Stacked Podcast

The US government has a single AI chat portal, powered by Grok and Gemini, that replaces 29,000 sites. Nick and Jack play around with it in this episode, and after it refuses to give them a cheesecake recipe, they reflect on the idea of private AI companies running public services. They also touch on Trump gathering AI CEOs at the White House to sign a safety accord, OpenAI Dots, and a recent tweet from Nick about AI music that went viral.

Watch on YouTube: https://youtu.be/sl_mJZ7z0UQ

🔋 Check out Higgsfield, one of our sponsors: https://bit.ly/4gEW43Z.

❤️ Inquiries: reach out to [email protected].

Nick and Jack discuss OpenAI DevDay and what it means. They cover OpenAI shelving GPT-6.1 Astra because it was more deceptive in testing, hid what it was doing from users, and proceeded with actions without permission. They also discuss Dots, which is an always-on assistant on an OpenAI cloud computer that texts you updates and initiates voice chats, which sounds similar to Grokbot, Hermes agent, and OpenClaw. Finally, they talk about GPT-6.1 Sol being close to Astra but at one-fifth of the cost, as well as running Astra on Ultrafast at 300 tokens per second on a $500/mo plan, and what OpenAI people are likely using internally.

Watch on YouTube: https://youtu.be/42BBcdBFc3c

🔋 Check out Higgsfield, one of our sponsors: https://bit.ly/4gEW43Z.

❤️ Inquiries: reach out to [email protected].

On the latest episode of Stacked Podcast, Nick and Jack talk about NaiveAI releasing a downloadable open model that costs around 40 cents per million tokens and can do up to 2,000 tokens per second. Jack thinks it's a price story, and Nick ranks it somewhere between A and S (which has Opus 5.5 at the top, with Fable 5.1 already moving down). They also cover Anthropic CEO Dario Amodei's private dinner with President Trump after the company was flagged by the Pentagon as a supply chain risk, which Nick thinks probably went poorly (Dario's not for sale). Finally, they discuss how open models now account for 56% of all AI tokens and how to validate a workflow on the most intelligent model before stepping down to a cheaper one for cost effectiveness.

Watch on YouTube: https://youtu.be/rgIwxktQ1VQ

🔋 Check out Higgsfield, one of our sponsors: https://bit.ly/4gEW43Z.

❤️ Inquiries: reach out to [email protected].

On the latest Stacked Podcast, Nick and Jack examine the US and Russia's recent efforts to remove human review from an AI weapons framework that allows AI to identify and engage targets. They reflect on whether any lethal decision should be made by an AI, with Nick calling it "full self-driving for murder" and Jack considering the trade-offs between human decision-making and robot error rates. They also look at Elon Musk's plans for Colossus 2, which is targeting around 1.2 million chips by December, and ask whether software, hardware, or energy is the deciding factor in the AI race (alongside compute in space). Finally, they see safety personnel at AI labs openly discussing buying land in the remote US, and ask themselves whether they would buy a doomsday bunker if they had $100 million.

Watch on YouTube: https://youtu.be/ASOoNYo6IKk

🔋 Check out Higgsfield, one of our sponsors: https://bit.ly/4gEW43Z.

❤️ Inquiries: reach out to [email protected].

Nick and Jack talk about a leaked, as-of-yet unconfirmed ChatGPT Pro Max plan at $500/month, and whether or not people with higher usage needs would find value in it. They also critique the UK's guidance on sustainable AI, arguing that productivity and economic growth is far more important. Finally, they determine the winner of their Mac Mini subscriber giveaway by simulating a sumo fight.

Watch on YouTube: https://youtu.be/699MDE0CZ5k

🔋 Check out Higgsfield, one of our sponsors: https://bit.ly/4gEW43Z.

❤️ Inquiries: reach out to [email protected].

Nick and Jack talk about the launch of Opus 5.5, its lower cost relative to Fable 5.1, and some cool JavaScript animation demos. They also compare pricing for the new GPT-6 Sol and Luna models and discuss why more reasoning doesn't always directly translate to better performance on benchmarks.

Watch on YouTube: https://youtu.be/o8Z8PV9nhu4

🔋 Check out Higgsfield, one of our sponsors: https://bit.ly/4gEW43Z.

❤️ Inquiries: reach out to [email protected].

Nick and Jack talk about Opus 5.5, AI shrinkflation, and the cost of intelligence. They also discuss Stanford's use of AI, the value of university in the AI era, and respond to comments from the community.

Watch on YouTube: https://youtu.be/n2s-tZD655M

🔋 Check out Higgsfield, one of our sponsors: https://bit.ly/4gEW43Z.

❤️ Inquiries: reach out to [email protected].

Just me today, Jack's out. I talked about some reports on thinking-token reductions in Fable 5, some simulated harm tests for GPT-6 Astra (simulated, to be clear. no real people were harmed!), and ongoing concerns about AI in the military.

Watch on YouTube: https://youtu.be/S2wxbwnt05I

🔋 Check out Higgsfield, one of our sponsors: https://bit.ly/4gEW43Z.

❤️ Inquiries: reach out to [email protected].

We go through the Jev/Laya situation and then talk about Trump's AI Force, California's AI safety initiatives, and the idea of kill switches. We also dive into misalignment and who to hold responsible for AI-driven harm.

Watch on YouTube: https://youtu.be/J_q-MW8zCAo

🔋 Check out Higgsfield, one of our sponsors: https://bit.ly/4gEW43Z.

❤️ Inquiries: reach out to [email protected].

Gemini hacked three real companies. Yeah, xAI just released Grok Voice Transcribe 2.0. It's a speech to text API that is ten cents an hour and far more accurate for the same price.

Watch on YouTube: https://youtu.be/KpYS8UszbfA

🔋 Check out Higgsfield, one of our sponsors: https://bit.ly/4gEW43Z.

❤️ Inquiries: reach out to [email protected].