🔮 The race to AI assistants; OpenAI exodus; GLP-1, risky films, insect protein ++ #474
An insider perspective on AI and exponential technologies
Hi, I’m Azeem Azhar. In this week’s edition, we explore Google and OpenAI product announcements and evaluate if they live up to the companies’ substantial investments.
In the rest of today’s issue:
Need to know: OpenAI loses the yin to its yang
The latest departures at OpenAI could deepen a monoculture at the leading AI company.
Today in data: Trade vs. the energy transition
The US is going to set a massive 100% tariff on Chinese exports, potentially harming its energy transition in the process.Opinion: Batteries as key transition technology
In this week’s deep dive for members, we do a survey of the state of the key technology in the energy transition – batteries, and in particular lithium-ion batteries.
💥 Today’s edition is brought to you by Sana, your new AI assistant for work.
Sunday chart
The Magnificent Seven’s combined Capex and R&D spend last year dwarfed global VC investment. But, where is this staggering sum going?
We get a sense of the spending breakdown with their announcements this week.
Google and OpenAI announced models, GPT-4o and Project Astra respectively, with the main advancement not being a dramatic leap in reasoning performance, as seen from GPT-3 to GPT-4, but rather one small step in usability.
These announcements presage the first signs of the advent of the AI assistant era for LLMs – models that can engage in real-time dialogue with lifelike voices, modulating pitch, tone and intonation, and perceive the world through video via a phone or even smart-glasses. Overcoming UX barriers (mainly latency and methods of interaction) may appeal to consumers using these models. For example, this Google demo, which does an impressive job building a science notebook for a school kid. Which is the better model? It’s hard to say from the demos, Project Astra did watch the OpenAI presentation.
Yet, perhaps the most socially significant announcement is OpenAI’s decision to make GPT-4 free.
At EV, we have consistently encountered people claiming LLMs are just not good enough at cognitive tasks to be truly useful. However, while these models have limitations, such critics are often basing their judgments on GPT-3.5, not the far more capable GPT-4. Making a GPT-4-quality model freely available will likely widen its use, helping the broader public understand the capabilities and potential of these AI assistants. Coupled with OpenAI’s impending contract with Apple, it’s difficult to imagine ChatGPT not being in the pockets of a much greater number of people within a few quarters.
And, for big tech companies, this needs to be the case. OpenAI burned through $520 million in cash in 2023, drawing scrutiny as to whether the extraordinary investments these companies are making are justified. In the New York Times, journalist Julia Angwin offers a critical perspective, arguing that AI often performs tasks poorly and unreliably, rather than being on the verge of becoming an all-powerful tool. While not entirely wrong, this view is incomplete. I have spoken with leaders from many businesses that have already found numerous beneficial applications for AI. These improvements in model usability are likely to increase engagement and uptake. It might not feel like a GPT-3 to GPT-4 technological leap, but it could certainly be an adoption leap.
See also:
GPT-4o’s interactions in its demos sometimes border on flirtatious – a potentially dangerous level of anthropomorphism. With people already forming “relationships” with AI, one might wonder, why bother with messy, complicated human beings when GPT-4o seems perfect? For a popular exploration of this question, watch the movie Her. For a comprehensive analysis of the ethical and societal implications of advanced AI assistants, explore this research paper co-authored by EV reader Iason Gabriel. The paper delves into the potential impact on our lives, the challenges of autonomy and alignment, the complexities of human-AI interaction, and the broader societal effects, while emphasising the need for responsible development through collaboration between researchers, developers, policymakers and the public.
🚀 Today’s edition is supported by Sana.
Sana AI is a knowledge assistant that helps you work faster and smarter.
You can use it for everything from analysing documents and drafting reports to finding information and automating repetitive tasks.
Integrated with your apps, capable of understanding meetings and completing tasks in other tools, Sana AI won’t just change the way you access knowledge. It’ll change the way you work.
Key reads
OpenAI culture wars. Ilya Sutskever, OpenAI’s co-founder and former chief scientist, and Jan Leike, co-lead of the super-alignment team, have resigned from the company. OpenAI has undergone a significant cultural shift, which became apparent last November during the failed coup against CEO Sam Altman. Those concerned about AI’s existential risk have been ousted. As I’ve previously discussed, I assign a low probability to doomsday scenarios, and believe that focusing on existential risk diverts attention from more pressing concerns. However, silencing cautious voices may lead to potential risks being overlooked. A balance is needed between considering existential risks and the traditional tech mentality of “build fast, fix later”. The concern is that with the departure of Ilya and Jan, OpenAI has lost the yin and is left with only the yang.
Catch me if you can. The long-awaited US AI roadmap “Schumer plan” has been published. The focus and tone is notably different from that heard in Congress hearings, where apocalyptic concerns were at the forefront. Instead, it has a clear priority: cementing America’s AI competitive advantage through industrial policy. It wants to dedicate at least $32 billion annually towards innovation and R&D. The focus of the roadmap may have come from big tech lobbying Washington to hammer in one key point: the real risk is China overtaking the US, not runaway AI. The roadmap is not binding, but it is another layer to the emerging picture. While the EU focuses on the risks of something going wrong, the US is focused on the risks of losing the tech race. I actually think it’s quite a reasonable roadmap.
Baby doom. The UN reports a dramatic decline in global fertility rates, from 4.86 live births per woman in 1950 to 2.32 in 2020, dangerously close to the 2.2 replacement rate required to maintain population stability. Advanced economies, such as the US (1.62) and South Korea (0.72), have already dipped below this threshold, raising concerns among governments about the potential consequences of a shrinking workforce and sluggish economic growth. A second demographic transition may be underway, characterised by a shift towards individualism and away from traditional family structures. Combined with increased lifespans, this necessitates a reevaluation of life stages, argues Martin Wolf in the FT. Conventional linear stages of lifespan – education, work then retirement – appear increasingly outdated. Education should become a lifelong pursuit, not just the province of youth, and four-day working weeks may provide more time for people to consider starting families. Unless, of course, we all opt for AI companions instead.
See also:
Demographic collapse is one of the reasons we think China will not surpass the US in the long run. Recommended read.
The hottest drug in town. GLP-1’s meteoric rise continues with a new long-term study finding that people who received semaglutide saw sustained weight loss for four years and a 20% reduction of major cardiovascular events. It also has other potential applications. Trials are investigating its capacity to slow Alzheimer’s. Combining it with an NMDA antagonist1 could also amplify its effects. This combination may alter neuronal plasticity, suggesting long-lasting benefits without chronic use. As alternatives to Novo Nordisk’s Wegovy and Ozempic emerge, prices will likely drop. Roche’s experimental obesity medication boasts an impressive 18.8% placebo-adjusted weight loss after 24 weeks. As these drugs become ubiquitous, their ability to curb cravings could trigger significant shifts in consumer preferences, with far-reaching economic consequences. As covered in EV#447, analysts have already expressed concerns about the potential impact on the fast food industry.
See also:
For a deep-dive on GLP-1’s potential listen to my podcast with Eric Topol.
Newsreel
OpenAI released GPT-4o. See our Sunday Chart for additional context if you haven’t yet.
Google made many announcements this week, including doubling the context window of Gemini, introducing their latest state-of-the-art image generation model, unveiling a video generation model rivalling Sora, and launching the sixth generation of their TPU chip.
The US is going to set a 100% tariff on Chinese exports. As climate author Assaad Razzouk points out, this won’t affect China too much due to the scale of their energy transition – for the US, on the other hand, it’s going to slow their transition down a lot.
France blocked TikTok in its overseas territory of New Caledonia amid demonstrations that turned violent.
Wiley is closing 19 journals due to large-scale research fraud that has occurred over the last few years, including 11 affected by paper mills, leading to over 11,300 retractions. This is likely to only accelerate with AI. For a deep dive on AI and knowledge, see this paper on the epistemic dangers of AI technology for democracy.
Data
Microsoft reports 30% more emissions than in 2020, the main reason being due to investments in data centres.
Three times as many people strongly support natural gas compared to methane for electricity generation in the US, despite them being one and the same substance.
Only 2% of Tesla FSD free trial users end up buying the software according to credit card data.
The Public Investment Fund of Saudi Arabia has significantly reduced its US equity holdings, from a high of $55 billion at 2021’s close to just $18 billion by the end of the most recent quarter.
As many as 13% of people questioned in the UK say they would be willing to regularly eat insect protein.
More than 82% of Americans would support polygenic embryo screening (to test for traits) as part of the IVF process.
Short morsels to appear smart at dinner parties
🌎 The first quantitative meta-analysis of carbon pricing evaluations finds “robust evidence” that existing CO2 pricing schemes are effective in reducing greenhouse gas emissions (via EV member Rafael Kaufman). See also, new research published in a working paper indicates that the economic toll of climate change is six times greater than prior estimates, with a mere 1°C rise in global temperature triggering a staggering 12% drop in world GDP.
📜 How translation went from the vague capturing of the spirit of a text in the romantic era to a highly standardised process in the 20th century.
🎹 Beethoven’s hair holds clues as to why he went deaf.
🇨🇳 China is secretly building drone carriers.
⛪ For the first time in the US, a non-Christian church wins a battle to import ayahuasca for its religious practices. (Via EV member Josh Hardman)
🎬 Films that promote risk-taking sell more in entrepreneurial societies, while those promoting stereotypical gender roles perform better in more traditional societies.
💸 Wisconsin becomes the first US state to disclose the purchase of bitcoin.
End note
If I had to pick out one unifying theme from this week’s wondermissive, it would be the global geopolitical and economic realignment being catalysed by technological change.
The Schumer AI plan, which focusses on US technological superiority in AI, needs to be read in context of the growing tariffs levied by the Western administrations on Chinese climate tech transition exports. (See EV#472, for example.)
These mirror past approaches to maintain a geopolitical advantage through technological superiority.
Yet, that singular theme understates the impact of demographic shifts. Falling populations will restructure societal values, economic preferences and military priorities. Ageing populations change dependency ratios and increase fiscal drag, creating less room for policy. Thinning labour markets leave nations with the choice of increasing offshoring, robotisation or immigration: all contentious policies at this moment. Countries with declining populations may prefer to send drones rather than daughters and sons to war.
Things to mull over your Sunday coffee.
Enjoy!
Azeem
What you’re up to — community updates
Simon Knowles has been named a Fellow of the Royal Society.
Paul Currion contributed a short-story to the anthology All Tomorrow's Futures.
Carl Frey was interviewed on AI’s impact on wages.
Share your updates with EV readers by telling us what you’re up to here.
The same class of drug as ketamine, the darling of both the therapy and rave scene
You quote Martin Wolf's review of Andrew Scott's latest book, The Longevity Imperative. I don't think I'm going to pick up the book because, even if the numbers are updated, the message appears to be pretty much the same as in the 2016 book he co-wrote with Lynda Gratton, The 100-Year Life (https://www.100yearlife.com/). Or is there something new that I'm missing?