AI/ML News & Innovations Hub

AI/ML news, top picks, and generated innovation digests.

★ Visit ai-karthik.com
426Sources
90239News Items
8Top Picks
445Blogs
failedLast Run

Latest AI/ML News

90239 matching items

The list of EVs killed off this year
Business Insider AI 2026-11-11 09:52 UTC Score 37.0 USR-0098-20261111-global-ai-ne-537c4dcf Full article

The list of EVs killed off this year

Carmakers are rejigging their EV lineups in the post tax credit world. Here's a list all-electrics canceled this year — including the Cybertruck.

AI Alignment Forum 2026-11-11 09:49 UTC Score 60.0 USR-0151-20261111-community-fo-37031e0c Full article

How I'm thinking about different types of automated alignment

I used to think of "automating alignment" as a natural category. But when I tried to think through the strategic implications, I realised that it depends a lot on the type of work. I'm now thinking of it in three buckets: Empirical alignment research : design and execution of alignment-relevant experiments (for a given notion of what we’re aiming for) Conceptual alignment research: theoretical or judgement-driven work to improve our understanding of what we’re aiming for and how to achieve it (including how to measure it) Direct alignment work : making particular AI systems more aligned (including via inputs such as training data) In the rest of this post I'll walk through my current takes on these in terms what’s needed for automation, how good or bad they look, what our priorities might be (to the extent we're interested in automating alignment [1] ), etc. Automating empirical alignment research This is the most obvious fit for the AI capabilities we’re getting at the moment. You want to know more about the relationship between some alignment-relevant variables? Great, an AI agent can design and run that experiment for you. You want to know what training methods help your alignment benchmark go up? In the not-too-distant future, AI agents will probably be able to do this too. When we have the automated capabilities researchers we seem on track for, this is exactly the kind of thing they’ll likely be able to do a lot of. Note that I’m not counting the design of benchmarks/m…

LessWrong AI 2026-11-11 09:49 UTC Score 82.0 USR-0152-20261111-community-fo-3e017448

How I'm thinking about different types of automated alignment

I used to think of "automating alignment" as a natural category. But when I tried to think through the strategic implications, I realised that it depends a lot on the type of work. I'm now thinking of it in three buckets: Empirical alignment research : design and execution of alignment-relevant experiments (for a given notion of what we’re aiming for) Conceptual alignment research: theoretical or judgement-driven work to improve our understanding of what we’re aiming for and how to achieve it (including how to measure it) Direct alignment work : making particular AI systems more aligned (including via inputs such as training data) In the rest of this post I'll walk through my current takes on these in terms what’s needed for automation, how good or bad they look, what our priorities might be (to the extent we're interested in automating alignment [1] ), etc. Automating empirical alignment research This is the most obvious fit for the AI capabilities we’re getting at the moment. You want to know more about the relationship between some alignment-relevant variables? Great, an AI agent can design and run that experiment for you. You want to know what training methods help your alignment benchmark go up? In the not-too-distant future, AI agents will probably be able to do this too. When we have the automated capabilities researchers we seem on track for, this is exactly the kind of thing they’ll likely be able to do a lot of. Note that I’m not counting the design of benchmarks/m…

AvenuesAI Q2: Profit rises 6% to Rs 72 cr, revenue surges 47%
Medianama AI 2026-11-11 09:44 UTC Score 40.0 USR-0211-20261111-regional-new-71f87071 Full article

AvenuesAI Q2: Profit rises 6% to Rs 72 cr, revenue surges 47%

AvenuesAI reported a 6% rise in net profit to Rs 71.7 crore in Q2 FY27, while revenue grew 47% to Rs 2,879.2 crore. The company also raised its FY27 revenue guidance to Rs 13,000 crore. The post AvenuesAI Q2: Profit rises 6% to Rs 72 cr, revenue surges 47% appeared first on MEDIANAMA .

JetBrains AI Blog 2026-11-11 09:38 UTC Score 41.0 USR-0065-20261111-ai-specialis-9a4967ac Full article

JetBrains IDEs Transformed for the Agentic Era of Development – With Air

As agents write more code, individual productivity goes up, but the bottleneck moves to review. 49% of developers now spend four or more hours a week reviewing AI output and correcting prompts (Atlassian State of Developer Experience Report 2026). We believe this means you need to understand what changed, decide whether it belongs in your […]

'제미나이'가 기억하는 내 정보 한눈에 보여준다…수정·삭제도 가능
Korea AI Times 2026-11-11 09:23 UTC Score 40.0 USR-0048-20261111-global-ai-ne-79bd014c Full article

'제미나이'가 기억하는 내 정보 한눈에 보여준다…수정·삭제도 가능

구글이 \'제미나이\'가 이용자에 대해 기억하는 정보를 한눈에 보여주는 \'메모리 요약(Memory summary)\' 기능을 도입했다. 이용자는 AI가 파악한 자신의 직업과 관심사, 생활 습관 등을 직접 확인하고 잘못된 내용을 수정하거나 삭제할 수 있다. \'챗GPT\'와 \'클로드\'에 이어 구글도 개인화 AI의 기억을 이용자가 직접 관리하도록 하는 기능을 강화한 것이다.나인투파이브구글은 9일(현지시간) 구글이 제미나이에 \'메모리 요약(Memory summary)\' 기능을 제공하기 시작했다고 보도했다. 웹 버전의 설정에서 \'개인 인텔리전스(P

'페이블 5.2' 프롬프트 가이드 공개…"코드 맡기면 정리 5배 증가"
Korea AI Times 2026-11-11 09:05 UTC Score 40.0 USR-0048-20261111-global-ai-ne-764caee5 Full article

'페이블 5.2' 프롬프트 가이드 공개…"코드 맡기면 정리 5배 증가"

앤트로픽이 최신 모델 \'클로드 페이블 5.2\'의 프롬프트 작성법을 공개하며 기존 모델에 사용하던 지침을 재검토하라고 권고했다. 특히 코드의 소유권을 어떻게 표현하느냐에 따라 AI의 작업 방식이 크게 달라지는 것으로 나타났다.앤트로픽은 10일(현지시간) 공식 개발자 문서를 통해 페이블 5.2의 성능과 비용을 최적화하기 위한 프롬프트 가이드를 공개했다. 장시간 작업의 조기 중단, 불필요한 승인 요청, 코드 수정 범위, 하위 에이전트 활용 등 실제 개발 과정에서 나타나는 문제와 해결 방법을 담았다.가장 눈길을 끄는 것은 코드 소유권에 관

InfoWorld AI 2026-11-11 09:00 UTC Score 29.0 USR-0126-20261111-global-ai-ne-59881b92 Full article

Software developers need an Agentic manifesto

I can’t decide whether Agile development is completely dead or about to make a huge comeback. Or maybe Waterfall will. Who knows? Coding agents are changing everything — and I don’t think we’ve quite figured out how they are going to update our methodology, or how we are going to organize ourselves to take advantage of them. We have barely scratched the surface. Folks who have been around a while will remember the Waterfall method for developing software. Because coding was the hardest part of any project, Waterfall called for figuring everything out before you even started writing code. You would write a huge specification, detailing everything, then turn it over to the development team. The development team would build to the spec and pass the software on to the testing team who would verify it, and finally the project would be delivered to customers. Never mind that by the time the thing was built and tested, what the customer wanted had probably changed dramatically. And if the customer decided they wanted a widget here and a sprocket there after the specification was written? Well, that would be a formal change request and would require a whole lot more time and money, thank you very much. Yeah, they really did it that way. Waterfall bad, Agile good It didn’t take long for everyone to figure out that the Waterfall way wasn’t going to cut it. Projects were almost always late and not at all what customers wanted. There had to be a better way, and eventually a bunch of sof…

InfoWorld AI 2026-11-11 09:00 UTC Score 57.0 USR-0126-20261111-global-ai-ne-66b5f789 Full article

MCP and A2A alone won’t scale AI agents

Two protocols define how AI agents connect to internal resources and collaborate with other AI agents. The Model Context Protocol (MCP) is well known because it’s promoted as a general protocol for connecting AI agents to resources and other AI agents. But that’s not completely accurate. It’s really the Agent2Agent protocol (A2A) that’s enabling AI agent orchestration . The confusion stems from how quickly platforms have adopted these protocols and marketed them as features. Users building their own AI agents with Claude, OpenAI, Microsoft Copilot, and other models are already being exposed to customizing skills, connectors, and other plug-ins that use MCP and A2A. Vibe coding and AI app-building tools like Bolt.new, Lovable, Pave, Replit, and v0 can connect to MCP servers and include built-in connectors. AI coding editors like Claude Code, Cursor, Devin Desktop, Gemini CLI, GitHub Copilot, and Kiro have MCP clients. These protocols will shape how enterprises scale and manage AI agents. Gartner predicts that the average global Fortune 500 enterprise will run more than 150,000 AI agents over the next two years. But according to Collibra’s 2026 Hallucination Tax Report , 76% have hit critical roadblocks while attempting to move from pilot to full production during the past 12 months. Reviewing the two protocols and which services are your responsibility to layer on top of them is key to scaling enterprise-ready AI agents. Differences between MCP and A2A Scott Gerlach, cofounde…

CIO AI 2026-11-11 09:00 UTC Score 30.0 USR-0125-20261111-global-ai-ne-09ec4283 Full article

The AI prototype worked. Who promised the launch?

“Sure, we can figure it out.” Earlier in my career, I used those words to avoid an uncomfortable conversation. I meant I would assess the request and explore what was possible. The person asking could hear that I had accepted the work and would deliver what they wanted. I learned to be explicit, sometimes to the point of repeating myself. What felt redundant to me helped someone else understand what I was actually promising. Consider this scenario. Joe, a colleague in operations, builds an AI assistant that answers questions about internal procedures. He tests it against a small collection of documents. His colleagues find the answers useful and want to use it. Their manager suggests launching across the department next month. IT agrees to assess how it would connect to the company’s knowledge base. By the next executive meeting, the update reads: “Assistant complete. Awaiting IT integration.” A demonstration has become a launch commitment. IT’s offer to investigate has become the remaining task. Clarify the promise before the date travels Before explaining what Joe’s assistant still needs, I would ask what his prototype reveals about us. He understood a recurring frustration: colleagues spending time searching for procedures or asking someone who knew where to look. He produced something they found useful. If that improvement would have spent months in our backlog, we should examine why. Did we underestimate the problem? Was our intake process too demanding for the size of…

구글, '제미나이 로보틱스 2.5' 공개…공장 작업 성공률 96%
Korea AI Times 2026-11-11 08:51 UTC Score 51.0 USR-0048-20261111-global-ai-ne-e172dc6b Full article

구글, '제미나이 로보틱스 2.5' 공개…공장 작업 성공률 96%

구글이 실제 자동차 공장에서 휴머노이드 로봇의 작업 성공률을 96%까지 높인 새로운 물리적 AI 모델을 공개했다. 기존 모델보다 정교한 손동작과 작업 속도를 개선하고, 최신 AI 모델을 로봇 훈련에도 활용해 산업 현장 적용 가능성을 높였다.구글 딥마인드는 10일(현지시간) 공식 블로그를 통해 최신 비전-언어-행동(VLA) 모델 \'제미나이 로보틱스 2.5(Gemini Robotics 2.5)\'를 발표했다. 로봇이 수행할 수 있는 작업의 범위를 넓히고, 새로운 동작을 배우는 속도와 작업의 정확성, 안전성을 개선한 것이 특징이다.지난 7

메타, AI 에이전트 '뮤즈' 애플워치·웨어OS로 확대
Korea AI Times 2026-11-11 08:41 UTC Score 38.0 USR-0048-20261111-global-ai-ne-69001d5f Full article

메타, AI 에이전트 '뮤즈' 애플워치·웨어OS로 확대

메타가 소비자용 AI 에이전트 \'뮤즈(Muse)\'를 애플워치와 구글의 웨어OS(Wear OS) 스마트워치로 확대했다. 지난 9월 출시 이후 스마트폰과 태블릿에 이어 웨어러블 기기까지 지원하면서, 이용자가 일상에서 AI 에이전트를 쉽게 사용할 수 있도록 서비스 접점을 늘리고 있다.메타는 9일(현지시간) X를 통해 스마트워치용 뮤즈 서비스를 발표했다. 사용자는 스마트워치에서 음성이나 문자로 AI 에이전트에 작업을 요청하고 답변과 진행 상황을 확인할 수 있다. 애플워치에서는 화면에 뮤즈 실행 버튼을 배치해 빠르게 서비스를 호출하는 기능도

문샷, 중국서 엔비디아 최신 AI 칩 임대 정황…금융 문서 확인
Korea AI Times 2026-11-11 08:33 UTC Score 41.0 USR-0048-20261111-global-ai-ne-ca84cd84 Full article

문샷, 중국서 엔비디아 최신 AI 칩 임대 정황…금융 문서 확인

중국 문샷AI가 미국의 중국 수출 규제 대상인 엔비디아의 최신 AI 칩 \'B300\'을 중국 내에서 임대하려 했다는 금융 계약 문서가 발견됐다. 문샷은 관련 거래를 부인했으며, 블룸버그의 취재 이후 공개 문서에서 문샷의 이름이 사라져 실제 거래 여부와 장비 반입 경로에 관심이 모이고 있다.블룸버그는 10일(현지시간) 중국 금융기관에 제출된 계약 문서를 확인한 결과, 문샷에 엔비디아 B300 GPU를 탑재한 슈퍼마이크로 서버 32대의 컴퓨팅 서비스를 제공하는 계약이 명시돼 있었다고 보도했다. 해당 서버는 중국 서부 간쑤성 칭양시에 보관

LessWrong AI 2026-11-11 08:12 UTC Score 50.0 USR-0152-20261111-community-fo-7a6273cf

Transfinite Procrastination

Heavily inspired by a conversation with Keshav Srinivasan and Richard Ngo during MATS 10.0. The week ahead of you is packed. This isn't your usual brand of busy—you're phoneless, so you can't check the details, but (as far as you can remember) every hour of every day is spoken for. Tonight will be particularly hectic… You're mid-conversation when Someone Important walks by. They pause, turn around, and ask: "Hey, when are you going to finish That Thing you said you'd do?" Level 0 procrastination: "I'll finish it by tonight!" Problem: you're busy tonight. Level 1: "I'll finish it by tomorrow!" Problem: one day is obviously not enough to finish That Thing. And besides, you might be really busy tomorrow! Level 5: "I'll finish it within 5 days!" Problem: this is starting to feel hopeless—you don't have your calendar on you right now, and you barely even have an order-of-magnitude estimate for how long doing That Thing will take you! Would even 100 days be enough? … Eureka! Level procrastination: "By tonight, I'll let you know when I will finish That Thing by!" If we had said "I'll finish it within days" for any finite , we would have only days in which to complete That Thing; by choosing level instead, we have attained the freedom to mull things over tonight (with our calendar!), decree "I'll finish it within days" then, and subsequently enjoy the extra day of free working time! The symbol refers to the smallest transfinite ordinal , heuristically defined to be a new symbol such…

TSMC, 자체 칩 개발 나선 오픈AI·앤트로픽에 최대 1.3조원 투자
Korea AI Times 2026-11-11 08:05 UTC Score 40.0 USR-0048-20261111-global-ai-ne-3fc0b915 Full article

TSMC, 자체 칩 개발 나선 오픈AI·앤트로픽에 최대 1.3조원 투자

세계 최대 파운드리 업체 TSMC가 자체 AI 반도체 개발에 나선 오픈AI와 앤트로픽에 최대 10억달러(약 1조3400억원)를 투자하기로 했다. 두 AI 기업이 엔비디아 등 외부 반도체 업체에 대한 의존도를 낮추기 위해 자체 칩 개발을 추진하면서, TSMC도 반도체 공급을 넘어 이들 기업의 지분을 확보하며 관계를 강화하려는 움직임으로 해석된다.TSMC 이사회는 10일 앤트로픽 주식 매입에 최대 5억달러(약 6680억원), 오픈AI 우선주 투자에 최대 5억달러를 각각 승인했다. 앤트로픽 투자는 기업공개(IPO) 공모가를 기준으로 보통

[게시판] 패스트캠퍼스·엔비디아 해커톤서 신약개발 AI 에이전트 우승 등 단신
Korea AI Times 2026-11-11 08:00 UTC Score 40.0 USR-0048-20261111-global-ai-ne-270c5383 Full article

[게시판] 패스트캠퍼스·엔비디아 해커톤서 신약개발 AI 에이전트 우승 등 단신

■ 데이원컴퍼니(대표 신해동·김동혁)의 실무 교육 브랜드 패스트캠퍼스는 엔비디아와 공동 개최한 ‘에이전틱 AI 해커톤’ 결선을 ‘엔비디아 AI 데이 서울 2026’에서 진행했다고 밝혔다. 최종 우승은 신약개발·의료 AI 연구자와 현직 약사로 구성된 ‘플라이게이트’ 팀이 차지했다. 엔비디아 ‘바이오니모 NIM’과 ‘네모트론’을 활용해 신약개발부터 시판 후 안전성 검토까지 서로 다른 업무를 하나의 AI 에이전트 워크플로로 연결해 주목받았다.■ 코난테크놀로지(대표 김영섬)는 AI 기반 다국어 안내 시스템 ‘링고엑스’의 운영 성과를 부산교

소프트뱅크, AI 발전 설비 선주문에 5조 손실…분기 적자 9조
Korea AI Times 2026-11-11 07:56 UTC Score 38.0 USR-0048-20261111-global-ai-ne-9006b3db Full article

소프트뱅크, AI 발전 설비 선주문에 5조 손실…분기 적자 9조

소프트뱅크가 미국 AI 데이터센터에 필요한 발전 설비를 대규모로 미리 주문했다가 약 5조원의 손실을 입었다. 전력 설비가 부족해질 것에 대비해 거액의 선급금을 지급했지만, 사업 계획이 바뀌면서 일부 계약을 취소하거나 변경해야 했기 때문이다. 여기에 투자 자산의 가치 하락과 금융 비용 증가까지 겹쳐 2년 만에 분기 적자를 기록했다.소프트뱅크는 10일 2026 회계연도 2분기(7~9월)에 70억달러(약 9조원)의 순손실을 기록했다고 발표했다. 전년 동기 157억달러의 순이익에서 적자로 돌아선 것으로, 시장 예상치보다 크게 나쁜 실적이다

구글, AI로 암호 약점 발견 이어 시그널 보안 검증…650만줄 증명
Korea AI Times 2026-11-11 07:49 UTC Score 40.0 USR-0048-20261111-global-ai-ne-39462f43 Full article

구글, AI로 암호 약점 발견 이어 시그널 보안 검증…650만줄 증명

구글이 AI를 활용해 RSA 암호의 약점을 찾아낸 데 이어, 이번에는 메신저 시그널(Signal)의 실제 암호화 코드가 핵심 보안 조건을 충족한다는 사실을 수학적으로 입증했다. 오픈AI와 앤트로픽이 AI의 암호학·수학 연구 성과를 잇달아 공개하는 가운데, 구글도 암호 분석에서 실제 소프트웨어의 보안 검증까지 AI 활용 범위를 넓히고 있다.구글 딥마인드는 10일(현지시간) 공식 블로그를 통해 다중 AI 에이전트 기반 정형 검증 시스템 \'알파프루프 코드(AlphaProof Code)\'로 시그널의 핵심 암호화 라이브러리인 \'립시그널(li

Medianama AI 2026-11-11 07:48 UTC Score 33.0 USR-0211-20261111-regional-new-90b42e72 Full article

Madras High Court flags fake, AI-generated books in public libraries, orders FIR

The Madras High Court has ruled that AI-generated and plagiarised books cannot be procured for public libraries and ordered an FIR into alleged fraudulent book purchases in Tamil Nadu The post Madras High Court flags fake, AI-generated books in public libraries, orders FIR appeared first on MEDIANAMA .

New aviation deals raise flight limits between China and Central Asia
Euronews AI 2026-11-11 07:45 UTC Score 40.0 AI-164-20261111-regional-ai--d84dbd3e Full article

New aviation deals raise flight limits between China and Central Asia

Uzbekistan and Kazakhstan agreed to higher weekly flight limits with Beijing, while a separate agreement allows more services between Tashkent, Almaty and Astana. Airlines are also exploring new regional routes, although rising operating costs remain a concern.

X ‘locked’ Times of India photojournalist’s account after Delhi smog posts: Report
Medianama AI 2026-11-11 07:45 UTC Score 40.0 USR-0211-20261111-regional-new-b88cbf13 Full article

X ‘locked’ Times of India photojournalist’s account after Delhi smog posts: Report

While the TOI photojournalist said it was his duty to show "the truth of the post-Diwali situation" in Delhi through his camera, X cited a "violation of its rules" as the reason for locking the account. The post X ‘locked’ Times of India photojournalist’s account after Delhi smog posts: Report appeared first on MEDIANAMA .

Democrats urge Rubio to act on Ethiopia
Semafor Technology 2026-11-11 07:43 UTC Score 49.0 USR-0094-20261111-global-ai-ne-fffae22a Full article

Democrats urge Rubio to act on Ethiopia

Leading Senate Democrats want the US to restrict arms to Ethiopia and expand TPS amid escalating tensions.

오픈AI, 오데시아 인수하며 여행 AI 서비스 개발 본격화
Korea AI Times 2026-11-11 07:29 UTC Score 38.0 USR-0048-20261111-global-ai-ne-074be8b3 Full article

오픈AI, 오데시아 인수하며 여행 AI 서비스 개발 본격화

오픈AI가 여행 전문 AI 에이전트 스타트업 오데시아(Odessia)를 인수했다. 에어비앤비와 유사한 숙박 스타트업 손더(Sonder)를 창업한 프랜시스 데이비드슨도 영입해 여행 분야 AI 서비스 개발에 나설 전망이다.데이비드슨 창립자는 10일(현지시간) X를 통해 오픈AI에 합류해 \"여행의 미래를 만드는 일을 돕게 됐다\"고 밝혔다.구체적인 인수 조건은 공개되지 않았지만, 창업자와 핵심 인력을 영입하는 인재 인수(Acqui-hire)에 가까운 거래로 풀이된다.오데시아는 약 1년 전 설립된 스타트업으로, 세쿼이아 캐피털이 주도한 60

‘Constructive’ Abu Dhabi talks resume, but a ceasefire is nowhere near
Euronews AI 2026-11-11 07:21 UTC Score 40.0 AI-164-20261111-regional-ai--a32d0ee7 Full article

‘Constructive’ Abu Dhabi talks resume, but a ceasefire is nowhere near

Good morning. I’m Luca Bertuzzi, holding the pen for today’s newsletter. In this edition: a preview of the second day of peace talks in Abu Dhabi, the backlash to Merz’s budget gambit, and why aid to Ukraine is becoming a sore point for several EU countries.

중기부, 2030년까지 ‘AI 유니콘 20곳·데카콘 1곳' 키운다
Korea AI Times 2026-11-11 07:17 UTC Score 40.0 USR-0048-20261111-global-ai-ne-346f2af9 Full article

중기부, 2030년까지 ‘AI 유니콘 20곳·데카콘 1곳' 키운다

정부가 2030년까지 기업가치 1조 원 이상 AI 유니콘 20곳과 10조 원 이상 데카콘 1곳을 키운다.중소벤처기업부는 11일 서울 용산구 그랜드 하얏트 서울에서 열린 딥테크 스타트업 포럼 \'딥 다이브 인투 2026\'에서 국가인공지능전략위원회와 함께 ‘AI 스타트업 비욘드(Beyond) 유니콘 육성 전략\'을 발표했다.이번 육성 전략은 발굴에서 도약까지 전주기 지원, 혁신성장 가속화, 글로벌 시장 도약 촉진 등 3대 전략과 12개 세부과제로 구성됐다. 매년 50곳 규모의 ‘AI 초기 유니콘\' 후보군을 발굴하고, 투자·고용 등에서 상위

On the Computational Tractability of Robust Bandits
LessWrong AI 2026-11-11 07:11 UTC Score 65.0 USR-0152-20261111-community-fo-0b8189ad Full article

On the Computational Tractability of Robust Bandits

Vanessa Kosoy and I recently uploaded a new paper to arXiv. In this post, I briefly describe what the paper does, and attempt to explain in a self-contained way why this is a useful thing to do for ASI alignment. For a more complete background, see several previous articles . The technical contribution of the paper is two-fold. First, we identify a class of instances of the robust bandits problem that can be learned both statistically and computationally efficiently. Previous work on the problem had only identified a class that was statistically tractable. Second, we show that most ways to generalize this class of instances makes the problem NP-hard. Thus in some sense this class is at the boundary of what's computationally tractable. But what does the computational tractability of robust bandits have to do with ASI alignment? The thesis motivating this paper is that for ASI alignment, we need to build learning algorithms that: a) come with proofs of their statistical and computational efficiency, and b) work in environments where classical learning theoretic frameworks fail to capture phenomena that arise due to embedded agency. The "robust" in robust bandits is a way to handle embedded agency, and similar ideas have been used before for learning problems other than bandits , albeit with different names. Bandits are one of the simplest learning problems to study that still involve interactions with the environment, thus they are a good stepping stone before solving robust v…