This article gives a good overview about (most of) the startups building in the ecosystem and raises some interesting questions:
- Is AI the next big platform after Social, Mobile and Cloud
- How will value be created and captured (defensibility / moat / stickiness, winnter-takes-most market?, will there be a new market built on top of the IaaS Players?)
The question for me is, whether the advances in AI can shrink the needed training dataset further and therefore make it way cheaper / easier for smaller teams to create new models. At the core the question is: Does it make the most sense to have all the training data and then derive one or a small number of models from that, which then can be used by every industry / company. OR does it make more sense to have these models being trained on smaller more specific data which then would lead to every vertical having their own models. The latter case could mean that there will be a lot of markets and the companies succeeding there not only can train and build the models but more importantly have the industry knowledge and integration to build the “right” models.
Still fascinated by this and hope it goes mainstream soon. It seems like a perfect win-win-win with the only caveat that the consumer must be willing to wait for a period of time (months).
The idea is that the consumer doesn’t straight away buys a product but indicates that she’s willing to purchase it at this price in the future. And therefore will start a savings plan. This leads to better forecasting of future sales for the merchant / producer, it also makes some financial engineering possible on the merchant or payment processor level (as in using the saved cash or giving some percentage back to the consumer). Think of it as facturing but the other way round!
Especially with things like marriages or getting a child where lots of big item purchases might occur towards a specific date, this Save Now Buy Later system can ease the cash flow on the consumer side but also allowing her to be done with the purchase and cross it of the list.
For me specific we need to buy a need child safety seat in May next year, I already know which one, but it doesn’t make sense to buy it now and let it sit here (apart from inflation angst). If someone gives me 2-5 % discount to save for it now and receive it in May, I’d be willing to commit now.
A more in-dept view of how the cloud computing market was created and evolved. With the goal of drawing conclusions to how AI might develop. (very similar to the first link)
Always funny to read something like this.
Two things confused me about AI: First is that it seems to be more democratized as previously thought, as open source AI models like stable diffusion emerge, also the cost associated with creating these models is continuing to get cheaper. And on the other hand, I assumed that structured jobs like programming would be easier to do for an AI, but it turns out that because the task is very structured errors can be easily spotted. It turns out that current AI models have a pretty high failure rate. On the other hand, where these “mistakes” are not that important is in all the creative tasks, from writing to painting to soon designing videos or music. These mistakes are viewed by the human as creative freedom, which a artist would also take.
In particular, that creative endeavors across the board — whether visual, textual, or musical — are likely to be disrupted by AI long before systems building.
In addition to the correctness argument we use above, it also may be the case that combining and recombining all prior art may be sufficient for the practical range of creative outputs. The music and film industries, for example, have historically produced countless knock-offs of popular albums and movies. It’s entirely conceivable that generative models could help automate those functions over time. However, the remarkable thing about so many of the images produced by Stable Diffusion and DALL-E 2 is that they’re really good and genuinely interesting. It’s not difficult to envision an AI model producing genuinely interesting fusions of musical styles or even “writing” feature-length movies that are intriguing in how they tie together concepts and styles.
Another Generative AI article predicting the usage in games.
The opportunities are massiv as the market is very big (lots of creators, so a lot of place for tools, forms of expression, reducing time to produce and distribute content) and most are under-monetized (getting paid, but also financing, enabling other income streams like creating merch or products). The challenges are that it is difficult to reach the creators and they behave more like consumers than SMBs, also all creators want to have a direct relationship with their audience so any tool that gets in between this will have adoption problems.
particularly across creativity tools that leverage new technology, monetisation platforms and tools that look to capitalise on the most popular creators, the use of creator-specific data to expand access to financing, and companies that support the development of creator-led D2C brands . Products that ultimately help creators to (1) save time and (2) make more money — both in obvious ways — will also thrive in the creator economy and will help to mitigate against the challenges of creator acquisition.
Also there is always the “graduation problem” if a creator is getting big of your platform, she might choose to build most of the tools you’re offering herself, as she might need to customize them more and also doesn’t need to pay you for features she’s not using.
Quick and nice overview / comparison on the current startup climate