825 karma · joined August 11, 2017
I.e. 10 LOC -> 100 LOC -> 1000 LOC -> 10K LOC -> 100K LOC.
I failed to see AI moving behind 1000 LOC. I.e. 10*100LOC << 1000LOC.
Also, the AI is statistical, it does not have any notion of the real world. I.e. he might know how to write a function, but it does not know why you need the function.
Can the machine translate requirements to code? Can the machine actively maintain 100K / 1M / 10M code base? Can the machine even understand multithreaded code?
Let say that I want a new feature, and the machine cannot do it, what now? Are you going to reverse engineer your new code base by human? What happen if human introduce a bug during a bug fix? Who will fix it? The machine?
Are you a programmer?
Nobody get replaced by GPT3. Its like stock photo site.
Did commercial photography got affected by stock photos'? Did it replaced anybody?
I am not sure what this AI system can do in the real economy? Have you seen those deployed?
The only thing that I see affected is google.
The front end is less of a concern.
But in general the site is huge (at least 500K machines).
Also, 90% of the structured data is unlabeled. Hence, your calculation should be for "labeled" unstructured data, is this 90%?. I would argue that outside big tech, this is 0.1%.
You competition is not Databricks. Databricks main use case is tabular data (both in the delta lake and in ML). I.e. Databricks compete with snowflake. It tries to be a database. I.e. it tries to get out of the data lake.
I think that your competition is with S3 and R2 from the storage side, and with transformer based models (Hugging face). Correct me if I am wrong, but the whole idea with transformers models is that the training was already done , and you can use small amount of domain specific data? I.e. you do not need a lot of storage?
The problem with all those innovations, is that they take time.
In addition, he got a reputation to keep.
Most of the deep learning applications have little operational use cases.
It is hard to invent a new deep learning architecture without a PHD.
It is NOT hard to train a deep learning model.
The AI value chain is moving into data driven vs model driven. I.e. the modeling part is solved, or is being automated.
I.e. when commercial companies can no longer sustain open source projects.
In general I would buy patent insurance.
In general before you do any sale you need leads. I.e. you problem is not selling, but prospecting. I.e. you need prospects at the top of the funnel.
There is very good book - "fanatical prospecting" :
https://www.amazon.com/s?k=fanatical+prospecting
So your first step is to raise awareness of your product.
To do that there are two way - inbound (blogs,etc) and outbound(cold call / cold email). For inbound, you need to provide value to customers which are not related to your product. I.e. first earn trust and give value.
For outbound, the key is lead filtering, I.e. looks for signals that would qualify your leads (e.g. sector, traffic). But in general outbound is a numbers game.
This is why there is a lot of companies that did (2), but we will not have self driving car in the next 10 years.
Also, I am not sure why a future robot will have only 2 legs and 2 hands ?
Isn't the whole point is the be productive, why not 4 legs and 4 hands?
I am sure that if you run RL experiments in a human house, you would find much more efficient design (I.e. the human design is based on human living in nature).
This was a POC ? Not even sure what was it. Kind of a demo robot with some labs videos.
The real question is what happen to Tesla after Toyota comes with cheap EV.
Gen Z are much smaller than Boomers. This is a structural issue that can be fixed only via immigration.
What could have been automated probably was automated.
Also, due to china failures, there is going to be boom of insourcing not outsourcing.
The reason is that you do not want to start over each month sales wise.
Again, as a consumer you can "consume" what is offered in the market.
I.e. if you have an option to not buy, the seller has the option to maximize its profit.
The best way is to become a B2B creator in linked in.
As a creator you should be ready to add value to your network in any post. It is a lot of free work, but the main goal is to gain trust.
Once you have the trust, you can start asking the network to be customers, etc.
You should use machine learning. I think that those are ranking algorithm. Basically create a training set of known examples that you have, and try to predict the same behavior.
I do not think that this can be done via rule based approach.
For example, go to as many chess tournaments as you can.