🔮 Who is winning the AI talent wars? EV#467
Sunday briefing from Exponential View: Perspective, news and data digest to make you smarter about AI and exponential technologies
In this week’s edition, we explore who is winning the AI talent wars, and…
What Taiwan can teach us about the new era of disinformation.
Balancing AI assistance and learning in education.
The pitfalls of “conceptual borrowing” between AI and cognitive sciences.
🐙 Special thanks to our sponsor OctoAI. Mixtral on OctoAI: 12x lower costs, 30% faster speeds vs GPT-4-Turbo.
Sunday chart: The AI industry is nothing without its people
Microsoft is dominating the AI talent battlefield. Their tactics, most recently the strategic acqui-hiring of Inflection, have amassed significant talent pool. Even excluding their close partnership with OpenAI, they have one of the industry’s most gifted in-house teams.
Microsoft and the other four largest hirers—Google, Meta, Amazon, and Apple—have collectively hired 10x more AI talent than the next five largest hirers, cementing their position as the industry’s most voracious consumers of AI expertise.
Disruptor labs such as OpenAI, Anthropic, Mistral, and Cohere are notably absent from this equation. Their headcount is tiny relative to the big tech firms, although their people are amongst the most sought after. Geographically, while the US leads, Paris and London are building AI ecosystems around Meta and DeepMind alumni.
Poaching is becoming rampant in the industry. Companies are resorting to increasingly aggressive tactics to secure and retain top performers.
Meta has abandoned its long-standing policy of not matching rival companies’ offers and is now willing to exceed them to keep valued employees. Zuckerberg himself is penning heartfelt epistles to woo potential recruits. Google’s Sergey Brin is personally reaching out to departed employees, imploring them to return—a tactic that has proven effective. OpenAI is dangling compensation packages ranging from $5 million to $10 million to entice top Google researchers. Even freshly minted industry PhDs commanded salaries upwards of $800,000 when joining OpenAI and Anthropic last year.
Yet, amidst this frenzied competition, it would be a mistake to assume that the tech giants have wholly absorbed the pool of AI talent. A significant 88.6% of AI hires are outside the Big 5, highlighting a vast untapped reservoir of expertise. An AWS survey reveals that employers are prepared to offer a 30% salary bump for IT professionals with AI skills—although, given the exorbitant salaries commanded by PhDs, they may need to dig far deeper to secure the crème de la crème.
See also: China has eclipsed the US as the biggest producer of top AI talent, generating almost half of the world’s top AI researchers compared to 18% from US undergraduate institutions.
🚀 Today’s edition is supported by OctoAI.
OctoAI is the fastest, most affordable option to run Mixtral 8x7B, and the engineering effort to switch is negligible as the API formatting is the same as OpenAI. Customers making the move see comparable quality, 12x savings, and 30% lower latency when switching from GPT-4-Turbo.
Try OctoAI Text Generation and get $10 in free credits when you sign up today.
Key reads
A new propaganda playbook. As has become abundantly clear, AI-generated content is the new chameleon in the information landscape, making falsehoods more sophisticated, elusive and virulent. Microsoft researchers Madeleine Daepp and Robert Osazuwa Ness highlight Taiwan’s recent elections as a cautionary tale. During Taiwan’s 2020 presidential election, the fact-checking platform Cofacts responded to 80% of queries within 24 hours. However, by the next election, their response rate plummeted to 15% on election day, overwhelmed by the sheer volume of misleading content. Additionally, attempts to crash Taiwanese news websites spiked by 3,370% in the three months leading up to the elections, illustrating a calculated push towards the chaos of social media for news dissemination. The solution lies in proactive collaboration and adaptation, leveraging alliances between big tech and civil society, and employing tactics such as “prebunking” videos to combat disinformation preemptively. In an age where AI redefines the landscape of information, adaptability becomes indispensable.
See also:
The UK intelligence community must evolve from a closed, centralised model to an open, collaborative and cross-sector approach to effectively navigate the information age.
Russian disinformation groups amplified social media posts about the Princess Kate rumours, fuelling social tensions.
The student, not the LLM, is the real parrot. Darvishi et al. found that students who relied on AI for peer feedback struggled to maintain comment quality when the support was withdrawn, exposing the AI’s inability to cultivate authentic learning in feedback improvement. As AI permeates education and work, we must strike a balance between its assistance and the cultivation of genuine human competence. To maintain agility as AI continues to advance, fostering metacognition and self-regulation is not a suggestion, but a survival skill. On the other hand, let’s not overlook the unique potential of LLMs to assist learning. Another study, for instance, shows how LLMs can act as storytelling alchemists, transforming complex legal knowledge into accessible insights for non-experts.
Conceptual baggage. Floridi and Nobre argue that the “conceptual borrowing” between AI and Brain & Cognitive Science (BCS) has led to anthropomorphism in AI and computational reductionism in BCS. This cross-pollination has spawned misleading assumptions: in cognitive science, “attention” denotes selective concentration, while in AI, it refers to a neural network mechanism. The borrowed terminology falsely implies human-like capabilities in AI systems. While AI’s anthropomorphic rhetoric breeds inflated expectations, the jargon of BCS similarly falls short by neglecting the mind’s phenomenology. The authors envision a future where this confusion is dispelled: just as “horsepower” became a standard unit without reference to horses, so too may AI be assessed rationally, without imputing cognitive properties to computational systems.
See also: A paper by EV reader Murray Shanahan (who is also a research scientist at Google Deepmind and a Professor at Imperial College, London) suggests that the question of consciousness in AI systems, especially exotic forms like embodied generative agents, will be determined by our ability to meaningfully interact with them in a shared world and the evolution of our language to accommodate these strange new entities, rather than by some hidden inner quality.
Newsreel
Nobel laureate Daniel Kahneman has passed away. This 2021 FT interview with Kahneman is a charming testament to his genius.
Amazon pours another $2.75 billion into Anthropic. See my conversation with Anthropic co-founder and CEO Dario Amodei.
China bans Intel and AMD CPUs in government computers.
The EU Commission has opened non-compliance investigations against Apple, Alphabet and Meta under the Digital Markets Act. The US Justice Department is after Apple for its monopoly in the smartphone market.
Renewable energy sources have generated more power than natural gas power stations in the UK during the 2023/24 winter.
Data
Chinese-made electric vehicles are set to hold 25% of the European market this year. BYD, China’s leading EV manufacturer, has manufactured over 7 million electric cars so far.
In a study of 820 participants, GPT-4 with access to personal information increased agreement with its arguments by 81.7% compared to human debaters, while GPT-4 without personalisation still outperformed humans, but the effect was smaller and not statistically significant.
Micron, a leading memory chip manufacturer and a major supplier of high bandwidth memory (HBM), a crucial component for AI accelerators, has sold out its entire HBM supply for 2024 and most of 2025.
Russian weekly gasoline production has experienced a significant decline, falling 14.3% below last year’s levels, largely attributed to the impact of Ukrainian drone strikes.
Keytruda, a cancer immunotherapy, was the world’s best-selling drug in 2023 grossing $25 billion in sales.
Short morsels to appear smart at dinner parties
🫖 This New Yorker piece is the best thing I have read on 15 years of UK politics and is great for non-Brits.
🍻 An AI can predict what a beer tastes like and how to improve it, based on the chemical makeup. See also, researchers use ASCII art to jailbreak AI.
🧠 Researchers found that brain inflammation is essential in creating long-term memories.
🦠 Humans pass more viruses to other animals than we catch from them.
🍄 Gene-edited fungi are emerging as nutrient-rich, minimal-impact food sources.
🔬 Tweeting about your research paper will boost engagement, but not citations. On a similar note, giving senior scientists more money doesn’t incentivise them to produce more risky or impactful research.
💊 Amazon hopes that anti-obesity drug demand will boost its online pharmacy business.
End note
Madhumita Murgia of the FT and I will be in conversation in London next month. Madhu’s new book, Code Dependent, is debuting. She and I will chat about AI and how it has already infiltrated the everyday lives of humans worldwide.
The conversation will be held in the gorgeous Daunt Books in Marylebone. You can get tickets here. Hope to see you there.
And for the long Easter weekend, for those who have one, I’m releasing my next playlist, Straight Outta Cricklewood. High energy for your Bank Holiday party or time in the gym.
Cheers,
A
What you’re up to — community updates
Matt Clifford and Ian Hogarth received CBEs from King Charles III.
Share your updates with EV readers by telling us what you’re up to here.
Three ways you can work with us:
Book my keynote — I will bring insight and clarity about AI and exponential technologies to your next event.
Sponsor my newsletter or podcast — get your brand in front of 250,000+ early adopters and senior leaders.
Bring your team to Exponential View — organisations invest in their people’s development through discounted group access to Exponential View.
Stunning fact: "Microsoft and the other four largest hirers—Google, Meta, Amazon, and Apple—have collectively hired 10x more AI talent than the next five largest hirers"
However, how talent is distributed in the robotics world may be more important than LLM wars. Clearly, LLMs will be critical to robotics, but the big breakthroughs may include better models of the world—logical and physical. Since I've thought the same for a few decades and did not wake up to LLMs until 30 November 2022, my hopes are higher than my confidence.
Thank you for that amazing newsletter again 🤯
Just FYI your link on “Sponsor my newsletter or podcast — get your brand in front of 250,000+ early adopters and senior leaders” brings me to the App Store somehow 🤔