🔮 What is Tesla?; LLM gene editor; China rivals GPT-4; crypto, Kant & conscious molluscs ++ #471
Hi, I’m Azeem Azhar. In this week’s edition, we explore all the layers that make up Tesla’s identity.
And in the rest of today’s issue:
Need to know: Research breakthrough
Biotech company Profluent has achieved the first successful precision editing of the human genome using a programmable gene editor designed with large language models.Today in data: AI as fast as Bolt
Groq is serving Llama 3 at more than 800 tokens per second. For comparison, GPT-4 Turbo runs at 23 tokens per second.Opinion: The bedlam of AI
Where does AI development go from here?
Sunday chart: Tesla’s identity crisis
What is Tesla? Its ambition defies simple categorisation as it straddles the boundaries of the automotive, AI, robotics and energy industries – and, now, the taxi industry, too. Elon Musk said in Tesla’s latest earnings call that it is best to identify Tesla as an AI and robotics company. Yet its historical revenue streams paint a decidedly different picture — that of an electric vehicle manufacturer.
Why the new identity? Perhaps a combination of vision and necessity. Gone are the days when Tesla had the EV market to itself. Today, the arena has become fiercely competitive, with incumbent ICE automakers and a new breed of Chinese companies vying for dominance. Tesla has faced a rocky start to the year, seeing profits decline by over half. The EV market — in stark contrast to the winner-takes-all dynamics that internet-era companies have grown accustomed to — is shaping up to be a relentless battleground where no one’s dominance is assured. If Tesla settles for only being an auto company, it won’t be a big company.
However, Tesla has grand visions. The company possesses unique, albeit currently unmonetised, characteristics that hold immense potential if properly aligned. Having collected over 300 billion miles of data from its “Full-Self Driving” systems, Tesla is amassing the raw material needed to refine its AI models — a crucial advantage in the autonomous vehicle race. And, by the end of 2024, Tesla’s acquisition of over 85,000 H100 GPUs will position the company as a top consumer of these powerful processors, rivalling the major tech giants. This amount of compute and data gives it a real edge over its competitors.
If successful, this technology could be a paradigm shift for car ownership, replacing individual possession with efficient, city-wide fleets that can be summoned on demand. Musk even envisions idle Teslas serving as a distributed compute network, akin to Amazon’s transformation of excess capacity into AWS. However, this grand vision hinges on solving the complex puzzle of autonomous driving. Numerous edge cases need to be conquered, as a recent prank on Waymo demonstrated — when a T-shirt with a stop sign on it tricked one of its robotaxis into stopping.
Yet, if this fails, Tesla has another potential identity — that of an energy business. While generating relatively small sales, it is experiencing rapid growth and a record margin of 24.6%. The world needs batteries, and Tesla’s are quite good. Combined with the widespread charging infrastructure it has established, the company could find a decent business in energy.
These incarnations all hold promise but therein lies the crux of Tesla’s identity crisis: there is no grand future for the company as a mere EV manufacturer, yet as its revenues show, it cannot escape its roots, still reliant on the core business of selling cars. The EV competition has triggered somewhat of a rush to cement its new identity, but until the way is clear, Tesla must continue to sell cars.
Key reads
Genomic Genie. Biotech company Profluent has pioneered a method for creating novel gene editors using LLMs. These AI-designed proteins can be employed with CRISPR technology to precisely edit the human genome, surpassing the capabilities of naturally occurring gene editors. Profluent has open-sourced its most successful creation, OpenCRISPR-1, which operates similarly to CRISPR-Cas9 but with improved precision1. This breakthrough marks a significant milestone in AI-aided genetic engineering, tangibly demonstrating the potential of biological generative models, as highlighted in our previous Chartpack. While clinical trials are required to validate the technology’s efficacy and safety, Profluent’s achievement in generating 4.8 times more protein clusters than found in natural CRISPR-Cas families is testament to the power of generative AI in pushing the boundaries of what is possible in genetic engineering.
The end of the world as we know it (and I feel fine). The world is fine; it’s time to bury the existential risk debate. The doomer crowd’s use of Bayesian reasoning to legitimise these fears is little more than pseudo-scientific hand-waving. Even by playing their game, my forecast for the probability of existential catastrophe by 2050 is a mere 0.000131%. And even the most pessimistic super-forecasters and AI experts put the risk at just 0.38% and 3% by the end of the century. These small probabilities, projected over such vast timescales, contingent on so many factors, should not dominate the conversation.
As Jack Stilgoe astutely observes in his op-ed: this irrational fixation on existential risks “sidesteps questions of who the winners and losers will be”, allowing AI companies to evade responsibility for the real and present dangers they pose, hiding behind the grandiose excuse of planetary salvation. Policymakers should focus their attention on the voices of thinkers such as EV reader Rumman Chowdhury, who grapple with the tangible and imminent challenges of AI safety.
See also: A tome on The Ethics of Advanced AI Assistants. Timely, given the renewed focus on agents. Google, Microsoft, Meta and OpenAI all reportedly have task-fulfilling AI in the works.
Free (the labour) market. This week the US Federal Trade Commission passed a ban on non-compete agreements2 across the country. These agreements have shackled nearly 20% of workers and have long been viewed as obstacles to the free exchange of talent and ideas between companies and startups. California, the epicentre of American ingenuity, has already proven the efficacy of such a ban. Businesses, predictably, are up in arms, arguing that non-competes are essential for safeguarding trade secrets. However, their true concern lies in the potential shift in bargaining power towards workers, who will now have the freedom to seek higher-paying opportunities with competitors.
Alternatives such as patents, copyrights, NDAs and equity plans exist to protect intellectual property without stifling employee mobility. This move could propel the US to even greater heights despite the inevitable legal challenges from business lobby groups. Non-compete agreements have historically favoured established companies at the expense of smaller businesses. The FTC projects a 2.7% increase in new business formation as a result of this ban, further strengthening the US’s innovation potential.
See also: Inside the struggles of chip manufacturer TSMC’s expansion into the US.
Newsreel
The US will force TikTok owner ByteDance to sell TikTok or face a ban in America. ByteDance says it would prefer to shut the app down if legal options to fight the ban fail.
Microsoft unveiled Phi-3, a 3.8bn parameter model — achieving GPT-3.5 level performance on agriculture benchmarks. Apple has also released OpenELM — a series of small models ranging from 270m to 3bn parameters. As I said in my yearly outlook: “Smaller models will be in demand: they are cheaper to run and can operate in a wider array of environments than mega-models.”
Chinese company SenseTime has released SenseNova 5.0, which is reportedly better than GPT-4 in a range of scenarios.
After a six-year hiatus, Stripe will start taking crypto payments again.
Scotland’s coalition government has collapsed amid disagreements over climate targets.
Sequoia Capital plans to back Elon Musk’s AI startup xAI.
Meta’s AI spending spree spooks investors despite strong revenue growth. It has also introduced multi-modal AI to its Ray-Ban smart glasses.
Data
Hedge fund Tiger Global was SurveyMonkey’s biggest customer in 2019, spending over $1.7 million yearly on its products.
Global military spending surged to $2.44 trillion in 2023, marking a 6.8% real-term increase from 2022 — the sharpest annual rise since 2009.
According to a survey by Morgan Stanley, after starting GLP-1 treatment (like Ozempic, for example), there was a 16% drop in respondents who smoked cigarettes weekly and a 14% drop in those who used e-cigarettes weekly. Meanwhile, around 56-62% of alcohol consumers reported drinking less alcohol, with 14-18% cutting their alcohol consumption entirely.
Groq is serving Llama 3 at more than 800 tokens per second. For comparison, GPT-4 Turbo runs at 23 tokens per second.
Battery storage was the fastest-growing energy technology in 2023, with deployment more than doubling year-on-year to add 42 GW globally.
Last week, battery storage became the largest source of supply to the California grid for the first time.
Short morsels to appear smart at dinner parties
💬 How workers in Africa are unwittingly crafting the language of AI.
🌚 The reality of terraforming on Mars. Not a short read, but an amusing one.
🦦 Scientists have extended the list of animals they believe have a “realistic possibility” of being conscious, from fish to cephalopod molluscs.
📚 Hypnovels leverages AI to create immersive previews of books. Potentially a new way of marketing literature?
🤔 Kant’s 300th birthday is marked this week. This is a great Sunday read on his philosophy of thought.
🪙 Jack Dorsey is apparently inching closer to creating the first standalone mining chip.
End note
We’re playing the game of “Waiting for GPT-5”. But unlike Vladimir and Estragon we won’t be left on a limb. Sam Altman is preparing us for its arrival.
OpenAI’s boss has said: “We can say right now, with a high degree of scientific certainty, GPT-5 is going to be a lot smarter than GPT-4 and GPT-6 will be a lot smarter than GPT-5, we are not near the top of this curve.” The curve Sam is talking about is an exponential one. He viewed GPT-1 to GPT-2 as baby steps on this curve, GPT-2 to -3 and -3 to -4 in similar manner. Yet, as end users we know the difference between those models didn’t feel like baby steps.
Sam tends not to hyperbolise OpenAI’s products. I put this down to being a product person (both before and during his stint at Y Combinator). So I would be surprised if GPT-5 felt like business-as-usual but a bit shinier. I’d expect something meaningful. I’d say that since the original GPT-4 was launched we’ve seen three meaty improvements: multi-modality; extremely long context windows; and faster, smaller models of similar output quality. Let’s see what -5 has in store for us.
cheers,
A
P.S. I will be speaking at several major events in Europe and the US in June, I hope to see you there:
June 5-6, Autonomous Innovation Summit hosted virtually.
June 11, Twilio’s Transform Together in London.
June 18-19, Future Talent Summit in Stockholm, alongside Yuval Harari and Vivienne Ming. EV readers get 30% off tickets via this link.
June 26, DASH 2024 by Datadog in NYC.
What you’re up to — community updates
Scott Farquhar is stepping down as co-CEO of Atlassian after 23 years. I hosted Scott on the podcast three years ago, when we discussed the evolution of teamwork.
Gianni Giacomelli will be speaking at the Autonomous Innovation Summit in early June as well!
Adam Oskwarek will be speaking on a panel discussing the question “Is the finance industry ready for the great wealth transfer?” during SF Climate Tech Week.
Ussal Sahbaz is moderating a digital policies roundtable at the Tech.eu Summit in London next month.
Share your updates with EV readers by telling us what you’re up to here.
They didn’t open-source the LLM that created it.
A non-compete agreement is a contract that prohibits an employee from working for a competitor or starting a competing business for a specified period after leaving their current employer.
Your concise comment hit home. "Linear thinking no longer applies in a world that is changing at an exponential rate. Since 2015, I have been researching, interpreting and discussing how emerging technologies will dramatically change every aspect of our lives.
Your insight is spot on. I had a left brain stroke while producing one of my TEDx Talks event. It immediately took my two left brains off line impacting my linear, logical, sequential thinking. My right non-Linear as I see them"Nous" brains have offset my damaged two left brains, and are wide open online and generating an infinity of ideas faster than I can apply them.
Cheers for the Deep Mind doc, shall read