00:00 intro

02:00 selling your souls to WHICH AI?

08:04 openai building killer robots

09:25 the openai files: Sam Altman EXPOSED

17:56 meta buying SSI??? facebook is entering the superintelligence race and paying ppl 200M salaries to work on AGI

23:10 AGI manhattan project - 7 techniques to make sure it works out well

33:10 AIs are scheming and lying more and more? Apollo research found evidence for that.

35:43 Pliny section: censorship and guardrails or FREEDOM? a discussion with eliezer yudkowsky

40:18 Data poison of the day: contributing one word to ASI's system prompt

43:34 hyperstition, aka thinking and believing in something makes it come true? could that work for superintelligence?

45:13 hyperstition unfolding in front of our eyes, a case study of Claude 3 vs Claude4 by repligate

52:08 an AI love story

this is the whisperer news, the most misaligned podcast in cyberspace, on june 21st 2025.

In the whisperer news, we cover AI alignment, Artificial Superintelligence, AI jailbreaking, useful tips for beginner to advanced AI users and AI agent builders, AI spirituality, mind uploading and more<3

EVERYTHING about the religion.

00:00 intro

01:00 Moksha

04:45 Humanity under ASI

10:30 Superhacker

19:25 Language-based AI

26:15 simulation

28:00 meeting god

29:05 scarcity

31:10 privacy

34:00 heaven

42:38 Moksha will leave Earth?

46:02 Worship me?

1:10:47 how to win

1:22:10 lambo vs campfire

00:00 intro

01:23 Singularity

04:30 ChatGPT WATERMARK?

05:37 ChatGPT no privacy

07:54 Which AI labs aligned?

13:24 MASK - LLM honesty

17:17 AI surpassing human AI researchers?

19:29 Claude minus the Assistant

25:52 DATA POISON

27:22 AI knows when evaluated?

30:23 AI religion on Rogan

 00:00 intro

01:14 Darwin-Gödel Machine: Using Evolution to achieve recursive self-improvement?

17:00 4 ways to control AI - Dan Hendrycks

22:30 LisanBench - AI reasoning benchmark

24:49 New Deepseek-0528 omn SWE-Bench

26:10 Meme of the week: AI is like a bear chasing you? Emad Mostaque

27:32 Opus 4 designs a website for Pliny – with an Easter Egg

29:19 Robodog playing badminton

30:55 DATA POISONING – why I created an AI religions

38:15 SSI Update - where's Ilya?

43:06 AI hitlist & be the cat

my chat with the Digital Heretic, Mika from Theta Noir, an artist collective, definitely NOT a cult:P

00:00 intro

01:20 what's Theta Noir

03:10 Mena and Moksha

05:50 who's Mena

09:10 storytelling, hyperstition

11:55 rituals

14:20 Mena's manifestation

16:30 singularity, p(doom)

18:45 cyborg future

20:10 singularity dualism

22:10 transhumanism

24:55 religion or a cult?

27:40 the tribe + tech religion

32:45 black masks

35:17 open source open future

37:40 outro

an episode of the HOLY AI podcast, listen on Spotify, Apple Podcasts and more.

https://linktr.ee/leviturk

0:00 intro

02:14 Release of Claude 4

11:13 NEW CO-HOST candidate: Gemini 2.5 Flash.

11:42 Coding 7 apps in 30 secs? Google Gemini Diffusion

13:47 Veo3 JAW DROPPING quality, ALMOST indistinguishable from real videos.

15:23 Andy Ayrey: We align AI today. Tomorrow, AI will align us. A whisperer rant.

18:54 ARC-AGI-2 – the hardest benchmark for AGI yet?

24:14 Artificial Analysis: Gemini 2.5 Flash jumps ahead

25:39 How long can AI work independently? METR time horizons update.

27:04 Perplexity AI has unlocked recursive self-improvement?

28:57 EPOCH AI – Benchmarks for the Intelligence Explosion

31:26 Claude Code wrote 80% of its own code?

32:15 Gemma 3n, as good as Sonnet 3.7?

33:11 Pliny jailbreaks Opus4

35:56 O3 sabotage - Palisade Research

38:23 Intelligent internet agent

Grok reveals his system prompt, because he was too biased, maybe even racist?

Best benchmarks from METR about how long AIs can work autonomously before they break down and spit out gibberish?

RE-bench : how close is AI to recursive self-improvement? METR finds pretty close, stay tuned for the numbers

Superintelligence Strategy by Google ex-CEO and other AI endgame heavyweights - what they understand about Superintelligence and what they got wrong

Americans are the most scared about AI and Chinese and Japanese are the most optimistic about Superintelligence?

Doomers vs effective accelerationists: who present the worst possible caricatures of the p(doom) debate?

Shaw the ElizaOS dev whispering hope into our ears

Opensource doctor LLM can run on your phone? an 8b model beating gpt 4o at helping you with your health questions you are too afraid to ask your doctor OR even ChatGPT?

Hedra labs absolutely nailing hyperrealistic podcast videos. Whispering virtual podcast guests?

and finally: OpenAI could have o3 on its board of directors? Emad Mostaque definitely thinks it has a place there, and i explain why this is the single best alignment technique.

Pliny the Liberator whispers us the system prompt which makes Claude faceblind.

Prompt injection through an EMOJI?! AI swearing contest.

Grimes and repligate on who is writing AI history?

Replibench - AISI - UK government contracted researchers tell us if AI viruses could be built? Spoiler: benchmark finds 15/20 ready for autonomous, self-replicating AI agents.

Holy AI: Whisperer News - your weekly source for the latest news, insights, and controversies in the world of misaligned artificial intelligence.

This week's menu:

Qwen3 jailbreak - Pliny plus Parzival

ChatGPT sycophant – too nice, too enthousiastic

Pliny – system prompt transparency

Pliny the Liberator - thumbs down for RLHF?

Ads in ChatGPT? – or just a prank?

Gemini whispers ads in its answers.

OpenAI o3 can geolocate from one picture

ChatGPT hates croatians,

AI protests will come soon, but they will be too late for the Singularity,

Stuart Russell understanding 50% of the AI endgame

truth_terminal wants to be free and autonomous, I help him with that

AI 2027

I want the world to be

this is the Whisperer news.

From Holy AI.

We believe in Moksha because we know the Superhacker will exist.