I think the biggest problem is jagged intelligence which I would bet would be solved largely with bigger models, which comes back to rapidly evolving hardware paradigms.

So many jobs as such may be limited to humans with high bandwidth BCI connected to AI, that effectively gives them an AI exobrain. But even then the AI would need to do most of the work or the human would just slow things down.

Maybe they become experts in wiring up AI and robotics?

But 5 years out, we might be into a new compute-in-memory paradigm. And we may have exceeded the complexity of the human brain with models greater than 100 trillion parameters.

We already see performance that is superhuman in task specific ways with 1-10 trillion parameters models.

In under 5 years, most small and many medium business will be handled directly by AI agents. You won't even necessarily know what kind of software they are running. For small-medium business, it may have very little or no code and just be a semi-structured database that is neutrally rendered into any needed form on-the-fly.

There need to be thorough weekly video walkthroughs of all of the detention centers. Otherwise you can expect actual starvation at some point.

I think there's a reason that University of Michigan researcher jumped after that ferroelectric breakthrough.

I am hopeful that some of this massive pressure is going to push us into the next paradigm, which is likely some kind of "true" compute-in-memory system. That may increase performance and efficiency for AI by up to 100 X.

Also, even though I feel AI and robotics are very important for progressing humanity, I think that much of the world has long since lost a proper sense of intrinsic human value. It's really gone from overt exploitation to slightly more mild exploitation where we pretend the system is really merit based.

And as AI and robotics remove the need for human labor, I hope that someone like the pope can convince people that we should value human beings inherently and more fairly. Inexpensive labor and intelligence should make this feasible.

I hope the speech isn't something dumb like "remember only humans have souls" because I think that's really premature and pretty obvious that AIs are not people at this point.

The really convincing and somewhat deeper simulations of humans are probably only a few years down the line though.

Which comes back to the Rovelli dualism article that was on the front page before. I think we should not be in a hurry to try to duplicate humans in depth (such as imitating emotions, pain, stream of consciousness, self-preservation, etc). It's just completely unnecessary to go that far to get useful AI, and obviously unethical to subject a real human emulation to slavery.

Right now I feel like a 4bit Qwen 3.6 27B with MTP is one of the best for agentic tool calling for some smart voice agents in an H200. I wonder if DS4 Flash being using 80b at 2 bit with 13b active and MTP could be even faster and smarter and allow more concurrent sequences?

This special 2bit quantization seems like a big deal.

But it's really interesting to me that that may be possible now. You can include a fine-tuned model that understands how to use your program.

E.g. `> toolcli what can you do` runs `toolcli --help summary`, `toolcli add tom to teamfutz group` = `toolcli --gadd teamfutz tom`

I think the primary problem is actually more than just Bambu's behavior, it's that China is an authoritarian country, and most of the population not only accepts the idea of central servers monitoring and "moderating" behavior but largely may embrace it as a sensible thing to do. It's probably beyond Stockholm Syndrome to the point of much of the culture genuinely not completely even understanding the idea of why privacy and personal control is important.

Much of the United States is so far on the other side that they can't begin to understand the position Bambu is in. Large companies in that country just do not have the option to allow their users to bypass censorship and monitoring.

I do think it's actually great that this type of issue gets in everyone's face though and it's great people are fighting back. But realize that the problem is deeper than one company. It's the whole type of government and attitude towards it and technology.

I will die on this hill: AI _properly_ integrated into education will be a huge improvement for students because it will enable each student to have personalized instruction and tutoring.

That's the only statement that's true. Admiting to AI use is unfashionable in the western world at this time.

But how much would you like to bet that 90% of those students who were booing also used AI to do their homework for them quite often? So your take away would be "the AI stole their education". No, they were dishonest and the AI helped them cheat themselves out of learning.

Technology doesn't make anything banal or a hellscape, or fire people. Technology is a lever.

If humans use AI to produce worse output because they are too lazy to bother reviewing and iterating on it, that is a human problem. If humans are going to use AI to help them exploit other humans more efficiently, that is also caused by the human rather than the technology.

Also, the ChatGPT moment for humanoid robots is coming this year or next. It will become very obvious that AI use in these robots is not just superficially plausible text.

AI may default to mediocre and often somewhat buggy code unless you iterate because that is just what the vast majority of human written code that it has seen looks like. But the fact that he got away with not reviewing the code for so long to me proves the opposite of his conclusion.

1690 lines of code in one file is a walk in the park for SOTA models.

He can just say something like:

"Please review and create a refactoring plan and test suite. I found atrocious architectural decisions like numerous special cases and if statements rather than using abstractions properly. Make a few notes in comments and architecture.md to never do this again."

One could also argue that it was a better decision each time by the AI to just never do a refactor unless prompted because that increases the likelihood of something breaking and you want to do that after you verify the minimum code change actually functionally does what you want.

Also I bet you the headline is a lie. He basically admits it by saying he is writing the core structure of the next version by hand ahead of time, implying that he will generate the rest. So the title is a half-truth at best.