- Modern Chaos
- MC.20: Navigating the AI Impact on Economy - Insights from IMF Study and NFX's Guide About the AI Disruptive Qualities
MC.20: Navigating the AI Impact on Economy - Insights from IMF Study and NFX's Guide About the AI Disruptive Qualities
This week, delve into the anticipated AI shift in global economy, the startup vs incumbent battle in the AI era, Google's innovative Circle to Search feature, and Lumiere's stunning video generation model.
modern chaos issue 20
Modern chaos is a newsletter exploring tech and AI through the journey of a dev agency turning into a startup studio. We share our notes, analysis, and experiments, plus a bit of Emacs tips.
People are trying to anticipate the impact of AI on the economy and jobs, and it’s hard.
According to the IMF, it will be a major shift, especially for countries with advanced economies. They are talking about an astonishing 60 percent of jobs being impacted by AI.
Half of the exposed jobs may benefit from AI integration. For the other half, AI apps may execute key tasks currently performed by humans, lowering labor demand and reducing hires. In extreme cases, jobs may disappear.
In contrast, our poll from last week asking, "What impact will AI have on your workflow in the coming year?" shows that adoption is likely just starting. As readers of a newsletter about tech and AI, you are at the frontline and early adopters of the tech, yet roughly half voted that AI will not influence or will only have a minor impact on their workflow in the coming year.
We are still at an early stage, but the investments in capital are massive. I bet that the adoption, as mentioned in the study, will happen in the next 5 years. Every week, we see new tools, apps, and use cases. A lot of people are experimenting; the pace is literally a race.
AI will significantly transform my workflow: 18.20%
Only a minor impact: 36.40%
No, AI will not influence my workflow: 18.20%
Uncertain about AI's role in my workflow: 27.20%
A note on the IMF study (direct link):
Their model is calibrated to the United Kingdom.
The simulated scenarios assume a labor share decline in line with comparable historical episodes associated with automation.
It focuses on the impact over a period of the first 10 years following adoption. However, the pace of AI adoption is difficult to foresee, and they draw a parallel with the introduction of general-purpose technologies, such as electricity, which was hard to predict at the time as well.
In this issue:
🤖 NFX: Startups vs Incumbents in the AI Era
🕵️ Dark Visitors: A list of known AI agents on the internet
💻 The Verge: Meta will own more than 340,000 H100 GPUs
📊 Helicone: Monitor your LLM-application
🌕 Moondream: A tiny vision language model that runs everywhere
📦 FireLLaVA: the first commercially permissive OSS LLaVA model
📱 The Verge: Google’s Circle to Search — POLL
🎨 Chase Lean: 10 Useful DALL-E Prompting Tips
📹 Lumiere: A Space-Time Diffusion Model for Realistic Video Generation
Updates & tools
NFX: Startups vs Incumbents in the AI Era
Pete Flint wrote an extensive piece about the disruptive qualities of AI and how it plays out for startups and incumbents. Understanding these dynamics in a market is key to positioning your startup. MORE
Dark Visitors: A list of known AI agents on the internet
Insight into the hidden ecosystem of autonomous chatbots and data scrapers crawling across the web. What I'd like to see is the trend and the rate at which top sites are blocking AI agents. I'm also curious about how they come up with their current numbers (robots.txt analysis, I guess). MORE
The Verge: Meta will own more than 340,000 H100 GPUs
Some people may be jealous. Meta shows again how powerful the company is. Zuckerberg clearly states his intentions that he will not pass on artificial intelligence. We can hope that the strategy and resources from Meta will bring value to the open source community as they did with Llama 2. MORE
Helicone: Monitor your LLM-application
Monitor and save the conversations between your service and OpenAI so you can use them to fine-tune Mixtral and reduce your fees. They are launching a fine-tuning service, which seems to be their best bet for making money. Free for up to 50,000 logs. Follow the playbook from Zachary to lower your OpenAI costs by 99% using Helicone. MORE | PLAYBOOK
Moondream: A tiny vision language model that runs everywhere
1.6B parameter model built using SigLIP, Phi-1.5, and the LLaVa training dataset. The model is released for research purposes only; commercial use is prohibited. MORE
FireLLaVA: the first commercially permissive OSS LLaVA model Moondream and this model are early signals of the availability of locally usable multimodal models. They will trigger many interesting apps like the context switching tool we talked about, which are difficult to do without local models due to privacy concerns. MORE
The Verge: Google’s Circle to Search makes visual searches super simple
A new kind of UX from Google, enabled by AI. You can circle anything on your phone; it takes the picture and makes a search on your behalf that you can guide. It's an interesting experience powered by a multi-modal model. I'm curious to see a response from Apple. MORE | VIDEO
POLL: Do you believe Google's new Circle to Search feature will change user behavior?
Google's new AI-powered feature allows users to circle anything on their phone, which then triggers a search. Will this innovative UX alter usage patterns or is it merely a flashy addition?
Chase Lean: 10 Useful DALL-E Prompting Tips
Chase experiments daily with image and video generation, so he knows a bit. If you are using Midjourney, DALL-E, or others, it's worth a look. MORE
Lumiere: A Space-Time Diffusion Model for Realistic Video Generation
Despite the departure of numerous researchers to other firms or to initiate their own ventures, Google continues to demonstrate its ability to develop cutting-edge technologies. Check out the samples; they're jaw-dropping. MORE