AI versus Marshy - Current challenges in AI

Welcome to this standalone article about the challenges in AI. I’m Marshy, and I’ve been exploring the intersection of tech, marketing, AI tools, and their potential impacts in my previous newsletters. If you’re new here, don’t worry – this article will still make sense, but if you’re curious about my previous work, feel free to check it out. Let’s dive into the current challenges in AI. The thing about making money with art I don’t read from news portals any more, I stopped checking them during COVID and my wellbeing improved and the world didn’t end. So its impressive that this Guardian article - “AI is not a one-time bomb, but a slow burn of devastation that is consuming jobs and culture” got through my filters. A doom-mongering headline - well I never… One of the points it makes is that the progression of generative AI is going to contribute to a sustained erosion of things we love TV, music, art etc. and impact people who work in this fields. It then talks about the AI advocate response, and observes what’s being looked at from a regulatory perspective. I’ve got three spicy takes for you on this descending order of complexity: AI is accelerating that gap and decommodifying creating versus owning . Cixin Liu articulates what it’d be like for an alien race to see our art, and to contribute to it: After the tenth year of the Deterrence Era, besides additional scientific information, Trisolaris began to transmit cultural and artistic products done in imitation of human models: films, novels, poetry, music, paintings, and so on. Surprisingly, the imitations were not at all awkward or childish; right away, the Trisolarans produced sophisticated, high-quality art. Scholars called this phenomenon cultural reflection. Human civilization now possessed a mirror in the universe, through which humanity gained a new understanding of itself through a novel perspective. In the following ten years, Trisolaran reflection culture became popular on Earth, and began to displace the decadent native human culture that had lost its vitality. Reflection culture became the new source for scholars seeking new cultural and aesthetic ideas. Liu, Cixin. Death’s End (The Three-Body Problem) (p. 138) Generative AI and whatever AI ends up being is no different - its going to be reflecting back what we’re doing and shining a light on things that are already there - but might not be a focus. There’s always been a disparity between creating, art, and valued art - but it’s now taking us to uncomfortable places. One of the solutions against this is providing UBI (Universal Basic Income) so that there’s a “base” level of consciousness we can all enjoy without that influence.Mind-bending stuff but if it resonated let me know - and if it didn’t the next piece is FAR more approachable. The system is incentivised to continue production Attention becomes so commodified that incentives continue to capture it The market shapes and influences our very consciousness Tool of the week: Ahref’s AI writing tools I’ve really enjoyed watching the trajectory of AI writing tools and have no qualms with what they’re doing so far. Writing clearly is hard . It doesn’t come naturally to a lot of people, so I think it helps (with caveats*). I’ve been doing some auditing of competitors for a client and the free tools appearing are becoming more helpful. ​ Ahrefs is an expensive SEO tool - but the AI tools section has some writing help for when you need a quick bio or to spritz something up. I popped in an LI post on fractional execs as a test, not quite what I’m looking for but could prompt some fun for when you’re noodling on what to say on a page or ad. Not sure if rhyming poorly counts as a “hook” So what are the current problems researchers are trying to solve with AI? A thoughtful post by Chip Nguyen on what the current BIG challenges are with current AI technology in Language Learning Models. The challenges in a nutshell: Reduce and measure hallucinations Optimize context length and context construction Incorporate other data modalities Make LLMs faster and cheaper Design a new model architecture Develop GPU alternatives Make agents usable Improve learning from human preference Improve the efficiency of the chat interface Build LLMs for non-English languages Like all things in breakthrough tech - I think some of these will get solved quickly and others will present more complexity than people give credit. Non-English languages is a big one. The collective earth doesn’t “think” in English - so our technologies shouldn’t either. Sam Altman (CEO, Open AI) mentioned this is a big priority for them (see AI versus Marshy #1 ). From the first newsletter! Other ones - like “making agents usable” might be closer than we think. This demo from Humane (screenless wearable AI tech) is really impressive - starting from the ground up with a “more screens is not the answer” from someone who spent 22 years designing for Apple. The real story is his wife’s name is in lower-case - why? – What do you think there? I’m quite happy filtering through all this AI noise to give you more informed takes and the reality is none of us really know. This quote appeared from my highlights today, and I thought it was apt: This is what I’m striving to do I’m keen to continue pushing through these topics and share what I think and appreciate you coming along for the journey! Thanks! 🙏🏻 -Marshy p.s. I was up at 2am Sunday for a 90-minute webinar from Alex Hormozi. He wrote $100m Offers , and just launched $100m Leads . It was a ridiculous masterclass on Internet Marketing guru delivery - with 500k+ people on the call. You can watch the livestream here if you also feel like donating that amount of time at a more sensible hour 😂 ​ Originally appeared in newsletter : AI versus Marshy #10: the creativity and profit gap, a tweaky freebie, + current challenges