Worth noting that while AlphaGo and AlphaZero are incredible achievements, the amount of actual code to implement them isn't very much. If you have the research paper, someone in the field could reimplement them in a few days. Then there is the large compute cost for training them to produce the trained weights. So, opensourcing these bits of work without the weights isn't as major a thing as you might imagine.
If only there existed a distributed way to incentivize calculation of AI weightings while also providing a currency to encourage scale…
Please do the same for Alphastar, the starcraft AI, that would be great.
This would probably be against some Blizzard ToS
I thought that they had an open API, and I also thought we as players and AI enthusiasts would benefit somehow from this venture, it was all rather exciting. But the research was all behind closed doors, and nothing much came of it except for some impressive demonstrations.
There's an interesting bias I've noticed on HN where a lot of people still believe AlphaZero is the state of the art in computer chess, when that hasn't been true except for a short while after the release. AZ still gets posted and upvoted today while newer improvements in other engines are discussed far less. I guess it's because AZ came out of Google?
AZ revolutionized chess engine architecture and then moved on to other fields. Fast followers who incrementally improve upon the breakthroughs are very rarely recognized. E.g. inventor of the blue LED versus those who improve the efficiency by .1%
Stockfish' architecture is in no way revolutionised by AZ. It has gone its own way.
I thought that they had an open API, and I also thought we as players and AI enthusiasts would benefit somehow from this venture, it was all rather exciting. But the research was all behind closed doors, and nothing much came of it except for some impressive demonstrations.
In case you didn't know, SC2 does have a healthy AI bot community:
I feel like this headline is deceptive. This is a Jax reimplementation, and it was released a year ago. It is a cool library though. The basic operation of muzero is very simple, but training it efficiently is tricky.
I thought that they had an open API, and I also thought we as players and AI enthusiasts would benefit somehow from this venture, it was all rather exciting. But the research was all behind closed doors, and nothing much came of it except for some impressive demonstrations.
In case you didn't know, SC2 does have a healthy AI bot community: https://sc2ai.net/
I am aware of that and follow it but its way behind what Alphastar was doing.
The human in the loop reinforcement learning paper that powered chatgpt's training arose from deepmind's experiments with boardgames (and games). AGI is still an unsolved problem and deep RL that arose from the success of deepmind's experiments wth boardgames and games so far, will likely play a huge part in it
Like Bell Labs of old, Google has excellent AI researchers but they haven't built many (any?) AI products people can use.
Search? Gmail completing your emails? They have lots of AI products, they just wait until they are more solid before releasing stuff. Now they will rush out a competitor to ChatGPT before it is ready, but they would use it for something given time.
These particular programs were bleeding edge when they were announced, but have been recreated and improved on outside of DeepMind since then. For example, LCZero today is stronger at chess than any of the DeepMind chess programs that were shown to the public. Of course we don't know if they got further behind the scenes. Of course LCZero and everything else relied on the published papers of DeepMind. That is, the pa…
There's an interesting bias I've noticed on HN where a lot of people still believe AlphaZero is the state of the art in computer chess, when that hasn't been true except for a short while after the release. AZ still gets posted and upvoted today while newer improvements in other engines are discussed far less. I guess it's because AZ came out of Google?
Well, AZ could perform above Stockfish by using much simpler principles. A lot of tinkering was required to get Stockfish to where it was, and AZ pretty much rediscovered and improved on some of these refined fine-tuned principles, such as board evaluation, opening books or endgames, from its own training. I think that's very impressive.
There's an interesting bias I've noticed on HN where a lot of people still believe AlphaZero is the state of the art in computer chess, when that hasn't been true except for a short while after the release. AZ still gets posted and upvoted today while newer improvements in other engines are discussed far less. I guess it's because AZ came out of Google?
AZ revolutionized chess engine architecture and then moved on to other fields. Fast followers who incrementally improve upon the breakthroughs are very rarely recognized. E.g. inventor of the blue LED versus those who improve the efficiency by .1%
It changed engine analysis of games because it doesn’t generate a series of “lines” that each has a centipawn value. It just gives you the move.
Subjectively, the Monte Carlo moves seem so human-like in comparison to minimax. Minimax can suggest a move that no human would play because the depth of calculation at which that move is good is just impossible for people.