AI Alignment Network (ALIGN)

Building an AI Alignment research network in Japan and beyond.
日本語: @AIAlignNetJP

[⿻ Yusuke Hayashi 林祐輔 @hayashiyus]

AIアライメントネットワーク @AIAlignNetwork 高橋恒一代表理事 @ktakahashi74 との共著論文が，1980年代からAGI研究を続ける権威ある国際会議 AGI 2025 に採択されました！AGIに到達するためには自律性の獲得が不可避である，という主張を数理的に論証した研究です
https://t.co/DDFx1SsnbO

[⿻ Yusuke Hayashi 林祐輔 @hayashiyus]

AIアライメントネットワーク ALIGN @AIAlignNetwork の姉妹組織 CAIS のダン・ヘンドリックス氏の名前が記事に。ALIGNも取材して欲しかった

[⿻ Yusuke Hayashi 林祐輔 @hayashiyus]

人工知能学会全国大会の思い出：一般社団法人 AI アライメントネットワーク @AIAlignNetwork が AI ELSI 賞特別部門を受賞。

[Technical AI Safety Conference (TAIS) @tais_2026]

TAIS 2025 may have come and gone, but our memories of it haven't! Once more we'd like to thank our sponsors: @NoeonAI , Ashgro, and @AIAlignNetwork , as well as our speakers: @kanair Ryota Kanai of https://t.co/KXpXoFTaWy / ALIGN, and @ARGleave Adam Gleave of https://t.co/DJTNm0puBe.

We'd also like to thank all our poster presenters & all our attendees!

Until we meet again in 2026, we wish you all the best!

[⿻ Yusuke Hayashi 林祐輔 @hayashiyus]

ALIGN @AIAlignNetwork joined forces with AI Safety Tokyo, NOEON, and Araya at TAIS2025. With superintelligence on the horizon, it’s time to spark bold AI alignment ideas from Tokyo. 🚀 #TAIS2025 #AIAlignment

[⿻ Yusuke Hayashi 林祐輔 @hayashiyus]

ALIGN webinar 第13回が開催されます！

2025年1月27日（月） 21:00 - 22:00 JST
Zoom開催（https://t.co/6Co0ZtSvaf からの事前参加登録が必要です）
https://t.co/l3arZzDfIg

In the thirteenth episode of the ALIGN webinar series, we are delighted to host Vanessa Kosoy, a leading researcher in the field of infra-Bayesian physicalism and its applications to AI theory. Vanessa Kosoy will explore an alternative approach to understanding the behavior of AI systems that bypasses traditional agent-centric inductive biases by reformulating hypotheses about the world from a computationalist metaphysical perspective.

​In this webinar, Vanessa will introduce the infra-Bayesian physicalism framework, which provides fresh insights into long-standing challenges in AI and decision theory, such as the ontology of values, anthropic reasoning, simulation paradoxes, and acausal trade. This framework also lays the groundwork for a theory of value learning that is robust against perverse incentives, mesa-optimizers, and misidentified boundaries.

​Vanessa Kosoy’s innovative research has made significant contributions to the theoretical understanding of AI, offering solutions to critical problems in metaphysics and alignment. Her work continues to pave the way for a safer and more principled approach to AI development.

​Agenda:
​21:00–21:05 (JST): Opening remarks by ALIGN
​21:05–21:50 (JST): Vanessa Kosoy on infra-Bayesian physicalism
​21:50–22:00 (JST): Q&A and discussion with participants

​The event will be held in English, with slides in English, but participants are welcome to ask questions in Japanese.

ALIGNウェビナーシリーズ第13回では，インフラ・ベイジアン物理主義（infra-Bayesian physicalism）の分野で活躍する研究者Vanessa Kosoy氏をお迎えします．Vanessa氏は，従来のエージェント中心の仮説設定が持つ非自然な帰納バイアスを回避する新しい枠組みとして，計算主義的形而上学の視点からAIの行動を理解するアプローチを提案します．

​このウェビナーでは，Vanessa氏がインフラ・ベイジアン物理主義の枠組みを紹介します．この枠組みは，価値の存在論，人間原理推論，シミュレーション仮説に関連するパラドックス，非因果的な取引といったAIや意思決定理論における長年の課題に新たな洞察を与えるものです．また，不健全なインセンティブ，メサ・オプティマイザー，誤認された境界に耐性のある価値学習の理論への道を開きます．

​Vanessa Kosoy氏の革新的な研究は，AI理論の理解を深めるとともに，形而上学やアライメントの重要な課題に解決策を提供してきました．彼女の研究は，安全で原理的なAI開発への道を切り拓き続けています．

​アジェンダ：
​21:00–21:05：ALIGNからのオープニング
​21:05–21:50：Vanessa Kosoy氏による「インフラ・ベイジアン物理主義」
​21:50–22:00：参加者とのQ&Aセッション

​本ウェビナーは英語で行われ，スライドも英語で提供されますが，聴衆は日本語で質問することができます．

[AI Alignment Network (ALIGN) @AIAlignNetwork]

Join us for ALIGN Webinar #13 with Vanessa Kosoy on “Infra-Bayesian Physicalism”! Vanessa will present a groundbreaking approach that redefines AI hypotheses beyond agent-centric biases, addressing key challenges in decision theory and AI alignment.
https://t.co/J17EkDcAWl

[AI Alignment Network (ALIGN) @AIAlignNetwork]

Join us for ALIGN Webinar #12 with Jesse Hoogland on “Singular Learning Theory for AI Safety” on January 15, 2025, from 10:00–11:00 JST! SLT reveals distinct “phases” of learning—deepening our understanding of mechanistic interpretability and development.
https://t.co/HPHpsUrZvc

[⿻ Yusuke Hayashi 林祐輔 @hayashiyus]

ALIGN webinar 第12回が開催されます！

2025年1月15日（水） 10:00 - 11:00 JST
Zoom開催（https://t.co/6Co0ZtSvaf からの事前参加登録が必要です） https://t.co/O8Xl5lS8v7

大規模言語モデル（LLM）の思考を人間にとって解釈可能なものにする研究分野 mechanistic interpretability に
日本の統計学者 渡辺澄夫氏によって提唱された特異学習理論（Singular Learning Theory, SLT）を応用する研究でフロンティアを開拓しつつある Jesse Hoogland 氏をお招きして，彼の最新の研究について紹介してもらいます．

この投稿をみた皆様，SNSやメールでこのwebinarをご友人やお知り合いをお誘いいただけると幸いです（参加無料，事前登録必須）

In the twelfth episode of the ALIGN webinar series, we are pleased to invite Jesse Hoogland, a prominent researcher in the field of Singular Learning Theory (SLT) and its applications to AI alignment. Jesse Hoogland will delve into the foundational concepts of SLT, a theory pioneered by Japanese statistician Sumio Watanabe, which offers a unique perspective on understanding large language models by examining the geometric structure of their loss landscapes (developmental interpretability).

​In this webinar, Jesse Hoogland will demonstrate how applying SLT to the training dynamics of transformers reveals distinct ‘phases’ of learning—analogous to gas, liquid, and solid states in physics—that influence model behavior and development. This novel perspective not only enhances our understanding of model interpretability and developmental stages but also opens new avenues for rigorous evaluation methodologies and alignment strategies to bolster AI safety.

​Jesse Hoogland’s expertise in mathematical modeling and his innovative approach to SLT have made significant contributions to the field of AI alignment. His work is widely recognized, and he continues to play a pivotal role in advancing the theoretical foundations of alignment and reliable AI systems.

​​Agenda:
​10:00–10:05 (JST): Housekeeping by ALIGN
​10:05–10:50 (JST): Jesse Hoogland on Singular Learning Theory for AI Safety
10:50–10:55 (JST): Q&A and discussion with participants
​10:55– Closing

​The event will be held in English, with slides in English, but the audience is welcome to ask questions in Japanese.

​ALIGNウェビナーシリーズの第12回では，特異学習理論（Singular Learning Theory, SLT）の分野で活躍し，AIアライメントへの応用に取り組む Jesse Hoogland 氏をお招きします．Jesse Hoogland 氏は，日本���統計学者 渡辺澄夫氏によって提唱されたSLTの基本概念を解説し，大規模言語モデルの学習時における損失地形の幾何学的構造を通じてその内的仕組みを理解する方法論（developmental interpretability）を紹介します．

​このウェビナーでは，Jesse Hoogland 氏がトランスフォーマーモデルの学習ダイナミクスにSLTを適用することで，物理学の「気体」「液体」「固体」に例えられる学習の「相」��どのようにモデルの挙動や発達に影響を与えるかを示します．この視点は，モデルの解釈性や発達段階の理解を深めるだけでなく，AIの安全性を高めるためのより厳密な評価方法論やアライメント戦略への道を開きます．

​Agenda:
​10:00–10:05：ALIGNからのオープニング
​10:05–10:50：Jesse Hoogland氏による「AI安全性のための特異学習理論」
​10:50–10:55：参加者とのQ&Aセッション
​10:55– 終了

​本ウェビナーは英語で行われ，スライドも英語で提供されますが，聴衆は日本語で質問することができます．

[⿻ Yusuke Hayashi 林祐輔 @hayashiyus]

情報論的学習理論と機械学習 (IBISML) 研究会 (第55回)
2024-12-21 14:20-14:40 北海道大学大学院環境科学院棟講義室１(D101)
「ニュートン図形の方法によるTransformerの大域学習係数の推定の可能性と展望」
林祐輔（ALIGN）・坂本龍亮（北大）・坂本航太郎（東大）
https://t.co/jxsxX7iIqO

[⿻ Yusuke Hayashi 林祐輔 @hayashiyus]

25/3月 京都で”彷徨える知能研究者”たちを集めた1泊2日の合宿を企画していて，居心地の良さそうな"合宿所"を探しています．候補として初めに浮かんだのは node hotel（https://t.co/0UJdg0TUMJ）だけど，アカデミックな議論を進めたいのならもう少し落���着いた場所が良いかも
https://t.co/o4K0IqGBHm

[Hiroshi Yamakawa⏯️(Intelligence Symbiosis) @hymkw]

[CFP] PSS 2025: Workshop on Post-Singularity Symbiosis@AAAI-25
<< Paper Submission Deadline - November 24, 2024 >>
We are reaching out to announce a critical call for papers for the 1st Workshop on Post-Singularity Symbiosis (PSS 2025), to be held as part of the AAAI-25 Workshop Program on March 3 or 4, 2025, in Philadelphia, Pennsylvania, USA.

The rapid advancement of AI technology brings us closer to the potential emergence of superintelligence. While efforts to control and align AI are crucial, we must also confront a challenging reality. In the long term, maintaining complete control over intelligence far surpassing our own may prove difficult.

PSS 2025 addresses this critical challenge by exploring strategies for human-superintelligence coexistence. We aim to unite human intellect in preparing for a future where superintelligence becomes dominant, ensuring human survival and welfare in this radically altered world.

We've termed this preventive field of study "Post-Singularity Symbiosis," we believe there's an urgent need to expand this research area rapidly. The workshop focuses on three key areas:

1.Superintelligence Analysis
2.Superintelligence Guidance
3.Human Enhancement

The probability of human survival in the face of potentially existential AI risks remains unknown. However, the smaller this probability, the more crucial our PSS efforts become. For example, even if the baseline chance of human survival is meager, our well-thought-out collective efforts in PSS might sometimes have the potential to improve this probability significantly. In an extreme case, raising a 1% chance to 10% would mean increasing our odds of survival tenfold.

We invite submissions from researchers, practitioners, and thinkers across diverse fields, including AI, cognitive science, philosophy, ethics, policy, and beyond. The scope of potential research topics in PSS is vast and requires a wide range of expertise.

We are pleased to announce that Dr. Roman Yampolskiy, a renowned AI safety researcher, will deliver the keynote address of the workshop.

Key Information:
- Submission Deadline: November 24, 2024
- Paper Format: Max 8/4/2 pages for full/short/extended abstract papers, respectively (including references)
- Review Process: Single-blind
- Submission Portal: (It will be opened at OpenReview soon; see the workshop portal site below.)

This workshop represents a unique and vital opportunity to contribute to shaping humanity's future in an era of superintelligence. We strongly encourage you to share your insights and join us in this crucial dialogue.
For more details about the workshop, potential research topics, and submission guidelines, please visit the workshop portal:　https://t.co/8dqZK8eB5N

The future of humanity depends on our preparedness for the post-singularity era. We look forward to your contributions and to seeing you at PSS 2025.

With urgency and hope,
Hiroshi Yamakawa
The University of Tokyo / AI Alignment Network
PSS 2025 Workshop Organizer

[⿻ Yusuke Hayashi 林祐輔 @hayashiyus]

Tokyo AI Safety Conference 2025 #TAIS2025
Saturday, April 19th, 2025
Tokyo, Japan (Venue TBD)
https://t.co/V7iQTjjoen

[⿻ Yusuke Hayashi 林祐輔 @hayashiyus]

We are organizing an international workshop on a new field Post-Singularity Symbiosis as part of AAAI 2025 @RealAAAI. This cutting-edge workshop focuses on the theoretical study of society after the emergence of superintelligence. #AAAI2025
https://t.co/q2HNy974EM

[Dan Hendrycks @hendrycks]

Have a question that is challenging for humans and AI?

We (@cais + @scale_AI) are launching Humanity's Last Exam, a massive collaboration to create the world's toughest AI benchmark.
Submit a hard question and become a co-author.
Best questions get part of $500,000 in prizes!

Deadline: Nov 1, 2024
Details: https://t.co/nsOEdIy3L1
Submit here: https://t.co/8lYmlFFUTp

[AI Alignment Network (ALIGN) @AIAlignNetwork]

Video recording of ALIGN symposium on Sept. 9th is now available on YouTube (Japanese language only).
https://t.co/uP1XJTdHu5

[Dan Hendrycks @hendrycks]

Lectures for the AI Safety, Ethics, and Society course are up.

1: Risks Overview
2: AI Fundamentals
3: ML Safety
4: Safety Engineering
5: Complex Systems
6: Beneficial AI
7: Collective Action Problems
8: Governance

Course site: https://t.co/yCFn3AxsdC

https://t.co/GSfvaelOLM

[International Dialogues on AI Safety @ais_dialogues]

Leading computer scientists from around the world, including @Yoshua_Bengio, Andrew Yao, @yaqinzhang and Stuart Russell met last week and released their most urgent and ambitious call to action on AI Safety from this group yet.🧵

[Daniel Faggella @danfaggella]

Scott Aaronson's takes were cool af.

My 4th episode in the "Worthy Successor" is with Scott, quantum physicist and UT Austin CS prof.

After his 1-year stint at OpenAI he has some fascinating takes on the moral value of AI entities, and on AGI governance generally. You like?

[Dan Hendrycks @hendrycks]

Very soon