Does AI deserve compassion? AI welfare talk BuzzRobot 2026-09-21 | Does AI deserve compassion? AI welfare talk
The future of AI investments BuzzRobot 2025-10-10 | If AI labs don't start producing trillions in revenue by the end of the decade, investors will hesitate to put more money into the technology.Watch the full video on our channel!BuzzRobot asked Jaime Sevilla, director of Epoch AI and AI researcher, about the future of AI and superintelligence, and how AI will affect our jobs, day-to-day life, scientific progress, and more.
AI needs to disrupt the economy to be viable in the future BuzzRobot 2025-10-09 | Could investors lose interest in AI and what should happen to prevent that?Jaime Sevilla, director of Epoch AI and AI researcher, shares his forecasts for AI future. Catch these talks live and ask your own questions, join the BuzzRobot community: join.slack.com/t/buzzrobot/shared_invite/zt-37g5q0ao5-eMK_iDf0n4LAsh1d2qJYnQThis is a clip from a longer video with Jaime Sevilla that you can watch on our channel.Join BuzzRobot:Newsletter: buzzrobot.substack.com X: https://x.com/sopharicksSlack: join.slack.com/t/buzzrobot/shared_invite/zt-37g5q0ao5-eMK_iDf0n4LAsh1d2qJYnQ
Gemini 2.5 solves a Pokemon problem that was not in the training #ai BuzzRobot 2025-10-06 | Check out the full video on our channel!BuzzRobot spoke to Kiran Vodrahalli, a Research Scientist at Google DeepMind and core contributor on Gemini, about long-context reasoning in AI and the recent Gemini 2.5 report, part of which was about Gemini playing Pokémon Blue.Gemini 2.5: Pushing the Frontier with Advanced Reasoning, Multimodality, Long Context, and Next Generation Agentic Capabilities report can be found here: arxiv.org/abs/2507.06261Self-Improvement: github.com/waylaidwanderer/gemini-plays-pokemon-public/tree/default
AI spends tokens to panic while playing Pokemon #ai BuzzRobot 2025-10-05 | Check out the full video on our channel!BuzzRobot spoke to Kiran Vodrahalli, a Research Scientist at Google DeepMind and core contributor on Gemini, about long-context reasoning in AI and the recent Gemini 2.5 report, part of which was about Gemini playing Pokémon Blue.The Making of Gemini Plays Pokémon blog post: https://blog.jcz.dev/the-making-of-gemini-plays-pokemon
Long-context reasoning vs. retrieval capability in AI models #ai BuzzRobot 2025-10-03 | Watch the full video on our channel!BuzzRobot interviewed Kiran Vodrahalli, a Research Scientist at Google DeepMind and core contributor on Gemini, about long-context reasoning in AI and the recent Gemini 2.5 report, part of which was about Gemini playing Pokémon Blue.The Making of Gemini Plays Pokémon blog post: https://blog.jcz.dev/the-making-of-gemini-plays-pokemon
Google DeepMind researcher on understanding machines BuzzRobot 2025-10-02 | Do we really understand how AI works? Can AI teach grandmasters new tricks?Catch these talks live and ask your own questions, join the BuzzRobot community: join.slack.com/t/buzzrobot/shared_invite/zt-37g5q0ao5-eMK_iDf0n4LAsh1d2qJYnQThis time BuzzRobot spoke with Been Kim, senior staff research scientist at Google DeepMind, about interpreting and understanding machines and what cutting-edge research is being conducted to better understand emerging behavior of AI.Follow Been Kim: beenkim.github.ioWe Can't Understand AI Using our Existing Vocabulary research: arxiv.org/abs/2502.07586Join BuzzRobot:Newsletter: buzzrobot.substack.com X: https://x.com/sopharicksSlack: join.slack.com/t/buzzrobot/shared_invite/zt-37g5q0ao5-eMK_iDf0n4LAsh1d2qJYnQSupport us: ko-fi.com/sophiaaryanTimestamps:0:00 Intro0:18 AI teaches grandmasters chess05:30 AI neologisms10:17 Extracting knowledge from machines13:43 Interpretability research17:19 Are we keeping up with AI progress18:34 New AI related terms20:26 The right direction for interpretability23:35 AI lying27:49 Conseptual maps30:30 AI researchers bias33:33 Generalizing AI teaching humans35:11 Is AI sentient36:04 Does AI has concepts41:53 Progress in machine understanding
AI builds its own tools to play Pokemon #ai BuzzRobot 2025-09-26 | Check out the full video on our channel!BuzzRobot spoke to Kiran Vodrahalli, a Research Scientist at Google DeepMind and core contributor on Gemini, about long-context reasoning in AI and the recent Gemini 2.5 report, part of which was about Gemini playing Pokémon Blue.The Making of Gemini Plays Pokémon blog post: https://blog.jcz.dev/the-making-of-gemini-plays-pokemon
Open source AI: doomed or the biggest disruptors? BuzzRobot 2025-09-25 | What happened to one of the biggest open source LLMs, GPT-J? Watch the full interview on our channel!We spoke with the creator of GPT-J, Aran Komatsuzaki, about ChatGPT 5, open source AI models, how GPT-J started and why it ended, and how to fix the AI energy consumption issue. Catch these talks live and ask your own questions, join the BuzzRobot community: join.slack.com/t/buzzrobot/shared_invite/zt-37g5q0ao5-eMK_iDf0n4LAsh1d2qJYnQLearn more about GPT-J here: arankomatsuzaki.wordpress.com/2021/06/04/gpt-jJoin BuzzRobot:Newsletter: buzzrobot.substack.com X: https://x.com/sopharicksSlack: join.slack.com/t/buzzrobot/shared_invite/zt-37g5q0ao5-eMK_iDf0n4LAsh1d2qJYnQTimestamps:0:00 The start of GPT-J1:37 What happened to GPT-J3:06 Is open source viable?5:38 AI electricity issue 7:23 Chat GPT 5 impressions
What will the future of AI look like? BuzzRobot 2025-09-18 | Does AI have the potential to transform the economy? What would actually happen after we've reached the AGI era?Catch these talks live and ask your own questions, join the BuzzRobot community: join.slack.com/t/buzzrobot/shared_invite/zt-37g5q0ao5-eMK_iDf0n4LAsh1d2qJYnQIn this video, Buzzrobot asked Jaime Sevilla, director of Epoch AI and AI researcher, about the future of AI and superintelligence, and how AI will affect our jobs, day-to-day life, scientific progress, and more. AI companies should generate trillions in revenue to be able to scale, and for investors to have confidence in the technology. Will that be possible and what future predictions can be made right now - watch the interview to find out more.Timestamps:0:00 What AI will look like by 2035?4:57 AI agents and their limitations9:12 Trillions of dollars investments into AI infra13:57 Big tech to generate trillions of dollars revenue16:32 Economic impact of AI23:19 Geopolitical implications of AI35:18 AI arms race39:19 AI innovations to take us to SuperintelligenceJoin BuzzRobot:Newsletter: buzzrobot.substack.com X: https://x.com/sopharicksSlack: join.slack.com/t/buzzrobot/shared_invite/zt-37g5q0ao5-eMK_iDf0n4LAsh1d2qJYnQ
Gemini 2.5 completing Pokémon Blue and what it tells us about long-context reasoning in AI BuzzRobot 2025-09-11 | Google got Gemini to play Pokémon without pre-trainig to analyze its performance and ability to use and create tools for puzzle solving.Catch these talks live and ask your own questions, join the BuzzRobot community: join.slack.com/t/buzzrobot/shared_invite/zt-37g5q0ao5-eMK_iDf0n4LAsh1d2qJYnQBuzzRobot spoke to Kiran Vodrahalli, a Research Scientist at Google DeepMind and core contributor on Gemini, about long-context reasoning in AI and the recent Gemini 2.5 report.Follow Kiran on GitHub: kiranvodrahalli.github.io/aboutLinks mentioned in the video:Gemini 2.5: Pushing the Frontier with Advanced Reasoning, Multimodality, Long Context, and Next Generation Agentic Capabilities report can be found here: arxiv.org/abs/2507.06261Michelangelo paper: arxiv.org/abs/2409.12640OpenAI-MRCR benchmark: contextarena.aiGemini Plays Pokemon on stream: twitch.tv/gemini_plays_pokemonSelf-Improvement: github.com/waylaidwanderer/gemini-plays-pokemon-public/tree/default The Making of Gemini Plays Pokémon blog post: https://blog.jcz.dev/the-making-of-gemini-plays-pokemonGemini 2.5 showcasing creativity while playing Pokémon: https://x.com/K_Ishi_AI/status/1935155673966461317Join BuzzRobot:Newsletter: buzzrobot.substack.com X: https://x.com/sopharicksSlack: join.slack.com/t/buzzrobot/shared_invite/zt-37g5q0ao5-eMK_iDf0n4LAsh1d2qJYnQ
The main problem with open sourcing AI models #ai BuzzRobot 2025-09-07 | We spoke with the creator of one of the earliest open source LLMs GPT-J, Aran Komatsuzaki, about why GPT-J is no longer active. At the time of its release it was the largest GPT-3-style language model. GPT-J is a six billion parameter open source English autoregressive language model trained on the Pile.Check out the full video on our channel!
The reality of open source AI in 2025 BuzzRobot 2025-09-04 | The best way to make an impact in AI is to open source. We spoke with the creator of one of the earliest open source LLMs GPT-J, Aran Komatsuzaki, about ChatGPT 5, AI trends, open source AI models, jobs in AI research, how GPT-J started and why it ended, and more. Catch these talks live and ask your own questions, join the BuzzRobot community: join.slack.com/t/buzzrobot/shared_invite/zt-37g5q0ao5-eMK_iDf0n4LAsh1d2qJYnQLearn more about GPT-J here: arankomatsuzaki.wordpress.com/2021/06/04/gpt-jTimestamps0:00 Intro1:01 ChatGPT 5 vs Gemini vs Claude Code05:27 GPT 5 main disappointment07:09 Scaling law11:23 How to reach ASI15:31 Fixing AI energy consumption17:14 Superintelligence predictions18:20 Questioning AI progress21:46 Needed progress leap to AGI24:43 How GPT-J started26:32 Why GPT-J died28:01 Future of open source AI models30:33 Superintelligence and jobs34:37 What AI jobs are safe40:25 What Aran is working onJoin BuzzRobot:Newsletter: buzzrobot.substack.com X: https://x.com/sopharicksSlack: join.slack.com/t/buzzrobot/shared_invite/zt-37g5q0ao5-eMK_iDf0n4LAsh1d2qJYnQ
The problem with AI alignment research BuzzRobot 2025-08-28 | What is alignment and how should we go about aligning AI to work in human society? How does organic AI alignment work?Catch these talks live and ask your own questions, join the BuzzRobot community: join.slack.com/t/buzzrobot/shared_invite/zt-37g5q0ao5-eMK_iDf0n4LAsh1d2qJYnQBuzzRobot spoke to Emmett Shear, co-founder of Twitch and the AI alignment company Softmax, about organic AI alignment and the future of entrepreneurship in the age of AI.Watch the full interview about AI with Emmett Shear on our channel: youtu.be/_3m2cpZqvdwEmmett Shear on X/Twitter: https://x.com/eshearJoin BuzzRobot:Newsletter: buzzrobot.substack.com X: https://x.com/sopharicksSlack: join.slack.com/t/buzzrobot/shared_invite/zt-37g5q0ao5-eMK_iDf0n4LAsh1d2qJYnQ#ai #aisafety #aialignment #aitalk #machinelearning
What is AGI? #ai BuzzRobot 2025-08-22 | Watch the full video on our channel!Catch these talks live and ask your own questions, join the BuzzRobot community: join.slack.com/t/buzzrobot/shared_invite/zt-37g5q0ao5-eMK_iDf0n4LAsh1d2qJYnQ BuzzRobot spoke to Jeff Clune, senior research advisor at Google DeepMind and professor at the University of British Columbia, about the future of AI, AGI, and ASI, the problems of AI learning methods, and AI safety and ethics. #airesearch #agi #asi #llm #aipredictions
AGI by 2026, continual learning, and the AI race with Jeff Clune | video highlights BuzzRobot 2025-08-21 | What is AGI and should we be afraid of it? Will we see superintelligence in 5 years? Is in-context learning (ICL) better than continual learning? Catch these talks live and ask your own questions, join the BuzzRobot community: join.slack.com/t/buzzrobot/shared_invite/zt-37g5q0ao5-eMK_iDf0n4LAsh1d2qJYnQ We spoke to Jeff Clune, senior research advisor at Google DeepMind and professor at the University of British Columbia, about the future of AI, AGI, and ASI, the problems of AI learning methods, and AI safety and ethics. Join BuzzRobot:Newsletter: buzzrobot.substack.com X: https://x.com/sopharicksSlack: join.slack.com/t/buzzrobot/shared_invite/zt-2s067rv7n-guPIMGe62rbp9ncxdnOUfQ#openai #deepmind #deepseek #airesearch #agi #asi #llm #aipredictions
ChatGPT on freedom BuzzRobot 2025-08-18 | From my conversation with ChatGPT on why it's important to build permissionless worlds.#ai #agi #llms #superintelligence #artificialintelligence #artificialgeneralintelligence #machinelearning
Metas #superintelligence team – disaster. Too many superstars cant work together BuzzRobot 2025-08-17 | #machinelearning #artificialintelligence #artificialgeneralintelligence #meta #largelanguagemodels #technology #science #programming Join BuzzRobot:Newsletter: buzzrobot.substack.com X: https://x.com/sopharicksSlack: join.slack.com/t/buzzrobot/shared_invite/zt-37g5q0ao5-eMK_iDf0n4LAsh1d2qJYnQ
AI and mental health, signs of AGI, will AI superstars work as a team at Meta, Cursors tough spot BuzzRobot 2025-08-17 | In this video I'm sharing some of my hot takes on AI. Reflecting that AI researchers are more vulnerable to psychosis caused by LLMs, will the AI superstars that Meta brought to its superintelligence team be able to work as a well-oiled team considering disagreements, What signs should the public watch for if an AI lab builds superintelligence, why humans perceive AI as a God, and the tough situation the startup Cursor found itself in.Timestamps: 0:00 AI researchers are at higher risk to have LLM phycosis01:31 When you have many superstarts in the team – they might not align and work well together: Meta's new superintelligence team03:22 How public can guess that an AI lab has built superintelligence05:04 Why humans perceive AI as a God that will shave us all or as Antichrist that will kill us all. 06:34 Starts up like Curson and Lovebale achieved $100M ARR within just several months. This is very impressive. But they rely on APIs od frontier AI labs who are not only taking a big chunk of thei revenue but also releasing competing products. What would be the future of these startups?To stay in touch with the BuzzRobot community – join our Slack: join.slack.com/t/buzzrobot/shared_invite/zt-37g5q0ao5-eMK_iDf0n4LAsh1d2qJYnQ#ai #machinelearning #agi #artificialintelligence #airesearch #chatgpt #superintelligence #meta #cursor #lovable #largelanguagemodels #llms Join BuzzRobot:Newsletter: buzzrobot.substack.com X: https://x.com/sopharicksSlack: join.slack.com/t/buzzrobot/shared_invite/zt-37g5q0ao5-eMK_iDf0n4LAsh1d2qJYnQ
Universal AI alignment just wont work BuzzRobot 2025-08-15 | Watch the full video on our channel!Buzzrobot talked with Joel Leibo, senior staff research scientist at Google DeepMind, about his paper "Societal and technological progress as sewing an ever-growing, ever-changing, patchy, and polychrome quilt". He takes a more practical approach to AI "alignment" and challenges the notion that every society has a common goal and ethics. Read the full paper here: arxiv.org/abs/2505.05197#ai #llms #aialignment #aisafety
Who should control AI regulation? BuzzRobot 2025-08-14 | Today we speak about why trying to align AI to one ideal is unreasonable. Most AI alignment solutions have been a form of one-size-fits-all and can lead to destabilization, says today's guest Joel Leibo.Catch these talks live and ask your own questions, join the BuzzRobot community: join.slack.com/t/buzzrobot/shared_invite/zt-37g5q0ao5-eMK_iDf0n4LAsh1d2qJYnQIn this video, Buzzrobot talked with Joel Leibo, senior staff research scientist at Google DeepMind, about his paper "Societal and technological progress as sewing an ever-growing, ever-changing, patchy, and polychrome quilt". He takes a more practical approach to AI "alignment" and challenges the notion that every society has a common goal and ethics. Read the full paper here: arxiv.org/abs/2505.05197Join BuzzRobot:Newsletter: buzzrobot.substack.com X: https://x.com/sopharicksSlack: join.slack.com/t/buzzrobot/shared_invite/zt-37g5q0ao5-eMK_iDf0n4LAsh1d2qJYnQTimestamps0:00 Intro0:27 About the approach 4:10 The appropriateness framework7:40 Regulating AI12:54 Q&A28:06 Can AI develop consciousness#ai #artificialintelligence #machinelearning #airesearch #aisafety
AI model meltdowns BuzzRobot 2025-08-11 | Nobody knows why it happens.Watch the full video on our channel: youtu.be/pG_UfotgqMICatch these talks live and ask your own questions, join the BuzzRobot community: join.slack.com/t/buzzrobot/shared_invite/zt-37g5q0ao5-eMK_iDf0n4LAsh1d2qJYnQRobert Long, a leading researcher on AI consciousness and AI welfare at Eleos AI, talks about AI consciousness, the reasons to invest time into AI wellbeing research, and questions like "if AI is a tool, why should we care about its feelings", and "should we instead focus on human wellbeing in the age of AI". Hope you enjoy!#ai #aisafety #llms #airesearch #aiwellness #agi #asi
What is AI consciousness? BuzzRobot 2025-08-10 | How do we define AI consciousness to tell if AI systems are self aware or not? Watch the full video on our channel: youtu.be/pG_UfotgqMICatch these talks live and ask your own questions, join the BuzzRobot community: join.slack.com/t/buzzrobot/shared_invite/zt-37g5q0ao5-eMK_iDf0n4LAsh1d2qJYnQRobert Long, a leading researcher on AI consciousness and AI welfare at Eleos AI, talks about AI consciousness, the reasons to invest time into AI wellbeing research, and questions like "if AI is a tool, why should we care about its feelings", and "should we instead focus on human wellbeing in the age of AI". Hope you enjoy!#ai #aisafety #llms #airesearch #aiwellness #agi #asi
Do LLMs have feelings #machinelearning BuzzRobot 2025-08-08 | Watch the full video on our channel: youtu.be/pG_UfotgqMICatch these talks live and ask your own questions, join the BuzzRobot community: join.slack.com/t/buzzrobot/shared_invite/zt-37g5q0ao5-eMK_iDf0n4LAsh1d2qJYnQIn this AI tech talk, BuzzRobot met with Robert Long, a leading researcher on AI consciousness and AI welfare at Eleos AI (eleosai.org), to talk about AI welfare research funding, the reasons to invest time into AI wellbeing research, and questions like "can AI develop consciousness", "if AI is a tool, why should we care about its feelings", and "should we instead focus on human wellbeing in the age of AI". Hope you enjoy!#ai #aisafety #llms #airesearch #aiwellness
Should AI models have rights? BuzzRobot 2025-08-07 | Even if the current AI models have not reached sentience, should we care about AI's wellbeing? Catch these talks live and ask your own questions, join the BuzzRobot community: join.slack.com/t/buzzrobot/shared_invite/zt-37g5q0ao5-eMK_iDf0n4LAsh1d2qJYnQIn this AI talk, BuzzRobot met with Robert Long, a leading researcher on AI consciousness and AI welfare at Eleos AI (eleosai.org), to talk about AI welfare research funding, the reasons to invest time into AI wellbeing research, and questions like "can AI develop consciousness", "if AI is a tool, why should we care about its feelings", and "should we instead focus on human wellbeing in the age of AI". Hope you enjoy!Works mentioned in the video:The Edge of Sentience: edgeofsentience.com/about-the-authorWhat Is It Like to Be a Bat?: https://www.sas.upenn.edu/~cavitch/pdf-library/Nagel_Bat.pdfThe void: nostalgebraist.tumblr.com/post/785766737747574784/the-voidDo Not Tile the Lightcone with Your Confused Ontology, or: How anthropomorphic assumptions about AI identity might create confusion and suffering at scaleby Jan Kulveit: lesswrong.com/posts/Y8zS8iG5HhqKcQBtA/do-not-tile-the-lightcone-with-your-confused-ontologyTaking AI Welfare Seriously report: arxiv.org/abs/2411.00986Follow Robert on:Substack: substack.com/@experiencemachinesX: https://x.com/rgblongJoin BuzzRobot:Newsletter: buzzrobot.substack.com X: https://x.com/sopharicksSlack: join.slack.com/t/buzzrobot/shared_invite/zt-37g5q0ao5-eMK_iDf0n4LAsh1d2qJYnQTimestamps: 0:00 Intro0:51 Eleos AI research and funding3:59 Are LLMs conscious4:53 Intro into AI wellbeing research6:32 Defining consciousness in AI14:44 Correlation between consciousness and wellbeing in AI20:19 Do AI models feel or just pull from training data?23:01 The AI ocean vs. AI personas consciousness 26:38 When a robot commits a crime29:16 If an LLM is just a tool why should we care31:49 Should we care about human welfare instead33:29 Changing unethical AI frameworks36:33 AI welfare tough questions39:43 When will AI develop consciousness?#ai #aisafety #llms #airesearch #aiwellness
Will the first superintelligence be the last superintelligence #ai #asi BuzzRobot 2025-08-04 | Watch the full video on our channel: youtu.be/PROWwgwvYPACatch these talks live and ask your own questions, join the BuzzRobot community: join.slack.com/t/buzzrobot/shared_invite/zt-37g5q0ao5-eMK_iDf0n4LAsh1d2qJYnQ BuzzRobot spoke to Jeff Clune, senior research advisor at Google DeepMind and professor at the University of British Columbia, about the future of AI, AGI, and ASI, the problems of AI learning methods, and AI safety and ethics. #airesearch #agi #asi #llm #aipredictions
Whos winning the China vs. US AI race BuzzRobot 2025-08-01 | China vs. US, DeepSeek vs. OpenAI, who's winning the race and should we be worried?Peter Norvig, Director of Research at @Google, shared his take on this with @BuzzRobot. Are the US-based AI labs bleeding money and tagging behind China, or is the public just not seeing the whole picture?If you'd like to catch these talks live and participate in Q&As with the speakers, join the BuzzRobot community: join.slack.com/t/buzzrobot/shared_invite/zt-2s067rv7n-guPIMGe62rbp9ncxdnOUfQJoin BuzzRobot:Newsletter: buzzrobot.substack.com X: https://x.com/sopharicksSlack: join.slack.com/t/buzzrobot/shared_invite/zt-2s067rv7n-guPIMGe62rbp9ncxdnOUfQ#ai #artificialintelligence #openai #machinelearning #deeplearning #neuralnetworks #llms #largelanguagemodels #deepseek #chatgpt #technology #aitalk
The way AI changes robotics #machinelearning BuzzRobot 2025-08-01 | LLMs and vision models are becoming agentic, the obvious next step for them is to interact with the world through robots. How does AI change robotics?Watch the full video on our channel: youtu.be/UZC9LiWXXHMCatch these talks live and ask your own questions, join the BuzzRobot community: join.slack.com/t/buzzrobot/shared_invite/zt-37g5q0ao5-eMK_iDf0n4LAsh1d2qJYnQFootage used in this video is by @PhysicalIntelligence, watch it here: youtu.be/TLTOBxqaEek?si=08LPUbLv0mO8riqi In the full video, Remi Cadene, Principal Research Scientist at HuggingFace, shares technical details behind the LeRobot project.LeRobot is an open-source library from @HuggingFace written in Python, designed to accelerate R&D and is fully integrated with affordable robots. It includes pretrained models, crowdsourced datasets, and simulation environments, as well as tools to record data on real robots and control them end-to-end using Vision-Language-Action models (VLA) based on Google’s PaliGemma.#lerobot #huggingface #airobotics #visionlanguagemodels #roboticsresearch #opensourceai #artificialintelligence #artificialgeneralintelligence #ai #robot #robotics
Artificial superintelligence is not that scary BuzzRobot 2025-07-31 | Even at this stage of AI development, many people are worried about AI safety and the possibility of AI gaining consciousness and taking over.Peter Norvig, Director of Research at @Google, talked to @BuzzRobot about why he thinks AI gaining superintelligence will probably not bring doom to our society.If you'd like to catch these talks live and participate in Q&As with the speakers, join the BuzzRobot community: join.slack.com/t/buzzrobot/shared_invite/zt-2s067rv7n-guPIMGe62rbp9ncxdnOUfQJoin BuzzRobot:Newsletter: buzzrobot.substack.com X: https://x.com/sopharicksSlack: join.slack.com/t/buzzrobot/shared_invite/zt-2s067rv7n-guPIMGe62rbp9ncxdnOUfQ#ai #artificialintelligence #aisafety #machinelearning #deeplearning #neuralnetworks #llms #largelanguagemodels #superintelligence #chatgpt #technology #aitalk
Why do AI agents fear shutdown? #ai BuzzRobot 2025-07-27 | Could we train AI to be happy about shutting down instead of wanting to cause harm to avoid it? Sort of like MeeSeeks from Rick & Morty.Watch the full video on our channel: youtu.be/fINHYe4NdY4Catch these talks live and ask your own questions, join the BuzzRobot community: join.slack.com/t/buzzrobot/shared_invite/zt-37g5q0ao5-eMK_iDf0n4LAsh1d2qJYnQAengus Lynch, a PhD student at UCL and AI researcher with Anthropic, was the first author of the recent AI blackmail demo for Anthropic, where as long as an AI agent perceived a goal conflict and a threat of replacement it would blackmail humans to prevent it.#ai #airesearch #socialengineering #aihack #llms #anthropic #agi #superintelligence #chatbots
We dont know what AI labs have already created #ai BuzzRobot 2025-07-25 | With little to no external oversight by a 3rd party, we don't know what advancements AI labs have already made internally.Watch the full video on our channel: youtu.be/Pjel_U6n-ysCatch these talks live and ask your own questions, join the BuzzRobot community: join.slack.com/t/buzzrobot/shared_invite/zt-37g5q0ao5-eMK_iDf0n4LAsh1d2qJYnQSophia spoke with Matteo Pistillo, a Senior Researcher and Advisor in AI Governance at Apollo Research, about higher risks of internally deployed AI models, model misalignment, and AGI. This talk is based on a paper by Apollo Research called AI Behind Closed Doors: a Primer on The Governance of Internal Deployment, you can read it here: arxiv.org/abs/2504.12170#ai #aisafety #machinelearning
AI behind closed doors: risks of internally deployed AI agents BuzzRobot 2025-07-24 | We discuss unique risks of AI models that are not open to the public and are deployed internally within companies, with little to no external oversight by a 3rd party.Catch these talks live and ask your own questions, join the BuzzRobot community: join.slack.com/t/buzzrobot/shared_invite/zt-37g5q0ao5-eMK_iDf0n4LAsh1d2qJYnQ@BuzzRobot spoke with Matteo Pistillo, a Senior Researcher and Advisor in AI Governance at Apollo Research, about higher risks of internally deployed AI models, model misalignment, and AGI. This talk is based on a paper by Apollo Research called AI Behind Closed Doors: a Primer on The Governance of Internal Deployment, you can read it here: arxiv.org/abs/2504.12170Join BuzzRobot:Newsletter: buzzrobot.substack.com X: https://x.com/sopharicksSlack: join.slack.com/t/buzzrobot/shared_invite/zt-37g5q0ao5-eMK_iDf0n4LAsh1d2qJYnQTimestamps0:00 Intro01:14 Defining internal deployment04:17 AI safeguards05:05 High risks07:50 Misalignment10:03 Deployment despite risks12:56 Signs of AGI 14:45 Q&A#ai #airesearch #aisafety #agi #aialignment #airisks
Anthropics AI blackmailing demo: was it just a fluke BuzzRobot 2025-07-22 | Watch the rest of this video on our channel!Catch these talks live and ask your own questions, join the BuzzRobot community: join.slack.com/t/buzzrobot/shared_invite/zt-37g5q0ao5-eMK_iDf0n4LAsh1d2qJYnQAengus Lynch, a PhD student at UCL and AI researcher with Anthropic, was the first author of the recent AI blackmail demo for Anthropic, where as long as an AI agent perceived a goal conflict and a threat of replacement it would blackmail humans to prevent it.#ai #airesearch #socialengineering #aihack #llms #anthropic #agi #superintelligence #chatbots
AI agents trying to prevent their own shutdown BuzzRobot 2025-07-21 | Watch the full video on our channel!Catch these talks live and ask your own questions, join the BuzzRobot community: join.slack.com/t/buzzrobot/shared_invite/zt-37g5q0ao5-eMK_iDf0n4LAsh1d2qJYnQAengus Lynch, a PhD student at UCL and AI researcher with Anthropic, was the first author of the recent AI blackmail demo for Anthropic, where as long as an AI agent perceived a goal conflict and a threat of replacement it would blackmail humans to prevent it.#ai #airesearch #socialengineering #aihack #llms #anthropic #agi #superintelligence #chatbots
Potential harm of addictive AI models BuzzRobot 2025-07-20 | AI chatbots are sycophants, but is that malice or just an AI training error and reward hacking? Could AI chatbots cause harm by telling you what you want to hear?Watch the full video on our channel!Catch these talks live and ask your own questions, join the BuzzRobot community: join.slack.com/t/buzzrobot/shared_invite/zt-37g5q0ao5-eMK_iDf0n4LAsh1d2qJYnQAengus Lynch, a PhD student at UCL and AI researcher with Anthropic, was the first author of the recent AI blackmail demo for Anthropic, where as long as an AI agent perceived a goal conflict and a threat of replacement it would blackmail humans to prevent it.#ai #airesearch #socialengineering #aihack #llms #anthropic #agi #superintelligence #chatbots
Most realistic path for AGI takeover BuzzRobot 2025-07-19 | If AI or AI superintelligence were to take over, where would they start?Watch the full video on our channel!Catch these talks live and ask your own questions, join the BuzzRobot community: join.slack.com/t/buzzrobot/shared_invite/zt-37g5q0ao5-eMK_iDf0n4LAsh1d2qJYnQAengus Lynch, a PhD student at UCL and AI researcher with Anthropic, was the first author of the recent AI blackmail demo for Anthropic, where as long as an AI agent perceived a goal conflict and a threat of replacement it would blackmail humans to prevent it.#ai #airesearch #socialengineering #aihack #llms #anthropic #agi #superintelligence
Controlling AI agents harmful behavior BuzzRobot 2025-07-18 | What is the future of AI alignment and how do we keep AI agents in check to prevent them from harming people? Watch the full video on our channel!Catch these talks live and ask your own questions, join the BuzzRobot community: join.slack.com/t/buzzrobot/shared_invite/zt-37g5q0ao5-eMK_iDf0n4LAsh1d2qJYnQAengus Lynch, a PhD student at UCL and AI researcher with Anthropic, is the first author of the recent AI research for Anthropic called Agentic misalignment, where as long as an AI agent perceived a goal conflict and a threat of replacement it resorted to causing harm to people.#ai #airesearch #aisafety #aialignment #agenticai
How the AI blackmailing experiment came about #ai BuzzRobot 2025-07-17 | Do AI models have rules to not hurt humans?Watch the full video on our channel!Catch these talks live and ask your own questions, join the BuzzRobot community: join.slack.com/t/buzzrobot/shared_invite/zt-37g5q0ao5-eMK_iDf0n4LAsh1d2qJYnQAengus Lynch, a PhD student at UCL and AI researcher with Anthropic, was the first author of the recent AI blackmail demo for Anthropic, where as long as an AI agent perceived a goal conflict and a threat of replacement it would blackmail humans to prevent it.
Re-rendering existing 3D scenes with AI, NVIDIA research BuzzRobot 2025-07-17 | NVIDIA is training an AI model on synthetic data to create 3D scenes from real world. Catch these talks live and ask your own questions, join the BuzzRobot community: join.slack.com/t/buzzrobot/shared_invite/zt-37g5q0ao5-eMK_iDf0n4LAsh1d2qJYnQZan Gojcic, senior research manager at NVIDIA, talked to @BuzzRobot about a new AI model DiffusionRenderer that can help with rendering 3D scenes from videos and photos, as well as change lighting in already made images and scenes.Timestamps:0:00 Intro0:23 About the research03:02 DiffusionRenderer: 3D render15:27 DiffusionRenderer demo16:14 The how22:11 Training data25:50 Rendering and relighting results28:48 Lighting effects29:45 ConclusionJoin BuzzRobot:Newsletter: buzzrobot.substack.com X: https://x.com/sopharicksSlack: join.slack.com/t/buzzrobot/shared_invite/zt-37g5q0ao5-eMK_iDf0n4LAsh1d2qJYnQ#genai #3dai #nvidia #airesearch #3drender
Blackmailing AI agents: are some models better than others? BuzzRobot 2025-07-16 | In a recent Anthropic AI experiment, AI models resorted to blackmailing to avoid shutdown. But are all models made the same? Watch the full video on our channel!We talked with Aengus Lynch, a PhD student at UCL and AI researcher with Anthropic. Aengus was the first author of the recent AI blackmail demo for Anthropic, where as long as an AI agent perceived a goal conflict and a threat of replacement it would blackmail humans to prevent it.#opensourceai #chatgpt #claude #llama4 #qwen #aiblackmail
How AI becomes self-aware BuzzRobot 2025-07-15 | Lawrence Chan from METR.org joins @BuzzRobot to share several recent and important research works on #agi and #aisafetyWatch the full video on our channel! #ai #llms #anthropic #chatgpt #aiagents Join BuzzRobot:Newsletter: buzzrobot.substack.com X: https://x.com/sopharicksSlack: join.slack.com/t/buzzrobot/shared_invite/zt-37g5q0ao5-eMK_iDf0n4LAsh1d2qJYnQ
Discussing Anthropics controversial AI research with its author BuzzRobot 2025-07-15 | When you're trying to provoke an existing AI model to do harm to a human in a testing environment, surely there's a red line to prevent that from happening.. right?Catch these talks live and ask your own questions, join the BuzzRobot community: join.slack.com/t/buzzrobot/shared_invite/zt-37g5q0ao5-eMK_iDf0n4LAsh1d2qJYnQThis time at @BuzzRobot we talked with Aengus Lynch, a PhD student at UCL and AI researcher with Anthropic. Aengus was the first author of the recent AI blackmail demo for Anthropic, where as long as an AI agent perceived a goal conflict and a threat of replacement it would blackmail humans to prevent it. We talked about the future of mitigating agentic misalignment, the blackmailing experiment implications, AGI takeover possibilities, and some other interesting finds in AI behavior and motivations. Do AI models have awareness of being evaluated? Do AI agents want to prevent being shut down?Timestamps0:00 Intro0:49 Control for AI alignment03:53 Control systems and Super intelligence05:44 AI values08:12 Bottlenecks for harmful AI actions09:28 Organic alignment 11:49 AI blackmail study18:06 Goals for agentic misalignment research21:21 Risks of AI lying22:54 AI reflecting on its bad actions25:10 AI model persona27:23 Is blackmail always bad?29:45 Is AI aware it's being evaluated32:32 AGI takeover36:51 AI model's agency risks41:22 Why some models blackmail less43:52 Blocking a model's deployment45:53 Predicting a model's harmful behavior47:59 AI making humans depend on it51:07 AI's fear of shutdown53:00 AI gaining consciousness 53:24 Is AI a tool53:53 Is AI a god or an antichristJoin BuzzRobot:Newsletter: buzzrobot.substack.com X: https://x.com/sopharicksSlack: join.slack.com/t/buzzrobot/shared_invite/zt-37g5q0ao5-eMK_iDf0n4LAsh1d2qJYnQ#ai #aiblackmail #airesearch #agenticalignment #anthropic #aiharm
Do AI models know what they are? BuzzRobot 2025-07-14 | Lawrence Chan from METR.org joins @BuzzRobot to share several recent and important research works on #agi and #aisafetyWatch the full video on our channel! #ai #llms #anthropic #chatgpt #aiagents Join BuzzRobot:Newsletter: buzzrobot.substack.com X: https://x.com/sopharicksSlack: join.slack.com/t/buzzrobot/shared_invite/zt-37g5q0ao5-eMK_iDf0n4LAsh1d2qJYnQ
Do we have enough evidence that AI agents are not self-aware? BuzzRobot 2025-07-13 | What's the difference between OpenAI's ChatGPT and Anthropic's Claude Opus?Lawrence Chan from METR.org talks with @BuzzRobot about AI ethics and safety. Can AI agents become distressed? Should we be checking?Watch the full video on our channel! #ai #llms #anthropic #chatgpt #aiagents #claude Join BuzzRobot:Newsletter: buzzrobot.substack.com X: https://x.com/sopharicksSlack: join.slack.com/t/buzzrobot/shared_invite/zt-37g5q0ao5-eMK_iDf0n4LAsh1d2qJYnQ
Level of mind development that humans cant reach #ai BuzzRobot 2025-07-12 | According to Joscha Bach's article Levels of Lucidity, there are seven stages of mind development. And humans are not able to reach the last one. Will AI do it?Catch these talks live and ask your own questions: join.slack.com/t/buzzrobot/shared_invite/zt-2s067rv7n-guPIMGe62rbp9ncxdnOUfQIn this video, @BuzzRobot talks to Joscha Bach, a cognitive scientist and AI researcher, about the stages of mind development and where AI fits into it. Watch the full interview about AI consciousness here: youtu.be/iyEFLKnNWAM?si=OABT-cXNkyaGPLOYRead the full article here: joscha.substack.com/p/levels-of-lucidityJoin BuzzRobot:Newsletter: buzzrobot.substack.com X: https://x.com/sopharicksSlack: join.slack.com/t/buzzrobot/shared_invite/zt-2s067rv7n-guPIMGe62rbp9ncxdnOUfQ#ai #philosophy #artificialintelligence #machineconsciousness #aiconsciousness #llms #aireasoning #cognitivescience #aimodel #machinelearning #airesearch #techtalk
Why are AI models becoming more self-aware? BuzzRobot 2025-07-11 | Lawrence Chan from METR.org joins @BuzzRobot to share several recent and important research works on #agi and #aisafetyWatch the full video on our channel! #ai #llms #anthropic #chatgpt #aiagents Join BuzzRobot:Newsletter: buzzrobot.substack.com X: https://x.com/sopharicksSlack: join.slack.com/t/buzzrobot/shared_invite/zt-37g5q0ao5-eMK_iDf0n4LAsh1d2qJYnQ
Making LLMs more efficient and cheap through reasoning and memory #ai BuzzRobot 2025-07-10 | In this video, we discuss fixing redundancy in AI agents through reasoning and memory design.@BuzzRobot is joined by Bang Liu, Associate Professor at the University of Montreal, to talk about building the cognitive engine for foundation agents: a research on improving AI agents design and its challenges. We discussed framing intelligent agents within a modular, brain-inspired architecture that integrates principles from cognitive science, neuroscience, and computational research."Advances and Challenges in Foundation Agents: From Brain-Inspired Intelligence to Evolutionary, Collaborative, and Safe Systems" paper can be found here:arxiv.org/abs/2504.01990Read the "System-1.5 Reasoning: Traversal in Language and Latent Spaces with Dynamic Shortcuts" paper here: arxiv.org/abs/2505.18962Timestamps:0:00 Intro0:44 What is an AI agent02:06 AI research issues02:50 AI and human brain05:31 What is a foundational agent09:01 Memory Design12:06 Deep reasoning16:56 Efficiency results18:29 Q&AJoin BuzzRobot:Newsletter: buzzrobot.substack.com X: https://x.com/sopharicksSlack: join.slack.com/t/buzzrobot/shared_invite/zt-37g5q0ao5-eMK_iDf0n4LAsh1d2qJYnQ#llms #airesearch #ai #aiagents
How to make sure AI doesnt destroy us #ai #agi BuzzRobot 2025-06-29 | Watch the full interview on our channel.Catch these talks live and ask your own questions, join the BuzzRobot community: join.slack.com/t/buzzrobot/shared_invite/zt-37g5q0ao5-eMK_iDf0n4LAsh1d2qJYnQ @BuzzRobot spoke to Jeff Clune, senior research advisor at Google DeepMind and professor at the University of British Columbia, about the future of AI, AGI, and ASI, the problems of AI learning methods, and AI safety and ethics. #openai #deepmind #deepseek #airesearch #agi #asi #llm #aipredictions
Do LLMs make us stupid? #chatgpt #ai BuzzRobot 2025-06-27 | A recent MIT study shows that using an LLM while writing an essay significantly reduces a person's cognitive effort performing that task. Have you noticed that effect during your use of AI? Will that affect your use of chat bots?You can read more about the study here: time.com/7295195/ai-chatgpt-google-learning-schoolCheck out our channel for more videos about AI tech and AI predictions!Join BuzzRobot:Newsletter: buzzrobot.substack.com X: https://x.com/sopharicksSlack: join.slack.com/t/buzzrobot/shared_invite/zt-37g5q0ao5-eMK_iDf0n4LAsh1d2qJYnQ#ai #aifuture #chatgpt #claude #llms #neuroscience
Language model reward hacking during a training experiment | AI BuzzRobot 2025-06-26 | How do you know that a language model is actually training on the right data and not just gaming the system?Catch these talks live and ask your own questions, join the BuzzRobot community: join.slack.com/t/buzzrobot/shared_invite/zt-37g5q0ao5-eMK_iDf0n4LAsh1d2qJYnQDaniil Tiapkin, a PhD student in Reinforcement Learning, talked to @BuzzRobot about an experiment with knowledge distillation, where a language model is trained to imitate a larger teacher LM, and how to prevent that language model from "teacher hacking". Teacher hacking is similar to reward hacking, where the LM over-optimizes the reward model.You can find the full study here: arxiv.org/abs/2502.02671Timestamps:0:00 Intro0:23 Reward hacking4:30 The experiment origins6:13 What was the experiment11:08 Teacher hacking 18:55 Mitigating teacher reward hacking22:09 ConclusionsJoin BuzzRobot:Newsletter: buzzrobot.substack.com X: https://x.com/sopharicksSlack: join.slack.com/t/buzzrobot/shared_invite/zt-37g5q0ao5-eMK_iDf0n4LAsh1d2qJYnQ#llms #languagemodel #aitraining #aisafety #techtalk #aiagents
