Talk politics.

In 2023, Meta released its AI code as open source. For those not familiar with it, "open source" means that the code is publicly available. DeepSeek has also released its code as open source.

Another factor is that the US bans exports of certain high-powered and AI-specific chips. This forced DeepSeek's developers to optimize the code to run on slower / standard hardware.

Meta sees this as a win. Why? Because releasing their code as open source allowed an innovator to optimize it in ways that they didn't. Meta's chief AI scientist (Yann LeCun), also says that thinking about this as "China beating the US" are looking at it the wrong way. It's about open source working better than closed source.

Now. Here's a question. In short:
1. Deepseek gets released to GitHub.
2. A developer in Venezuela forks it to add Taiwan responses back in.
3. Another developer in Iceland forks that and adds some data for a rare chemical process that they need to gather responses for to do their masters degree.
4. Another developer in Zimbabwe forks that and so on.
5. Any of these forks are trivial to download and run on a laptop with an rtx4060 and that happens all over the globe.
QuestioN: Is it still DeepSeek?

IBM pauses hiring for 7,800 jobs because they could be performed by AI

Except, that's not AI. The media are morons. Robotics and Expert Systems is not AI. AI is the ability for the machine/program to do more than what it was programmed to do. If a machine/program is only capable of doing what it was programmed to do, then it cannot be AI. We are not even remotely close to developing AI. Maybe when quantum computers start appearing on desktops or as home computers, but not before then. Current binary computers are not capable of creating AI.

If you thought the neural networks and machine learning stuff has come a long way, well, it hasn't. Not really. Let's not forget that LISP was developed in 1956 specifically for the purpose of developing AI. In 1964 they developed ELIZA at MIT, the very first "chatbot." Artifical neural networks predate computers and have been around since the 1940s. A few expert systems were developed in the 1970s, but by the late 1970s and early 1980s there was no work being done with AI at all. Then from the late 1980s through to the mid-1990s there was a flurry of activity. Neural networks became deeper (with more stacked layers), and back-propagation became slower due to the vanishing gradients problem. Then again from the mid-1990s until the mid-2000s there was nothing at all. Then by 2010 ImageNet sponsored a competition for using AI for visual recognition. The winner of the competition had a 15.3% error rate.

So AI has not "come a long way." In fact, it has barely progressed at all from the 1960s. It wasn't AI then, and it is not AI now.

I find this AI nonsense to be massively over-hyped and seriously misunderstood. When people see a cleverly written program they instantly think AI, when they should be thinking that REAL intelligence was used in the application's development and give the developer the credit they deserve instead of crediting the computer that is only following their programmed instruction.