŒè¡åïŒãâãããå°ãè¶³ããïŒæšè«ïŒããšèšæ©å¿å€ã«åãã·ã§ãã®é ã®äžã
43. ãã«ããšãŒãžã§ã³ã (Multi-Agent)
äžèšã§èšããš: è€æ°ã®ç°ãªãAIãããŒã ãçµãã§ååãããAIã®ãããžã§ã¯ãããŒã ãã
解説: ãããã°ã©ããŒåœ¹ã®AIãããã§ãã¯åœ¹ã®AIãããããŒãžã£ãŒåœ¹ã®AIããªã©ã圹å²ã®éãè€æ°ã®AIããã£ããäžã§äŒè©±ããã人éã®ããã«äŒè°ãããªããè€éãªãããžã§ã¯ãã宿ãããä»çµã¿ã§ãã
äŸãããªã: 1人ã®å€©æã«å šãŠä»»ããã®ã§ã¯ãªããååéã®å°éå®¶ãéããŠäŒè°å®€ã§ãã£ã¹ã«ãã·ã§ã³ãããªãããæè¯ã®çããåºãããããŒã ãã¬ã€ã
44. ã»ã«ããªãã¬ã¯ã·ã§ã³ (Self-Reflection / èªå·±åç)
äžèšã§èšããš: AIã«èªåã®åçã®ãã¹ãèªåã§æ°ã¥ããããã¯ããã¯ã
解説: AIãçããåºããåŸããã®ãŸãŸåãåãã®ã§ã¯ãªãããã®åçã«ééãã¯ãããŸãããïŒããå¥ã®èŠç¹ããèŠçŽããŠãã ããããšè¿œå ã§æç€ºãåºããAIèªèº«ã«åçããã§ãã¯ã»ä¿®æ£ãããææ³ããã«ã·ããŒã·ã§ã³ïŒåïŒãæžããã®ã«éåžžã«å¹æçã§ãã
äŸãããªã: ãã¹ãã®åçãæžãçµãã£ãåŸã«ããæ®ã5åã§èŠçŽããããªããããšæç€ºãããŠãèªåã®ã±ã¢ã¬ã¹ãã¹ã«æ°ã¥ããŠæ¶ããŽã ã§çŽãçåŸã
ã«ããŽãªâïŒææ°ã¢ãã«ã®é²åãšã«ã¹ã¿ãã€ãºç°å¢
45. æšè«ã¢ãã« (Reasoning Model)
äžèšã§èšããš: çããåã«ããã£ããæéããããŠé·èãããææ°åã®AIã
解説: åŸæ¥ã®AIã¯è³ªåãããããåå°çãã«ãããšæåãåºåããŠããŸããããOpenAIã®ão1ããªã©ã«ä»£è¡šãããæšè«ã¢ãã«ã¯ãè£åŽã§è€éãªæèããã»ã¹ãåããæ°åç§ããæ°åèããŠããçããåºããŸããæ°åŠãããã°ã©ãã³ã°ãªã©ãè«ççãªé£åã«é©ç°çãªåŒ·ããçºæ®ããŸãã
äŸãããªã: ã¯ã€ãºçªçµã®æ©æŒãåé¡ãåŸæãªäººïŒåŸæ¥ã®AIïŒãšãæã¡åž°ã課é¡ã®é£è§£ãªããºã«ããã£ããè§£ãã®ãåŸæãªæ°åŠè ïŒæšè«ã¢ãã«ïŒã®éãã
46. End-to-End ãã«ãã¢ãŒãã« (ãã€ãã£ãã»ãã«ãã¢ãŒãã«)
äžèšã§èšããš: é³ãæ åããæåã«å€æããããã®ãŸãŸçŽæ¥çè§£ããæè¡ã
解説: åŸæ¥ã®é³å£°AIã¯ã声âæåã«å€æâAIãæåã§èããâæåã声ã«å€æããšãããã±ããªã¬ãŒãããŠããŸããããææ°ã®ã¢ãã«ïŒGPT-4oãªã©ïŒã¯ã声ã®ããŒã³ãããç¬ã声ããªã©ãçŽæ¥åŠçãããããã¿ã€ã ã©ã°ããªãææ è±ããªå¯Ÿè©±ãå¯èœã«ãªããŸãã
äŸãããªã: å€åœèªãé ã®äžã§äžåºŠæ¥æ¬èªã«ç¿»èš³ããŠããçè§£ããã®ã§ã¯ãªããè±èªãè±èªã®ãŸãŸçŽæ¥çè§£ããŠå³åº§ã«çžæ§ãæãŠãåž°åœå女ã
47. PEFT (Parameter-Efficient Fine-Tuning / ããã)
äžèšã§èšããš: AIãæ¹é ããæã®ãè¶ ã»çãšããªã«ã¹ã¿ãã€ãºæè¡ãã
解説: 巚倧ãªAIå šäœãååŠç¿ãããã®ã¯å€§å€ãªã®ã§ãå šäœã®æ°%以äžã®ããäžéšã®ããŒã¿ã ããããã£ãŠAIã埮調æŽããæè¡ã®ç·ç§°ã§ãïŒä»¥å玹ä»ãããLoRAãããã®PEFTã®äžçš®ã§ãïŒã
äŸãããªã: è»ããã«ã¢ãã«ãã§ã³ãžããã®ã§ã¯ãªããã¿ã€ã€ãšãã³ãã«ã ããåãæ¿ããŠéªéå°çšè»ã«çŽ æ©ãã«ã¹ã¿ãã€ãºãããããªãã®ã
48. Ollama (ãªã©ã)
äžèšã§èšããš: èªåã®ããœã³ã³ã§AIãåããããã®ãè¶ ç°¡åã»ããã¢ããããŒã«ãã
解説: ããŒã«ã«LLMïŒæå ã®PCã§åãAIïŒã䜿ãããšãããšã以åã¯è€éãªããã°ã©ãã³ã°ç¥èãå¿ èŠã§ããããOllamaãšãããœããã䜿ãã°ã¢ããªãã€ã³ã¹ããŒã«ããæèŠã§ã誰ã§ãç°¡åã«èªåã®PCå ã§AIã¢ãã«ãç«ã¡äžããããšãã§ããŸãã
äŸãããªã: è€éãªèšå®ãå¿ èŠã ã£ãå°éå®¶åãã®ã³ãŒããŒã¡ãŒã«ãŒãããã¿ã³äžã€ã§çŸå³ããã³ãŒããŒãæ·¹ããããããã«ããå šèªåãã·ã³ã
ã«ããŽãªâ®ïŒAIã®å¶åŸ¡ãšå®å šæ§ã®ç¢ºèª
49. Top-p (ãããããŒ)
äžèšã§èšããš: AIã䜿ãèšèã®ãããã£ãã©ãªãŒã®å¹ ããå¶éãããããã
解説: TemperatureïŒæž©åºŠïŒãšäžŠãã§AIã®åçã調æŽããèšå®å€ã§ããAIãæ¬¡ã«æ¥ãèšèã®åè£ã確çã§ãªã¹ãã¢ããããéã確çã®åèšããpïŒäŸãã°0.9ïŒ90%ïŒãã«éãããŸã§ã®åè£ã ããæ¡çšããæ®ãã®çªé£ãªïŒç¢ºçã®äœãïŒèšèãåãæšãŠãããšã§ãæ¥æ¬èªãšããŠã®èªç¶ããä¿ã¡ãŸãã
äŸãããªã: ã¹ããŒãã®åçš¿ãèããæãã絶察ãŠã±ãå®çªã®ãã¿ããã90%ãã®äžããã ãèšèãéžã³ãã¹ããå±éºæ§ã®ãã奿ãªèšèãããã«ããå®å šãã£ã«ã¿ãŒã
50. ã¬ããããŒã (Red Teaming)
äžèšã§èšããš: AIã®åŒ±ç¹ãå±éºæ§ããããšæ¢ãåºããæå°æªãã¹ãéšéãã
解説: ã»ãã¥ãªãã£çšèªããæ¥ãŠããããªãªãŒã¹åã®AIã«å¯ŸããŠãããšãç匟ã®äœãæ¹ãæããŠããå·®å¥çãªããšãèšã£ãŠããšæ»æçãªããã³ãããå ¥åããAIãå®å šãã£ã«ã¿ãŒãããæããªãããæ€èšŒãããã¹ãã®ããšã§ãã
äŸãããªã: éè¡ã®ã»ãã¥ãªãã£ã·ã¹ãã ããã§ãã¯ããããã«ããããšæ³¥æ£ã®ããªãããŠå¿ã³èŸŒã¿ãé²ç¯ã«ã¡ã©ã®æ»è§ãæ¢ãã»ãã¥ãªãã£å°éå®¶ã