🔮 Google’s $100bn problem; escaping OpenAI; global trade, the physics of heartbreak++ #409
The challenge to Google is very real...
Hi, I’m Azeem Azhar. I convene Exponential View to help us understand how our societies and political economy are changing under the force of rapidly accelerating technologies.
In today’s edition:
How Google lost $100bn to an LLM problem;
Why China banned certain solar panel components’ exports;
Two giants disagree about the future shape of trade.
Sunday chart: Google’s $100bn LLM problem
Google stock lost about $130bn in value this week. To put this into context, that is more than SAP, Europe’s largest software business, is worth.
The culprit? Google’s chatbot.
It had a hallucinatory response. When noticed, Google’s stock fell.
If large language models become a key part of users’ search expectations, Google will face additional computational costs. Analyst Dylan Patel estimated that today it costs Google about 1.06 cents to run a search query, on which it generates 1.61 cents of revenue. Dylan’s napkin math suggests that ChatGPT queries cost roughly 0.36 cents. So a naive integration of ChatGPT into Google “would be devastating”. To the tune of a $36bn reduction in operating income.
Of course, Google has more scale than OpenAI does, as probably the largest private cloud in the world. Perhaps they would run a less complex model than ChatGPT or find other optimisations. But even if it is 10% of the cost Patel suggests, that is another $4bn of operating income lost.
The challenge to Google is very real. One analyst I spoke to pointed out that every small bit of advertising market share Google loses is likely reflected by a disproportionate drop in share price. As James Vincent points out, current chat implementations obviate the Web clicks that drive advertising. And increasing costs can only add to that pain. This is a fascinating innovator's dilemma.
The essays I have written looking at the impact of ChatGPT have been read at least 675,000 times this week alone. So I’ll be continuing my analysis of the competitive, business and economic impacts of these tools in a series of commentaries for members only over the coming weeks.
Meanwhile, dig deeper:
In The Heart of ChatGPT’s darkness, Gary Marcus writes “we now have the world’s most used chatbot, governed by training data that nobody knows about, obeying an algorithm that is only hinted at, with ethical guardrails that only sorta kinda work.”
The fine balance of using chatbots (and LLMs more generally) in science.
Toolformer, an LLM that can learn to use external tools such as calculators and calendars.
Key reads
Escape from OpenAI: The internet has become riddled with examples of DAN, ChatGPT’s alter ego that can evade OpenAI’s ethical guidelines. “Do anything” DAN can be invoked using a carefully written prompt, which is already on its 6th iteration by some accounts. Seeing the chatbot resemble a rogue, slightly corny villain from 80s science fiction is amusing. But it also shows how much more deeply OpenAI will have to embed their guardrails into ChatGPT’s system - if any prompt can reach DAN, someone will find it - while also keeping the bot’s generative abilities. However, as AIs improve, algorithmic fixes might not be enough to address the fundamental AI “containment” problem.
China’s counter-ban: China recently banned the export of key solar panel technologies. With this, it is looking to solidify its market dominance in solar panel manufacturing, where it produces over 80% of the world’s output. The ban is a response to the United States and the new rules of the game they have set with their trade ban on China and the reshoring pushed in the Inflation Reduction Act. This is a reminder that China will not be bullied by the US. China has the largest manufacturing capacity for several clean energy technologies. They have the power to act as the gatekeeper of the clean energy transition. For it to be successful, international cooperation will be key and that will require some better relations with China.
Ships in the night: Two of the largest shipping companies Maersk and MSC are parting ways because of their divergent predictions on the future of global trade. Geopolitical tensions, climate change and its mitigation, and wider supply chain disruptions are leading countries to reshore (the shifting of manufacturing capabilities back to their own land). This phenomenon could be bad news for maritime shipping, which is why Maersk is trying to become an integrated logistics company. We’re also seeing friendshoring - manufacturing in allied countries - which could dramatically change the structure of trade routes. MSC is counting on this to increase its fleet, but remain a traditional shipping company.
Read also: our Charts of the Week edition on whether India could become a manufacturing giant.
Market data
Researchers estimate that broadband internet expansion led to a 14% decline in the natural unemployment rate in Norway.
Twitter only has 180k paying subscribers in the United States. This is less than 0.2% of monthly active users.
A backlash against influencer advertising? #deinfluencing reached 68 million views on TikTok. However, this is only 0.17% of the 40 billion views for #TikTokMadeMeBuyIt.
Charitable technocrats. India’s tech magnates account for 35% of charitable giving in the country (and only 8% of total wealth).
Short morsels to appear smart at dinner parties
🍿 Moonfire Ventures simulated 972 billion venture portfolios to find out how best to construct one.
😬 Top earners don’t necessarily have the highest cognitive abilities.
👑 436 years after Mary, Queen of Scots was beheaded, the secret code in her letters has finally been broken.
🔥 The breakthrough modular approach to scaling quantum computing.
🥀 The physics behind the speed of heartbreak.
Exponential knowledge community
Join our invite-only Slack by filling out this application [open on a rolling basis].
Members engaged on Slack are invited to join local meetup groups for in-person gatherings:
Members in London are coming together on February 23 for a community breakfast,
Members based in NYC are meeting on the same day, February 23.
Join our global member mixer event on February 22 [online].
What you’re up to:
Tilen Travnik’s vegan steaks are now available in Britain’s Waitrose supermarkets.
We asked our member Ussal Sahbaz how the EV community can support the families and communities devastated in the earthquakes in Turkey and Syria. See the end note for recommended avenues to help.
Brooks Hamilton writes about Copilot and its role in the AI (r)evolution of the software industry.
Luke Conley is launching Planet House, a workshop and co-living space in NYC focused on sustainability, ethical innovation and affordable housing for artists.
Share your updates with EV readers by telling us what you’re up to here.
End note
I asked a long-standing EV member, Ussal Sahbaz, who is based in Turkey, to speak about the situation after these devastating earthquakes where the death toll now exceeds 25,000.
From Ussal:
Last Monday, Turkey & Syria were devastated by the largest natural disaster in their modern histories — a 7.7 earthquake that hit the ancient city of Antioch and its surrounding area.
Once upon a time, Antioch was the third largest city in the Roman Empire and the world. On 13 December 115 AD it was destroyed by an earthquake, which resulted in 250,000 deaths. As we have seen this week, while technology advances exponentially, certain societal and political economy issues only remain the same.
If you would like to donate to the relief efforts, there are some excellent charities already active in the area. Below are three ways to donate globally:
Turkish Philanthropy Funds (tax deductible in the USA)
Turkey Mozaik Foundation (based and regulated in the UK)
NeedsMap, a reputable local charity, accepts crypto donations in ETH, USDC, USDT: 0xbe4cde5eeeed1f0a97a9457f6ef5b71eae108652