Superhuman capability within the tasks that are well represented in the dataset, yes.
If one takes the view that intelligence is the ability to solve novel problems (ref F. Chollet), then the amount of data alone might not take us to superhuman intelligence. At least without new breakthroughs in the construction of models or systems.
R1 managed to replicate a model on the level of one one they had access to. But as far as I know they did not improve on its predictive performance? They did improve in inference time, but that is another thing. The ability to replicate a model is well demonstrated and quite common practice for some years already, see teacher-student distillation.
If OpenAI would publish the models before the competition, then one could verify that they were not tinkered with.
Assuming that there exists a way for them to prove that a model is the same, at least. Since the weights are not open, the most basic approach is void.
It would leave a lots of general purpose GPU-based compute. That is useful and enduring infrastructure? These things are used for many scientific and engineering problems - including medicine, climate modeling, material science, neuroscience, etc
A) We are out of the Internet-scale-for-free data. Of course the companies deploying LLM based systems at massive scale are of course ingesting a lot of human data from their users, that they are seeking to use to further improve their models.
B) Has learning though "self-play" (like with AlphaZero etc) been demonstrated working for improving LLMs? What is the latest key research on this?
What exactly do you mean by replace? Replace in particular roles (causing the humans to shift into new roles), or replace as in nothing useful left for humans to do? Those are two very different trajectories.
So far in the industrial revolution we have been experiencing the first. Waves of automation that displaces workers in one role, and then new roles open up for humans (often involving the machines). I believe that for the foreseeable future human + AI tools (symbiosis) will be a much stronger than AI alone. Of course it will tend towards more and more AI per human over time. The same way modern manufacturing is "machine tending", where a few workers might supervise a factory that outputs what would have taken thousands of workers in the past.
If the thing being produced is has unrealized demand, then we would expect that being able to do it more effectively would just mean more production, possibly to the extent that the same or even more humans are needed to produce it.
So the real questions are - which things does humanity have a lot more demand for (2-1000x) - and can be made more effective with AI in the loop / on the team.
And the flip side, which things are we near max demand for, and can also be made more effective with AI. Jobs in those areas are going to be decimated, move away as quickly as possible.
What would the "horse" equivalents be in this AI tech shift? Legit question - getting reasonable answers here might help us understand where we are going.
For sure it is not computers. Or software systems. We will for sure have way more of those. "Code" is more unclear, there will for sure be more of it - but for sure less human-hours per line of code. So the roles there will change quite a bit, though unclear exactly how. More focus on systems, business rules, specification, I suspect.
Translations (between human languages) seems like a thing that is highly risky. It seems that it could go the way of the horse, automated translations being 99% of the volume eventually?
Ahahaha, that is legit a nice framing and "business model" for a Patreon in the AI space. Drinking from the firehose and washpanning it, so that not everyone has to do the same.
It is pretty easy and convenient to write extensions using pybind11, including passing numpy arrays. It takes 10 lines in setup.py and around 10 lines in a .cpp file, run setup.py build_ext to build it. Not quite the convenience of inline - but in practice pretty nice.
My only nit in that compile time is around 3 seconds on my machine.
The Internet was also very expensive in its infancy. Dialup charged by the minute for mere kilobytes. The cost per MB dropped by a factor 1000x over the course of 30 years. It took billions in investments, and millions of people working on it to make it happen.
Give LLLms a couple of decades, and the price for a given capability will have increased by 1-4 orders of magnitude.
Lua is a great language, including for embedded. And I am sure that their Lua-based product is good also. But this article was not very convincing as to why it "beats MicroPython". The criticisms are vague or not correct.
Extending MicroPython in C is quite easy, and one can implement third party modules in the same way all the official modules are made. There is build variable to set for including that into your custom firmware build - nothing particularly tricky about it.
The article goes on to lament that popular libraries from Python world are not available, including numpy. But actually there is a very good reimplementation of numpy (and parts of scipy), called ulab.
No, just saying that a claim of qualia would require some sort of evidence or methodical argument.
And that LLM outputs professing feelings or other state-of-mind like things should by default be assumed to be explained by that the training process (perhaps inadvertently) optimized for such output. Only if such an explanation fails, and another explanation is materially better, should it be considered seriously.
Do we have such candidates today?
Nothing an LLM says can in itself, right now, be used as evidence of what they 'feel'. It is not established that there is any linking of their output to anything else than the training process (data, loss function, optimizer, etc.). And definitely not to qualia.
On the other hand, it is well know that we can (and commonly do) make them come up with any output we choose. And that their general tendency is to regurgitate any kind of sequence that occurs sufficiently often in the training data.
Big party for Altman et al these days, who have been advocating (lobbying) a lot for this.
The US, and the world at large needs more electricity, and nuclear probably does have it's place in the mix. Just hope we do not spend all the new capacity on "AI". And most importantly, that nuclear safety continues to be highly prioritized every step of the way.
Did 6 years as bicycle mechanic/sales in highschool, was ok, could do again. Electronics engineering - presumably in this scenario the software aspects might be reduced, but probably longer before LLMs can do board design, debugging, type approvals etc.
How good are current LLMs at translating problems given as text into something SMT solvers can operate on? Be it MiniZinc, Z3, Smtlib, Python bindings, etc. Anyone tried it out?
There is no LLM suffering today. There is no evidence that there will be such suffering this year. I have seen no credible claims that it is probable to exist this decade, or ever, for that matter. This is not an issue we need to prioritize now.
Just a quick prompt for reflection. I do not suggest anything how people actually live their lives. But many have said that they got caught up in being too busy to actually enjoy life. Don't fall into that by accident!
They might (or might not). Extraterrestrial beings might also need moral standing. It is ok to spend a bit of thought on that possibility. But it is a bad argument for spending a non-trivial amount of resources that could be used to reduce human or animal suffering.
We are not even good at ensuring the rights of people in each country, and frankly downright horrible for denying other humans from across some "border" similar rights.
The current levels of exploitation of humans and animal are however very profitable (to some/many). It is very useful for those that profit from the status quo, that people are instead discussing, worrying and advocating for the rights of a hypothetical future being. Instead of doing something about the injustices that are here today.
I do not think that those researchers are fungible. We could however allocate a few hundred million less to AI research, and more to fighting human exploitation. We could pass stronger worker protection and have the big corporations pay for it - which then they have less money to spent on investments (in AI). Heck we could tax AI investments or usage directly, and spend it on worker rights or other cases of human abuse.
One month at 20 USD seems like it should be plenty to try it out on a small project or two to decide wether it is worth trying 100 bucks/month?
Or one can just wait a couple of months as people report their learnings.
The actual harms being done today are still more pressing than the hypothetical harms of future. And should be prioritized in terms of resources spent.
We are slave masters today. Billions of animals are livestock - they are born, sustained, and killed by our will - so that we can feed on their flesh, milk and other useful byproduct of their life. There is ample evidence that they have "a form of consciousness". They did not consent to this.
Are LLMs worthy of a higher standard? If so, why? Is it hypocritical to give them what we deny animals?
In case anyone cares: No, I am neither vegan nor vegetarian. I still think we do treat animals very badly. And it is a moral good to not use/abuse them.
Does you Google actually respect the keywords? For me, most of the times it replaces words with "synonyms" (mostly wrong context or not really replaceable). And results are pretty crap as a result - no what I was looking for, but just much more common/generic stuff.
pybind11 is your friend. Focus on small self-contained functions first. For numerical functions you can then take it mostly our of a book. See if you can speed up some simple and common operation within your problem domain of interest.