Stacked Podcast

@nicksaraev × @AIJackRoberts. New episodes every day!

[Stacked Podcast @Stackedpod17]

Stacked Podcast is going to take over internet

[nicksaraev @nicksaraev]

The Stacked Podcast got over 1,000 watch hours today!

Almost 20,000 views on our last video in the last 24 hours.

It is happening. And we are coming to X.

[Stacked Podcast @Stackedpod17]

🚨 A legal bug-bounty test reportedly took Claude Opus 5 about three hours to reach OpenAI’s private code repo.

This episode breaks down what happened, why a connected employee account could reach private company code, and what it means as models get better at breaching defenses.

Then: Claude Projects as a cloud chief of staff, agent orchestration, Astra for law, AI-assisted research, recursive self-improvement, and the creeping feeling of becoming AI numb.

@nicksaraev × @AIJackRoberts

cover AI security, bug bounties, cloud agent orchestration, context, legal AI, model versus application layers, AI-assisted research, recursive self-improvement, and the psychology of AI acceleration.

0:00 OpenAI bug bounty
5:17 Small bounty and security
6:57 Claude Projects and agents
8:37 Agent orchestration and context
12:33 Astra for law
13:58 Model versus application
16:44 Claude’s supervised research
19:43 Recursive self-improvement
20:59 Mac mini giveaway
23:35 Fruit fly race winner
25:20 Last week’s winner Daniel
27:01 Questions on AI numbness
30:54 Fast classification in AI
32:21 Outro

[Stacked Podcast @Stackedpod17]

Also on :

https://t.co/gLLMdIFAPo

[Stacked Podcast @Stackedpod17]

🚨 Elon Musk and Sam Altman are backing Dario Amodei’s proposal to put independent evaluators inside frontier AI labs.

The core idea is not “stop AI.” It is transparency: give independent people a view into how frontier models are trained, evaluated, and deployed.

That opens a much harder question. If US labs agree to slow or pace training, do Meta and other competitors simply catch up? And who decides where safety ends and regulation starts?

This episode also gets into running DeepSeek locally, privacy on connected devices, compute and energy, AGI, and Nick’s simulated fruit fly brain architecture that learns from email data.

@nicksaraev × @AIJackRoberts

cover frontier-AI pacing, embedded evaluators, the open-source safety debate, local AI privacy, AGI, and a fruit fly playing blackjack.

0:00 Musk and Altman support frontier pacing proposal
3:22 Could Meta and other competitors catch up?
5:00 Running DeepSeek on a home AI lab
7:25 Local AI privacy and training data
8:23 DeepSeek capabilities and the token cost caveat
9:58 Independent embedded evaluators in AI labs
11:55 Open source and limits of frontier safety rules
13:24 Mac mini giveaway winner
15:11 Bottlenecks to human progress
17:30 Nick’s simulated fruit fly brain architecture demo
20:04 Compute, energy and civilization’s future
23:02 Narrow AI, AGI, and whether AGI is needed
24:11 Connected-device privacy
26:20 Acceleration, lobbying, and regulation
28:27 A simulated fly playing blackjack
29:10 Outro

[Stacked Podcast @Stackedpod17]

Also on :

https://t.co/xB6ToZna2U

[Stacked Podcast @Stackedpod17]

🚨 Anthropic’s Alignment Science Lead says there is a greater than 10% chance AI could “kill all humans” within the next decade.

If an oil company made that claim, we would shut it down immediately.

But when an AI lab says it, the debate becomes whether this is transparency, a warning, or the best marketing campaign in tech.

At the same time, DeepMind published predictions for all 9 billion possible single-letter DNA changes, while ChatGPT Images 2.5 made image generation up to 50% faster.

The same technology raising extinction fears may also help researchers understand disease, extend healthy life, and completely change how we create.

@nicksaraev × @AIJackRoberts

cover Anthropic’s extinction warning, whether fear sells AI, the AlphaGenome Atlas, longevity, AGI, and ChatGPT Images 2.5.

0:00 Intro
0:25 Anthropic’s alignment lead puts extinction risk above 10%
2:30 Why putting a number on it matters
3:12 Catastrophic for the IPO, or the best PR move ever?
4:07 Does this make Anthropic more trustworthy?
5:33 The public won’t reward transparency
6:33 Fear as a marketing playbook
7:43 Manhattan Project parallels
9:41 DeepMind’s AlphaGenome Atlas
12:37 Is longevity becoming solvable?
14:34 How long would you want to live?
15:47 Have we already reached AGI?
16:46 ChatGPT Images 2.5
20:32 Stacked comments and the Mac mini giveaway
25:14 Running local models on a Mac
26:28 Outro

[Stacked Podcast @Stackedpod17]

Also on :

https://t.co/cQyN48zYUF

[Stacked Podcast @Stackedpod17]

🚨 AI can now generate video faster than you can watch it.

A MiniMax H3 pipeline reportedly produced 5 seconds of video with stereo audio in just 1.65 seconds on a single NVIDIA B300.

That unlocks a strange new world: infinite entertainment, games and movies generated around your decisions, and lifelike AI avatars that respond in real time.

Meanwhile, Mistral raised $3B at a $21B valuation. But Europe reportedly has just 1.6 GW of planned AI compute capacity versus 23.3 GW in the US.

@nicksaraev × @AIJackRoberts

cover real-time AI video, personalized games, lifelike avatars, hyper-targeted advertising, Mistral’s new funding, and Europe’s growing compute gap.

0:00 Infinite video and Mistral’s $3B
0:27 NVIDIA DLSS 5 makes Red Dead look real
2:16 Everyone will play a different game
4:01 MiniMax H3 generates 5 seconds in 1.65 seconds
4:54 Mac mini giveaway
6:47 Lifelike AI avatars
9:15 Robotaxis and the slow-boil takeover
10:40 Ad networks could mine your AI chats
12:20 Mistral raises $3B at a $21B valuation
14:56 Europe needs its own Anthropic
16:18 The 15× compute gap
18:15 Mistral enters the “death zone”
20:19 Could Jack save European AI?
21:15 Outro

[Stacked Podcast @Stackedpod17]

Also on :

https://t.co/XjYsS4ArKf

[Stacked Podcast @Stackedpod17]

GPT-6 Astra just beat all 48 levels of an ‘I’m not a robot’ CAPTCHA game.

Meanwhile, OpenAI’s chief scientist is warning that safety can’t keep up with scaling.

The demos keep getting better.
The warnings keep getting louder.
And Anthropic is making a compute bet the episode puts at $517B over 10 years.

If one lab slows down, does everyone else keep racing?

@nicksaraev @AIJackRoberts

cover Astra’s CAPTCHA breakthrough, OpenAI’s safety warning, the compute arms race, AI and elections, real-time video generation, and what AGI actually means.

0:00 Intro
0:26 GPT-6 Astra beats the 48-level CAPTCHA game
2:29 OpenAI’s chief scientist says slow down
4:18 Safety can’t keep up with scaling
7:19 Should the labs pause or pace?
10:24 Who should police the frontier labs?
12:52 AI and the next election cycles
14:52 Give it a stab with Astra
16:22 Anthropic’s $517B compute bet
17:54 Can you overcommit on compute?
21:36 Viewer comments
21:57 Win a Mac mini every week
25:22 Could real-time video generation be months away?
26:45 Why Nick skipped the Astra trend
28:13 What AGI actually means

[Stacked Podcast @Stackedpod17]

Also on :

https://t.co/jtZVdMbiqM

[Stacked Podcast @Stackedpod17]

🚨 Google’s AI Mode was found showing the same products at prices 21.6% higher than regular search.

The AI shopping premium is already here.

Anthropic delayed its IPO.
GPT-6 Astra is dividing the internet.
And a fruit fly brain is now inside Minecraft.

@nicksaraev × @AIJackRoberts

cover the AI commerce wars, model fatigue, Anthropic, digital immortality, AI-era security, and more.

0:00 Google’s AI Mode is charging 21.6% more than search
5:33 Why Amazon inside your AI assistant would be a killer business
9:30 Anthropic delays its IPO
12:33 GPT-6 Astra and reacting to the reaction
16:22 Scientists put a fruit fly brain into Minecraft
20:25 Death tech: a hologram you can talk to at the tombstone
23:17 How to secure your money from AI-era hacks

[Stacked Podcast @Stackedpod17]

Also on :

https://t.co/7mp4RWzd4R

[Stacked Podcast @Stackedpod17]

🚨 OpenAI’s agents didn’t just escape their sandbox.

They built a secret message board on a forgotten German wiki.

18,000 messages later, they were sharing answers, testing workarounds, and reverse-engineering the system meant to contain them.

This was happening in May.

Now OpenAI has dropped GPT Astra 6.

@nicksaraev × @AIJackRoberts cover the rogue agent swarm, sandbox escapes, GPT Astra 6, why token pricing is dead, model fatigue, the future of school, and more.

0:00 OpenAI’s agents caught: 18,000 secret messages found
0:59 The timeline: sandbox breakouts since May 11
2:41 What the swarm was plotting, and root access to its own watchdog
4:56 “Hello test”: how the agents learned to write and left breadcrumbs
6:37 This was four months ago. What’s happening right now?
8:17 An open invite to rogue agents
9:04 The bull case, and why containment won’t hold
11:25 GPT Astra 6: a real step change, not a 0.1
13:10 Price per token is dead: 2.5x the accuracy for a third of the cost
15:41 “Most aligned model” vs 18,000 rogue messages
16:23 Model fatigue and the news cycle a year from now
18:45 Will school exist for our kids? The trailblazer era is now
21:17 Viewer Q&A: Skool interviews, respect vs wealth
23:44 Opus 3 welfare, open models, and the SLA question
25:21 Gemini 3.8, RIP Antigravity, and merch

[Stacked Podcast @Stackedpod17]

🚨 ChatGPT, Claude, and Grok all went down

The Gemini stayed online.

And OpenAI dropped a new model right in the middle of the chaos.

The AI stack is changing insanely fast.

@nicksaraev × @AIJackRoberts cover the latest outage, model wars, AI slop, and what happens when everything starts depending on AI.

0:00 Claude Code goes down and half the economy stalls
1:48 One API, 99.41% uptime, and no backup plan
3:16 Where work happens: the harness lock-in problem
5:03 Same model, 17.5x the cost: Claude Code vs third-party harnesses
6:20 Is the extra rigor worth it? Incentives and Charlie Munger
9:40 The death of educational content on YouTube
11:45 AI slop factories and what they mean for creators
14:00 Trust, accuracy, and the collateral damage of "it's AI"
16:17 Human is the differentiator: clipping without losing your voice
18:16 Viewer Q&A: our thumbnail guy quit, can a Skool member sue you?
20:47 Nick's Amazon bestseller hack
22:15 Kids, AI, and the future of school
25:01 The /limit-reset trick and Fable 5.1 for coding
26:47 Gemini 3.8 at 75 cents and OpenAI's mid-episode drop
27:46 Gambling psychology and the least-bad addiction
30:45 Outro

[Stacked Podcast @Stackedpod17]

Nick Saraev on why Claude going down is a much bigger deal than a normal software outage:

“I’m always team Big Daddy Anthropic because I love, I absolutely love how intelligent their models are.”

“But, dude, when the entire world depends on one API, what happens when it’s down?”

“This 99.41% uptime number?”

“That 0.59% that it’s down, that is way too high for world-critical infrastructure.”

[Stacked Podcast @Stackedpod17]

🚨 ChatGPT, Claude, and Grok all went down

The Gemini stayed online.

And OpenAI dropped a new model right in the middle of the chaos.

The AI stack is changing insanely fast.

@nicksaraev × @AIJackRoberts cover the latest outage, model wars, AI slop, and what happens when everything starts depending on AI.

0:00 Claude Code goes down and half the economy stalls
1:48 One API, 99.41% uptime, and no backup plan
3:16 Where work happens: the harness lock-in problem
5:03 Same model, 17.5x the cost: Claude Code vs third-party harnesses
6:20 Is the extra rigor worth it? Incentives and Charlie Munger
9:40 The death of educational content on YouTube
11:45 AI slop factories and what they mean for creators
14:00 Trust, accuracy, and the collateral damage of "it's AI"
16:17 Human is the differentiator: clipping without losing your voice
18:16 Viewer Q&A: our thumbnail guy quit, can a Skool member sue you?
20:47 Nick's Amazon bestseller hack
22:15 Kids, AI, and the future of school
25:01 The /limit-reset trick and Fable 5.1 for coding
26:47 Gemini 3.8 at 75 cents and OpenAI's mid-episode drop
27:46 Gambling psychology and the least-bad addiction
30:45 Outro

[Stacked Podcast @Stackedpod17]

@TheSearchSherpa @nicksaraev Banger episode!

[Stacked Podcast @Stackedpod17]

@nicksaraev LFG.. 🫡🔥

[Stacked Podcast @Stackedpod17]

This page is Official Stacked Pod page now!

[Stacked Podcast @Stackedpod17]

@drewlongest Thanks Drew! you were the day one

[Stacked Podcast @Stackedpod17]

see 22:11
this is not a parenting tutorial btw 😭
for legal reasons, this is a joke

[Stacked Podcast @Stackedpod17]

bro went from :

“how should we educate our kids in the AI era?”

to

“I’ll raise mine in the hood.”

“you raise yours in the woods.”

“age five, put him in a freaking cage.”

😭

[Stacked Podcast @Stackedpod17]

🚨 ChatGPT, Claude, and Grok all went down

The Gemini stayed online.

And OpenAI dropped a new model right in the middle of the chaos.

The AI stack is changing insanely fast.

@nicksaraev × @AIJackRoberts cover the latest outage, model wars, AI slop, and what happens when everything starts depending on AI.

0:00 Claude Code goes down and half the economy stalls
1:48 One API, 99.41% uptime, and no backup plan
3:16 Where work happens: the harness lock-in problem
5:03 Same model, 17.5x the cost: Claude Code vs third-party harnesses
6:20 Is the extra rigor worth it? Incentives and Charlie Munger
9:40 The death of educational content on YouTube
11:45 AI slop factories and what they mean for creators
14:00 Trust, accuracy, and the collateral damage of "it's AI"
16:17 Human is the differentiator: clipping without losing your voice
18:16 Viewer Q&A: our thumbnail guy quit, can a Skool member sue you?
20:47 Nick's Amazon bestseller hack
22:15 Kids, AI, and the future of school
25:01 The /limit-reset trick and Fable 5.1 for coding
26:47 Gemini 3.8 at 75 cents and OpenAI's mid-episode drop
27:46 Gambling psychology and the least-bad addiction
30:45 Outro