đ Quick Thoughts On Generative AI
Microsoft is the GOAT of bundling. I have never seen an organization, the size of Microsoft, integrating a new product/service in their stack at the speed and executional excellence that Microsoft demonstrated last week. Impressed and inspired!
I am very skeptic on the long-term viability of all the âcoolâ generative AI startups. The reason is because âmoat buildingâ in generative AI is extremely difficult + infrastructure costs are very high. Entrepreneurs that dontâ really understand the AI stack are doomed. Martin Casadoâs post is a must read.
There is an elephant in the room in the public discussion about the future of search. SEO. I am personally not convinced that SEO, the way is currently structured, has a future in the newly formed generative AI + search landscape. Why would anyone click on a link for an article that is perfectly summarized and even enhanced by AI? I see two potential outcomes: a) high-quality, long form content will be rewarded b) we may start seeing blockchain-based attribution systems where search engines are rewarding sources with tokens. There are some super early prototypes like this and this.
As with all things, itâs good to read a variety of different opinions about the current thing which in this case is AI. A good contrarian thinker is Gary Marcus.
The notion that AI will replace knowledge workers sooner than drivers for example was ridiculed a few months ago. Who is laughing now? Or as Sam Altman argues, what will be cheaper in a few months? Have someone build a mobile app or fix your toilet?
âđ˝ Reflections On Generative AI
On Consilience Mindset: for those who follow/read this blog, they already know that I am a big advocate of combinatorial and multi-disciplinary thinking. I had briefly written about this in my previous post but I want to double-click on it because a) I am coming across more individuals and initiatives that approach this type of thinking as âmeta-scienceâ and b) leaders and organizations seem to be on a âconsilienceâ race with new LLMs + X use cases coming up every day.
Meta-Science
Josh Wolfe, founder of Lux Capital, on how he leverages insights from fields like myrmecology (the study of ants) and physics to inform his work as a venture capitalist.
SantaFe Institute studying Applied Complexity
The Consilience Project publishing research on global risk mitigation, governance design and culture.
LLMs + X Use Cases
GraphGPT, an LLM building knowledge graphs. Organizational knowledge is still a relatively archaic field with tremendous potential for disruption and transformation. Such solutions could unlock organizational potential and productivity in unprecedented levels. Ernst & Young agrees.
Golden using retrieval transformers + knowledge graphs + human verification to solve the âconfident bullshitâ problem of LLMs. Imagine a much smarter and transformative Wikipedia.
Replit, a rapidly growing integrated development environment (IDE) that integrates conversational AI to make software building faster, smarter and simpler.
PlaygroundAI, text to image generator. Works like magic!
On AI Restarting the History In Tech: one of the most influential books of the past decade was Fukuyamaâs âThe End of History and the Last Manâ. Fukuyama actually argued that history should be viewed as an evolutionary process, and that the end of history, means that liberal democracy is the final form of government for all nations. There are many thinkers who believe that the same principles applied to Big Tech, especially the past decade. In a sense, Big Tech had been reaping the benefits of strong innovation, economies of scale and bundling moats for many years and had reached a point of complacency without an urgency to further innovate. The accelerated pace of innovation in AI, the release of ChatGPT and the amazingly fast âtime to marketâ mindset of Microsoft changed these dynamics almost overnight. Google publicly announced a half-baked competitor to Bing ChatGPT, very exciting startups emerged with vertical-specific LLMs and the AI war just started. The tech industry, and especially Big Tech, seems to have woken up again and I am confident that only good things can come out of this both for end users/consumers but also for the future of innovation all up. In a way, these huge organizations became what Samo Burja describes as âlive playersâ. Buckle up!
On Whether AI Can Impact the Real World: even if you are living in a cave, you must have read somewhere about Open AI, Chat GPT and the world-changing impact that Generative AI will bring. I am personally (moderately) bullish on Generative AI and the explosion of use cases that will be built upon the AI stack but I am also skeptical on whether it can dramatically increase productivity or significantly accelerate enterprise/industrial innovation. Eli Dourado recently wrote some âheretical thoughtsâ on why AI can NOT actually change anything in industries like Energy, Housing and Transportation or even in TFP (total factor productivity) numbers. These are industries that shape a very big part of the global economy and our daily lives and they seem to be stagnant because of âhuman-madeâ blockers like policymaking and polarization. Maybe a half-full way to look at it is that AI will become a driving force for cultural rather than technological change. As we increasingly get exposed to more LLMs and use cases, it will become harder for us to come back to the âoldâ world of manual, boring and repetitive tasks and as a result we will all push towards more innovative policies that will enable us move faster.
On Authenticity as a Service: I have been thinking a lot about the macro effects of a world fueled by Generative AI. An emerging pattern that seems to become more and more clear in my eyes is the increasing need for an âauthenticity layerâ that will distinguish human vs machine generated content. What is a ârealâ video vs a deepfake? Did a student use ChatGPT for her assignment? These are fairly simple challenges to address and OpenAI has already announced some clever solutions like Classifiers. But the problems will soon become much more complex. For example, how many real individuals/humans have voting tokens? Can anti-fraud systems detect generative AI bots that apply for loans? In a few words, we need a Proof of Humanity solution that will allow us to build a layer of undeniable and objective truth if we want to avoid a dystopian future. The good news that there are technologies that can enable us achieve this. Blockchain + zero knowledge technologies have the potential to build an Authenticity as a Service type of industry. Imagine this as an evolution of what Oracles currently do. Bonus: I had started formulating this concept in my mind after reading a very interesting book about the science of history. Bonus2: Balajisâ talk on oracles + definitive truths. Bonus3: without an authenticity layer, we will never know if we are dancing to âfakeâ David Guetta songs!
đĄMIND CRUNCHES
I have been a big advocate of biomimicry and I am always stoked when I see tech innovation leveraging insights from the âworld of natureâ. The latest one is about liquid neural networks based on wormsâ nervous systems.
Works In Progress film on ARIA, a UK modern version of DARPA, is a must watch. Bonus: ARIAâs chairman, Matt Clifford, has a really interesting blog and podcast. Highly recommended!
How to create more exceptional individuals. TLDR: homeschooling. Bonus: consider this post as a sequel to Erik Hoelâs âwhy we stopped making Einsteinsâ.
Byrne Hobart on how companies can fight (or manage) entropy. In a sense, this is another great article based on biomimicry.
âThe McKinseys and the Deloittes have no expertise in the areas that theyâre advising inâ - Mariana Mazzucato woke up and chose violence!
Brand-new report from the Collective Intelligence Project, an incubator for new governance models for transformative technology. Personal highlights: how to rebuild tech institutions leveraging tools like quadratic funding, DAOs and FROs. Reminder: I had also extensively written about the need for âgovernance innovationâ. Bonus: a new report on how collective response tools, like Pol.is, can build a Generative CI loop.
Why LLMs canât really help us understand biology.
A viral post about the complacent and bureaucratic mindset inside Google. Must read! Unfortunately, it applies to most Big Tech companies and is connected my point about the âend of historyâ above.
Web 2.5 is a term that I have been personally using, advocating and popularizing the past year because I truly believe it captures the current, true state of the Web3 industry. Apparently, I am not the only one! Really thorough blogpost.
John Carmack on AGI and the unavoidable groupthink effect. One of the best interviews on AI I recently read. Bonus: Carmack makes a point that AI isnât new and that it has gone through different cycles of hype, adoption and ridicule. It seems to me that Web3 is in an exactly similar situation.
Life hack that I am personally vouching for: âPick lottery tickets with high convexityâ
I am frequently asked whether there is any kind of meaningful relationship and solid business model between Web2 and Web3 organizations. This paper summarizes my thoughts very well.
Replit is an amazingly interesting startup and I expect them to become an increasingly popular choice for developers and builders. Amjad Masadâs, Replit CEO, interview about the future of technology is really good. Really really good.
Recommended book: Pieces of the Action, Vannevar Bush. A great collection of essays on how innovation really works.
Recommended podcast: Creators, creativity and technology with Bob Iger.
Recommended Newsletter: Experimental History, by Adam Mastroianni
Quote of the month: âGoogle is the 800 pound gorilla in search and I hope that with our innovation they will come out and dance and I want people to know that we made the danceâ Satya Nadella, during one of his Open AI interviews
Photo of the month: Sunset in Seattle (not generated by DALL-E)