AI's Version of Moore's Law? - Computerphile

https://www.youtube.com/watch?v=evSFeqTZdqs
https://metr.org

Note that the success rate on the default chart is only 50% and for 80% the score is much lower. But the interesting part is indeed the rate of progress.

#AI #LLM #OpenAI #Anthropic #METR

🚨 KI-Agenten exponentiell besser? METR zeigt steigende "Time Horizons" – aber 50% Erfolg = jeder 2. Versuch scheitert. Log-Skala: stabiler.

👉 Meine Einschätzung: Wirtschaftlich relevant, aber kein Beweis für baldige Agentenübernahmen.

(Picture Credits to METR, via metr org, abgerufen am 22.2.26, "Model Evaluation & Threat Research", Social-Media-Bearbeitung und Screenshot druch: Marlon Niklas Kaulich)

#KI #AIAgents #METR #KünstlicheIntelligenz

https://winbuzzer.com/2026/02/25/anthropic-drops-hard-safety-limit-responsible-scaling-policy-xcxwbn/

Anthropic Drops Hard Safety Limits From its AI Scaling Policy

#AI #Anthropic #ResponsibleScalingPolicy #AISafety #AIRegulation #AISafety #AIModels #AITraining #CatastrophicRisk #METR #TrumpAdministration #Claude

[Opus 4.6, 사람 기준 14.5시간짜리 문제를 푼다는 것의 의미 (METR Time Horizon)

METR 연구기관의 Opus 4.6 모델이 인간 전문가 기준 14.5시간짜리 문제를 50% 확률로 해결할 수 있는 능력을 평가한 연구 결과를 발표했습니다. 이 연구는 AI의 장기적·자율적 작업 수행 능력을 측정하며, AI가 고숙련 지식 노동을 대체할 수 있는 임계점에 도달했음을 시사합니다.

https://news.hada.io/topic?id=26872

#ai #metr #opus46 #automation #timehorizon

Cari #devs,

Uno studio #METR ha scoperto che gli sviluppatori esperti erano convinti che l’#AI li rendesse più rapidi del 20%.

Realtà dei fatti: impiegavano il 19% di tempo in più.

Percezione vs realtà

🔗 https://metr.org/blog/2025-07-10-early-2025-ai-experienced-os-dev-study/?utm_source=perplexity

#llm #claudecode #chatgpt #codex #gemini #agents #agentsai

Künstliche Intelligenz verstärkt Arbeitsbelastung statt sie zu verringern

Entwickler, die KI-Tools wie Cursor Pro mit Claude 3.5/3.7 Sonnet nutzten, benötigten 19 Prozent länger für ihre Aufgaben als ohne KI-Unterstützung.

https://www.all-about-security.de/kuenstliche-intelligenz-verstaerkt-arbeitsbelastung-statt-sie-zu-verringern/

#METR #entwickler #ki #kitools

https://winbuzzer.com/2026/02/06/metr-five-hour-ai-claim-misunderstood-graph-xcxwbn/

METR's Five-Hour AI Claim: Why Everyone Misunderstood the Graph

#AI #METR #Anthropic #Claude #ClaudeOpus45 #AIResearch #AISafety #AIBenchmarks #LLMs #AICoding #AIAgents #AgenticAI #AICoding #AISafety #AIEthics

If AI coding is so good … where are the performance numbers?

https://fed.brid.gy/r/https://pivot-to-ai.com/2026/01/13/if-ai-coding-is-so-good-where-are-the-performance-numbers/

Интересное в графике - не то что 8 часовые задачи (с успешностью 50%) прогнозируются в ~середине этого года, а то, как уныло выглядит график, если переключить на 80% успешность (там нечто вроде 15 минут на начало 2026, а не 4.5 часа как на 50%).

https://metr.org/blog/2025-03-19-measuring-ai-ability-to-complete-long-tasks/

#LLM #METR #evals #llm_evals #ai_evals