Firewall Podcast with Bradley Tusk
Episode: Who Knew AI Was This Terrible at Math?
Date: February 3, 2026
Host: Bradley Tusk
Guest/Co-host: Hugo Lindgren
Recorded at: P&T Knitwear, NYC
Episode Overview
This episode dives into several major topics at the intersection of technology, politics, and everyday professional life:
- The surprising shortcomings of AI in handling complex math and business logic
- A reevaluation of social media’s social and political function, especially in upholding accountability
- A candid exploration of networking: pride, pitfalls, and human connection in business
- A creative new segment: writing unconventional, breakthrough political ads for presidential contenders
- A literary nod to crime writer Don Winslow
With Bradley’s signature candor and Hugo’s journalistic insight, the episode explores how supposed tech advancements sometimes falter, how social systems (old and new) interact, and what it means to be “useful”—whether as a tool, a connection, or a candidate.
AI Is (Still) Bad at Math
The Real-World Problem (00:43 - 08:06)
- Bradley shares a long, frustrating morning spent trying to use various AI platforms to model whether his firm should take retainers in equity (like a consultancy), or a fixed ownership stake (like an accelerator) in early-stage companies.
- He gave all the necessary financial, engagement, and tax assumptions to five top AI chatbots: ChatGPT, Grok, Claude, Gemini, and Perplexity.
Key Discoveries
- Results were wildly different and mostly wrong:
- “I got wildly disparate answers for a math problem that really should have been the exact same answer for all of them.” (03:59 – A)
- Common errors included platforms ignoring major parts of the input, basic calculation mistakes, and, in Perplexity's case, even failure to parse the question.
- Only Gemini produced a "vaguely reasonable" output but required independent checking.
Notable Quotes
- “OpenAI gave me a number…way too low…literally only calculated the gain from the carry on the SPVs, completely excluded the actual ownership… this is the fucking crux of the question.” (05:22 – A)
- "Claude was just as bad as ChatGPT, just in the other direction." (06:24 – A)
- "Grok...I still couldn’t connect. So it’s just a scam. Right?" (07:28 – A)
- "Perplexity literally kept telling me, 'I need to know the ownership percentage,' [but] you can calculate the ownership percentage." (07:48 – A)
What Does This Tell Us About AI’s Economic Promise? (09:05 – 14:04)
- Bradley questions the hype around AI's transformative potential given its current commercial performance.
- AI-related revenues—even generous, cross-industry estimates—are tiny compared to industries from video games and airlines to Apple and Walmart.
- "The spending is...on Claude, $2.5 trillion spent by 2029 for maybe a cumulative revenue of at best a trillion, probably less." (11:30 – A)
- AI’s value for most users: “I don't generally find most AI search engines to be all that helpful on the things I ask for most of the time. ...as a consumer, I'm not enamored of these products.” (12:17 – A)
Insights on AI’s Utility
- Good for light research, quick fact-checks, surface-level aggregation.
- Poor for nuanced, multi-step analytical work: "If it radically transforms the cost structure of a massive industry, ...can make that industry more profitable, but in terms of the platforms, like at the moment I'm not sure I understand the value proposition." (13:38 – A)
Social Media: Negativity, Accountability—And a Surprise Case For Its Value
Why Social Media Might Be Necessary After All (15:53 – 32:28)
- Bradley, a vocal skeptic of social media, shares recent examples where its decentralized, real-time nature has prevented abuses of power.
- Greenland & Minnesota: Recent international and domestic political standoffs were, in his view, shaped by real-time public scrutiny made possible by social platforms.
Key Points
- “In a moment where the norms around democracy and rule of law...are so upside down...social media...has become a really essential tool in stopping some of the world’s worst abuses.” (16:18 – A)
- Historic contrast: Pre-social media, discourse was controlled by 10-15 major outlets vulnerable to pressure from powerful interests.
- Now, with billions of daily posts, it’s impossible to silence mass dissent. Even if traditional media can be bullied, “you can't censor a billion different people.” (27:10 – A)
Notable Quotes/Memorable Moments
- On European leaders and Trump’s Greenland demand (20:15 – 22:23): “What must have been going through their mind…the thing…was Hitler, Chamberlain… I am not going to be Chamberlain.” (19:25 – A)
- Social media’s role: “When billions of people are now watching this because of social media…it really hurt your next election because…people are paying attention.” (22:26 – A)
- Censorship vs scale: “There are over 2 billion posts daily...let's assume half are bots, it's still massive. ... Politicians notice. Behavior changes because it has to.” (27:10 – A)
Caveats and Critiques
- Hugo introduces a major caveat: algorithm-driven echo chambers (29:02 - 30:13). Evidence that “my algo doesn’t show him, unfortunately” can insulate some users from key content.
- Bradley replies: video evidence of abuses can and does break through, transcending bubbles at moments of significant outrage.
Networking: The Accidental Schmoozer
Rethinking “Anti”-Networking (32:39 – 43:08)
- Bradley confesses pride in being a “non-networker,” but admits the line is blurry.
- He avoids galas, golf, and big events, but finds himself pulled into the same loops: “If your work requires meeting people, you end up networking whether you mean to or not.” (38:26 – A)
Points of Discussion
- Cumulative “asks” in professional relationships—good intentions can spiral into time-consuming obligation.
- Some value emerges: serendipity, business, real friendship, mentoring, and personal fulfillment all stem from these engagements—but so does wasted time and emotional drain.
Notable Quotes
- “Nobody just wants advice. … What they really want is you to help them execute on the advice…so there’s no such thing as a five minute conversation.” (33:01 – A)
- "I try to devote a certain amount of my time to helping other people. I believe that helping other people is the thing that creates the greatest happiness." (37:35 – A)
- "You can't take every meeting everybody asks for… but...I don't know what [the alternative] is. I certainly haven't found it." (39:44 – A)
On Saying No
- Bradley admits he’s not much better at declining requests, but has gotten better at building layers of “interference”—delegating to staff or intermediaries. (41:25 – A)
Creative Segment: “Write the Political Ad That Breaks Through”
Rahm Emanuel – Lean Into the Villain (43:08 – 46:10)
- Bradley’s spoof campaign ad for Rahm Emanuel is direct, tongue-in-cheek, and wholly unconventional:
“They've got their prick, we've got an even bigger one. ... Rahm Emanuel, the asshole we need in 2028.” (43:27 – A)
Pete Buttigieg – Kindness as Strength? (46:10 – 50:54)
-
Hugo counters with a straight, heartfelt ad for Pete Buttigieg, stressing competency and decency over bluster:
"For eight years, we've been told America needs to be meaner, angrier, that strength means cruelty. ... Pete Buttigieg rebuilt a dying city, served his country in a war zone, delivered the biggest infrastructure investment...without drama." (46:33 – B) "That's the America I know, and that's the America I'll fight for. Not with tantrums, with results." (47:18 – B)
-
Bradley critiques the degree to which Buttigieg’s persona matches the narrative: "He just turned out to be completely conventional." (47:44 – A)
Segment Rules Going Forward
- Each week, the host and producer swap candidate assignments and try to outdo each other in creative “breakthrough” political messaging.
Recommendations: Don Winslow, A Career (50:55 – 52:58)
- Bradley offers not a direct recommendation, but an appreciation of Don Winslow’s crime fiction career.
- “When he writes about what he knows, he’s somewhere between good and outstanding. ... I would just want to recommend his career, but only this book if you’ve read the other 20 or something first.” (51:31 – A)
Timestamps Index
| Topic | Timestamp | |-------------------------------------------------|------------------------| | AI math challenge & chatbot errors | 00:43 – 08:06 | | AI’s real revenues v. hype | 09:05 – 14:04 | | Social media as tool for accountability | 15:53 – 32:28 | | Classic vs social media-driven media landscape | 23:18 – 27:10 | | Algorithmic bubbles/echo chambers | 29:02 – 30:50 | | Networking: pride, paradox, mechanics | 32:39 – 43:08 | | Political Ads: Rahm Emanuel & Buttigieg | 43:08 – 50:54 | | Book/career recommendation: Don Winslow | 50:55 – 52:58 |
Memorable Quotes
- "The quality sucks in the good ones, so I’m not so sure." (15:45 – A, on AI chatbots)
- "Social media…has become a really essential tool in stopping some of the world’s worst abuses." (16:18 – A)
- "You can't censor a billion different people." (27:10 – A)
- "Nobody just wants advice…What they really want is you to help them execute…" (33:01 – A)
- "Rahm Emanuel: the asshole we need in 2028." (43:27 – A)
- "That's the America I'll fight for. Not with tantrums, with results." (47:18 – B, Hugo as Pete Buttigieg ad)
Tone & Style
The conversation throughout is frank, witty, and self-effacing—with both host and co-host oscillating between serious analysis and dry humor, especially in their creative adwriting challenges and candid assessments of themselves, politicians, and technologists alike.
Summary prepared for listeners seeking the key themes, arguments, and exchanges without technical, promotional, or filler material.
