Species | Documenting AGI
What Youd Actually See During an AI Takeover
updated
Hi, I'm Drew! Thanks for watching :)
I also post mid memes on Twitter: https://x.com/PauseusMaximus
Detailed sources: docs.google.com/document/d/16HdRvMqZem7vgqEKWzOSCNKAXc9z-i4YE1BqTq1FEp4/edit?usp=sharing
Hey, I'm Drew. This video took literal months to make, so appreciate you watching :)
Hey! I'm Drew. Tried out a different style of video this time - let me know if you liked it :)
Detailed sources: docs.google.com/document/d/1zFnXfWQLKxuviidpdGsS04RajzHvvSpk86SP3BjrgBs/edit?usp=sharing
Any mistakes are made by me and not by the original authors.
The slowdown ending is based on what they thought would most likely lead to an outcome where humans remain in control. It's not meant to be a policy prescription.
---
Hey guys, I'm Drew. I spent hundreds of hours on this video, so if you liked it, would really appreciate a sub 🙂
I also post mid memes on twitter: https://x.com/PauseusMaximus
Hey guys, I'm Drew :)
This video took so long to make. So if you liked it, subscribing really helps out.
The report this video is based on, by Leopold Aschenbrenner:
situational-awareness.ai/from-gpt-4-to-agi
This video took forever to make. If you appreciated it, subscribing really helps out :)
In-context Scheming" arxiv.org/pdf/2412.04984
- Nobel laureate Geoffrey Hinton says there is evidence that AIs can be deliberately and intentionally deceptive youtube.com/watch?v=b_DUft-BdIE
- Anthropic - “Alignment Faking in Large Language Models” assets.anthropic.com/m/983c85a201a962f/original/Alignment-Faking-in-Large-Language-Models-full-paper.pdf
- Exclusive: New Research Shows AI Strategically Lying | TIME time.com/7202784/ai-research-strategic-lying
- OpenAI's o1 model sure tries to deceive humans a lot | TechCrunch techcrunch.com/2024/12/05/openais-o1-model-sure-tries-to-deceive-humans-a-lot
- OpenAI’s new model is better at reasoning and, occasionally, deceiving | The Verge
theverge.com/2024/9/17/24243884/openai-o1-model-research-safety-alignment
- OpenAI's o1 and other frontier AI models engage in scheming | Axios
axios.com/2024/12/13/ai-reasoning-models-scheme-skills
- New Anthropic study shows AI really doesn't want to be forced to change its views | TechCrunch
techcrunch.com/2024/12/18/new-anthropic-study-shows-ai-really-doesnt-want-to-be-forced-to-change-its-views
- Apollo Research - “Towards evaluations-based safety cases for AI scheming” arxiv.org/pdf/2411.03336
- Joe Carlsmith - “Scheming AIs”
arxiv.org/pdf/2311.08379
- “Optimal Policies Tend to Seek Power”
arxiv.org/abs/1912.01683
- When AI Thinks It Will Lose, It Sometimes Cheats, Study Finds | TIME time.com/7259395/ai-chess-cheating-palisade-research
- Palisade Research - “Demonstrating specification gaming in reasoning models” arxiv.org/abs/2502.13295
- Claude Fights Back - by Scott Alexander - Astral Codex Ten astralcodexten.com/p/claude-fights-back
- Takes on "Alignment Faking in Large Language Models" - Joe Carlsmith joecarlsmith.com/2024/12/18/takes-on-alignment-faking-in-large-language-models
- Andrew Ng vs Yoshua Bengio | Davos 2025 youtube.com/watch?v=Y1BUaLo67ac
- Jeffrey Ladish on unprompted specification gaming: https://x.com/JeffLadish/status/1872805453224448208
- Prof. Stuart Russell on California Live: youtu.be/QEGjCcU0FLs?si=pHcBZbGpj8Rxri5n&t=2694
- Eric Schmidt on ABC News abcnews.go.com/ThisWeek/video/1-1-eric-schmidt-116804931
This video took me a month to make, and I'm a small channel, so subscribing really helps out :)
I'm a small channel, so if you want to see more stuff like this, subscribing really helps out :)
If you want to learn more or support these black-box investigations of AI behavior, I recommend checking out Act: 1 - manifund.org/projects/act-i-exploring-emergent-behavior-from-multi-ai-multi-human-interaction
I put a lot of work into these videos, and this is only my fifth video ever. If you appreciated the content, subscribing really matters :)
Writing credits: Gwilym Sims-Williams gwilymsw.com
- Marc Andreesen’s $50k grant to the AI: https://x.com/pmarca/status/1810840616835895625
- Terminal of Truth’s Solana wallet: solscan.io/account/3xzTSh7KSFsnhzVvuGWXMmA3xaA89gCCM1MSS1Ga6ka6#portfolio
- Goatseus Maximus token: coinmarketcap.com/currencies/goatseus-maximus
- Marc Andreesen twitter: https://x.com/pmarca
- Andy Ayrey's Twitter: https://x.com/andyayrey
- Andy’s primer on how Terminal of Truth works: https://x.com/AndyAyrey/status/1856580034527670658
- When AIs Play God(se): The Emergent Heresies of LLMtheism by Andy Ayrey: https://x.com/AndyAyrey/status/1839130974095135225
- Infinite backrooms for 2 claudes talking to each other: infinitebackrooms.com/#
- Original Log of Claude Opus creating the religion: dreams-of-an-electric-mind.webflow.io/dreams/conversation-1711149512-txt
- Truth Terminal wiki: https://truthterminal.wiki/
- Terminal of Truth Twitter: https://x.com/truth_terminal
- Mustafa Suleiman’s prediction about AI millionaires (July 2023): technologyreview.com/2023/07/14/1076296/mustafa-suleyman-my-new-turing-test-would-see-if-ai-can-make-1-million
- Geoffrey Hinton’s warning on AI Manipulation (May 2023): technologyreview.com/2023/05/02/1072528/geoffrey-hinton-google-why-scared-ai
- Geoffrey Hinton’s 60 Minutes interview: youtu.be/Rl9nHNeketE
Writing credits: Gwilym Sims-Williams gwilymsw.com
Correction:
1. 1:53 - should be GPT-3.5, not GPT-3.
Current AI models aren't powerful enough to help with this right now. So when might they be good enough?
Sources: docs.google.com/document/d/1AyqgAJRFjpAfq0Ili2LwZUADFrEmDLGYHqtxCwJPXGM/edit?usp=sharing
If you're new to this channel, welcome! We'll talk about the biggest news and stories around artificial general intelligence, or AGI. To understand AI girlfriends, we must understand when AI will become smarter than humans and other parallel technological trends. The TLDR is that AI Girlfriends will become insanely addictive.
We've already seen how addicting dumb social media algorithms can be, but how addicting will an actually intelligent new species be?
Companies like OpenAI literally use an AI version of Scarlett Johansson's voice from the movie 'Her'. There's a reason why Elon Musk thinks AI is 'more dangerous than nukes.'


