532: Perplexity Buyout Talks, Meta Search Engine, Waymo vs Tesla, OpenAI Drops Foundry Dream, Anti-Natalist Policies, Samsung, and Podcasting Era

"lip service to family values"

Speak confidently as if you are right, but listen carefully as if you are wrong.

—Kevin Kelly

The vibe I was getting. Made with Midjourney 6.1 💚 🥃

📚📝🧃I recently had a valuable opportunity to share an important life lesson with my oldest boy—one that I didn't learn until much later in life than his current age of ten.

He was doing homework and kept saying how boring it was, how much it sucked, and generally grunting and grumbling.

Don’t get me wrong, I could sympathize — *I* did close to zero homework until university, and even then it was pretty hit-and-miss… But that’s not what I want to tell you about today.

I told my son that sometimes you have to do boring things and how bad they feel to you will largely depend on your self-talk and mindset.

When you approach homework while thinking ‘this sucks, this is terrible’, it’s going to *feel worse* than if you had more neutral self-talk like 'okay, let's get this done so we can move on to other things.'.

Ideally, I told him, we should strive for positive self-talk to make challenging tasks less painful. We can’t lie to ourselves and pretend it’s fun, but we can usually find positive and constructive ways to frame things; for example, something like “Two problems down, just three to go! I know how to do this, let’s see how efficient I can be at it. I'll challenge myself to complete this section without taking a break—I bet I can finish it in under ten minutes if I really focus.'".

Negative self-talk. Also made with MJ 6.1 💚 🥃

I explained to him how our brains are always listening to our internal dialogue, making it crucial to be mindful of our self-talk. To illustrate, I told him about my childhood friend who had the habit of calling himself ‘stupid’, ‘an idiot’, and ‘dumb’ whenever he made a mistake. It couldn’t have been good for his self-esteem, and over time he may have started believing himself.

This concept aligns with the principles of cognitive behavioral therapy, which suggests changing thought patterns to improve how we feel. It's not just that we have negative thoughts because we feel bad—we often feel bad because of our negative thoughts (causality goes both ways). Change the thoughts, change the subjective experience.

If instead of calling yourself a dumbass every time you make a mistake, consider more self-empathy: ‘Oh boy, I made a mistake here, but it happens to everyone. I’ll learn from this and be more careful next time.’ You probably end up both doing better and feeling better.

And it’s not just for you — self-talk creates ripples that affect our entire social ecosystem. Others also notice negative thought patterns, and it affects how they perceive us (which can become self-fulfilling), and it can change how they perceive themselves because we all serve as models for those around us—our children, family members, friends, and colleagues.

I don’t think my son internalized all this yet, but I’m planting seeds, and hopefully over time it helps him.

---

🏦 💰 Liberty Capital 💳 💴

🎧 How big can podcasting get? 🚀

I didn’t expect more growth before the pandemic than after. We’re clearly in the podcasting era of media.

The rising popularity of podcasts has created a new gateway to long-form content. The next logical step is for more of these podcast converts to also discover audiobooks. Too few people read books — there are such wonders to be found in them — and this low-friction way to read deserves to be more popular.

Ben Thompson (💚 🥃 🎩) has a good post about the impact of podcasts in this election cycle:

The reason I mention this is because of the effect this may end up having on presidential selection in the long run: a big reason the aforementioned televised debate helped Kennedy is because he was much more telegenic than Nixon. The currency of podcasts, on the other hand, is humor and story-telling and the ability to range widely across topics

This selection effect is interesting to think about.

It’s kind of bonkers that being a good podcast guest may become a necessary skill to become president of the US and commander in chief of the most powerful military in the world. But I suppose that politics has always had strange selection processes and this may not be worse than ‘Who would you rather have a beer with?’…

This raises questions about our leadership selection process: If you were investing your family's entire wealth in a single company, would you choose its CEO based on their podcast charisma? If your loved ones' lives depended on military leadership, would you select a 5-⭐️ general based on their podcast skills? There’s a disconnect between our political selection criteria and the actual responsibilities of leadership.

¯\_(ツ)_/¯

📄🔍🤖 Perplexity was an acquisition target for: X/Twitter, OpenAI, Notion, and Microsoft (at least) 🐦📎💰

I’m not the only one who likes Perplexity:

In the summer and fall of 2023, Srinivas told members of his executive team that Perplexity had received acquisition offers from three companies: X, OpenAI and Notion, according to a person with direct knowledge of the matter.

Srinivas indicated that Perplexity’s talks with X were particularly detailed as Elon Musk, the social network’s owner, was interested in using Perplexity’s technology to improve X’s search functionality, the person said. Perplexity had earlier launched Bird SQL, a tool for searching X’s archive of tweets using conversational language.

Srinivas also told employees Microsoft had expressed interest in acquiring Perplexity and had discussed how to structure a deal that would make it pass regulators, said the person.

All three offers were in the $150 million to $200 million range, the person added. Perplexity was valued at $150 million after a funding round announced in March 2023.

What a steal it would’ve been for $150-200m!

The latest round, which I wrote about in Edition #531, is in the $8bn range.

I’m curious if there have been more acquisition talks since then. I’m guessing that they’re probably always ongoing in the background for such a hot startup.

Will Srinivas have his Zuckerberg-with-Yahoo moment where he says no to a big offer to keep building? Back then it was $1bn for Zuck — the equivalent today is probably at least $10-20bn, I’d guess.

Speaking of Zuck:

🏗️👷‍♂️⚙️ ‘Meta Develops AI Search Engine to Lessen Reliance on Google, Microsoft’

As Meta Platforms tries to keep up with OpenAI in developing artificial intelligence, [it] is working on a search engine that crawls the web to provide conversational answers about current events to people using its Meta AI chatbot.

In doing so, Meta hopes to lower its reliance on Google Search and Microsoft’s Bing, which currently provide information about news, sports and stocks to people using Meta AI

This makes sense, and I wonder if this could also make Perplexity a target for Meta.

One of Zuck’s core skills is identifying startups that will make it big early and making them what seems like an overly generous offer at the time, but later turns out to be cheap:

• $1bn for Instagram in 2012

• $4 billion in cash + $12 billion in stock + $3 billion in restricted stock for WhatsApp in 2014.

Would Zuck offer Perplexity multiples of its current valuation to see if they bite?

The biggest question is: Would he be allowed to do it by regulators?

I doubt it. Meta has had problems buying a GIF company, so they probably can’t buy Perplexity. This probably means they’ll keep building their own…

🩻🔍⏳ Acquired’s 6-hour podcast about Meta & Zuckerberg! ⌛️

They did it, they made a longer Meta/Zuck podcast than the one I did with MBI:

• 🎧 Meta Platforms & Mark Zuckerberg: Going Deep with Mostly Borrowed Ideas (MBI)! 🕵️‍♂️🩻

At 1h35mins we weren’t that hard to beat…

I mean, I don’t want to make the running time the most notable thing about it, I’m just kidding. It’s a fun hook, but the *real* reason to listen to it is David & Ben’s deep research and genuine curiosity.

• 🎧 The Complete History & Strategy of Meta

And if that’s not enough for you, there’s also the live interview that they did with Mark Zuckerberg a few weeks ago:

• 📺 🎧 The Mark Zuckerberg Interview

👀🚙🤖 Waymo raises $5.6 billion, bringing total capital raised to $13 billion (Capital Efficiency vs Tesla)

This is all napkin math and based on estimates, so don’t expect this to be precise or even correct, but when I saw that Waymo is estimated to have raised $13bn so far, it made me look up how much capital Tesla has raised before it became more self-funding.

From IPO in 2010 to 2017, I added up $18.79 billion in equity and debt (I may have missed some).

Considering that Waymo was founded inside of Google in 2009 — later becoming a subsidiary of Alphabet — and has fewer than 1,000 vehicles deployed (afaict), this highlights how capital-efficient Tesla has been.

I know they’re not taking the same approach, but the breadth and scope of Tesla’s activities, including vertically integrated manufacturing around the world making millions of cars of multiple models, the autonomous driving software, AI projects, robots, the energy division, robotaxis, etc, is very impressive considering the inputs!

🤖💭 OpenAI gives up crazy foundry dreams and partners with Broadcom and TSMC for custom AI chip 🐜🏗️

A pragmatic pivot:

OpenAI has examined a range of options to diversify chip supply and reduce costs. OpenAI considered building everything in-house and raising capital for an expensive plan to build a network of factories known as "foundries" for chip manufacturing.

The company has dropped the ambitious foundry plans for now due to the costs and time needed to build a network, and plans instead to focus on in-house chip design efforts, according to sources, who requested anonymity as they were not authorized to discuss private matters.

The idea of building a bunch of global foundries from a blank page always seemed a little far-fetched, but I respect the ambition and large-scale thinking. Once in a while, one of these “crazy” ideas actually works and changes the world…

OpenAI has been working for months with Broadcom to build its first AI chip focusing on inference, according to sources. Demand right now is greater for training chips, but analysts have predicted the need for inference chips could surpass them as more AI applications are deployed.

The company has assembled a chip team of about 20 people, led by top engineers who have previously built Tensor Processing Units (TPUs) at Google, including Thomas Norrie and Richard Ho.

Sources said that through Broadcom, OpenAI has secured manufacturing capacity with Taiwan Semiconductor Manufacturing Company [TSMC] to make its first custom-designed chip in 2026. They said the timeline could change.

Focusing on inference makes a lot of sense — however, the fact that they’re apparently doing it without Microsoft (who’s also working on custom chips) shows how strained that relationship has been getting.

Nadella used to talk about designing chips specifically for OpenAI’s models not so long ago, and now he talks about models being commoditized and is funding large internal efforts to compete with OpenAI, even Github Copilot is going multi-model with support for Anthropic and Google models… while OpenAI barely mentions the company that provides them most of their funding and compute 🤔

At first, they were married, now they’re just roommates…

---

🧪🔬 Liberty Labs 🧬 🔭

The gray states are those where no parental leave laws have been enacted.

🇺🇸✋🤰🏻 How Anti-Natalist Policies Discourage Parenthood in the U.S.👩🏻‍🍼🚸

I recently talked to a friend who’s expecting their first child soon and it reminded me of how anti-natalist parental leave policies are in the US. They’re a low-hanging fruit to tweak if we’re looking to boost fertility.

Here’s the situation:

Federal Policy

The Family and Medical Leave Act (FMLA) is the primary federal law governing parental leave in the US, providing:

• 12 weeks of unpaid, job-protected leave

• Only applies to employers with 50+ employees within 75 miles

• Employees must have worked:

  • At least 12 months for their employer

  • Minimum 1,250 hours in the previous year

Due to these restrictions, approximately 40% of workers are not eligible for FMLA protections

In a nation that often gives lip service to family values:

“The United States is the only country among the 38 member OECD nations that has not passed laws requiring businesses and corporations to offer paid maternity leave to their employees”

Most states haven’t adopted policies, and others have just slightly increased the eligibility of the federal benefits — nothing too drastic.

There are also weird cases like this: “Hawaii, Puerto Rico, and the District of Columbia designate childbirth as a temporary disability thus guaranteeing mothers paid maternity leave through Disability Insurance (TDI) provisions” It really doesn’t send the right message and shows how dysfunctional the legislative process is.

“As of 2023, only 27% of private sector workers had access to any paid family leave, according to the Bureau of Labor Statistics.”

According to a study by the World Economic Forum: “Although 12 weeks are allowed to them, on average American fathers only take 10 days off, due to financial need”

How does this compare to other countries? Let’s have a look:

• France: 16 weeks (6 pre-birth, 10 post-birth), fully compensated

• UK: Up to 52 weeks, with 39 weeks paid (6 weeks at 90% of earnings, 33 weeks at £184/week)

• Germany: Up to 3 years parental leave for each parent (at 60-67% of average earnings capped at €1,800 monthly)

• Canada: 15 weeks of maternity leave plus additional parental leave (but in the province where I am, Quebec, you can get 50 weeks).

I don’t know what the perfect approach is. There’s probably such a thing as too much, but the US is pretty far from that!

🇰🇷 What’s wrong Samsung? (again) 🐜🏚️

Interesting quote about what may be going wrong inside of Samsung:

“The biggest factor affecting the current concerns surrounding Samsung is the company’s corporate culture of pursuing excessive meritocracy among its executives,” said Professor Lee Jong-hwan of Sangmyung University’s Department of System Semiconductor Engineering.

“The issues with HBM and foundry businesses stem from executives rushing to achieve quick results. When developing technology, it's crucial to consider mass production, meaning the manufacturing yield, but the meritocracy is driving executives to focus on immediate development achievements without considering mass production.”

Samsung has been keen to [be first to] introduce new semiconductor technology [to] the world. In February, Samsung announced that it developed the world’s first 12-layer HBM3e, but SK hynix said it began mass production of the 12-layer chip in September.

“One of the most crucial factors in semiconductors is mass production, which hinges on yield. Yield improves when all departments of a chipmaker work toward a shared goal in a coordinated manner,” Lee said. “However, in recent years, Samsung has been prioritizing the introduction of the world’s first technology without adequately considering mass production, as each executive focuses on gaining personal credit.” (Source)

I disagree with calling this “excessive meritocracy” — instead, I would say that they have created strong incentives around the wrong metrics and KPIs. This is causing execs to optimize for those at the detriment of the bigger picture and more important goals.

If they had pursued different goals with the same zeal — ie. rather than trying to be first or being the executive who crossed the finish line, if they had focused on mass production and commercial success — they may be in a very different place.

These things could’ve been accomplished in a meritocratic way, that wasn’t the problem.

🌋🌋🌎 Every Earthquake since 1980 🌏🌋🌋

Seismic Explorer allows you to explore earthquake date since 1980 in a visual way:

Geologists collect earthquake data every day. What are the patterns of earthquake magnitude, depth, location, and frequency? What are the patterns of earthquakes along plate boundaries?

Click the play button to see the earthquakes. You can drag the starting time to start playing earthquakes from a later date.

Use the Magnitude slider to choose the earthquake size shown on the map. Click the Show plate boundaries button to see the outlines of tectonic plates.

Make a cross-section to see a three-dimensional view of the earthquakes in a region. Click on the Draw a cross-section line then draw a line on the map. When you are done, click Open 3D view, and see the depths of the earthquakes in that cross-section. What does the pattern of earthquakes in a region tell you about the motion of tectonic plates?

The 3d view of the cross-section mode kind of blew my mind. You can not only see the location of earthquakes in X-Y coordinates, but you can also see their depth inside the Earth’s crust!

Here’s an example of a very active area in the Pacific Rim:

Very cool. h/t Kottke

---

🎨 🎭 Liberty Studio 👩‍🎨 🎥

🤖🎻🎼 Robot Plays Cello with Symphony Orchestra in Sweden

There are a lot of world firsts these days:

In a mesmerising piece written by Swedish-born contemporary classical composer and producer Jacob Mühlrad, the lines between acoustic and electronic music are skilfully blurred in this performance which took place in the Malmö Live Concert Hall.

In this clip of Jacob’s new piece, ‘Veer (bot)’, the orchestra creates a serene and hypnotic soundworld, before the cello joins and complements this with a sustained solo and laconic passage [...]

For Jacob, whose music draws on global and multi-facetted influences, this project is not about showing how technology could theoretically replace musicians, but rather a study of the cello’s possibilities when played by a tool with a very different anatomy from a human, and of how technology can complement human musicianship when man and machine performs side by side.

h/t MBI (🇧🇩🇺🇸)