> By making these fake images ubiquitous we are forcing people to quickly learn
That's quite the high opinion on the self-improvement ability of your Average Joe. This kind of behavior only comes with an awareness, previously learned, and an alertness of mind. You need the population at large to be able to do this. How if not, say, teaching this at schools and waiting for the next generation to reach adulthood, would you expect this to happen?
I agree that improvement for the Average Joe will be very hard. I also think that taking more attention to teach the younger generation is vitally important. But mostly I don't see an alternative. I don't think we can protect people from fake information without giving up our freedom, and that isn't a viable alternative in my mind. So what is left but trying our hardest to teach people to think critically?
Our institutions have been trying to get our kids to think critically for a while. At least when I was in school, we didn't focus a lot on memorization (sometimes we did, like memorizing the times tables or periodic table). My teachers tried to instill in us an understanding of the concepts, something I took for granted. Many of my classmates have gone on to become lawyers, doctors, other prestigious careers.
But I feel like we live in a different time now. I hear teachers tell stories about school admin siding with parents instead of teachers, and the kids aren't learning anything. Anecdotally of course.
I think our teachers really want the kids to think critically. But parents and schools don't seem to value that anymore.
Is any of this intrinsically a strength of Nano Banana, while not of other models/generative tools? Have you tried doing the same with say Klein, ZIT, etc.?
Nano Banana was the first generative image tool that seemed to understand architecture. I have some 18th century engravings that I've been trying to get AI tools to visualize as though they were photographs of real structures, and the results were comically bad until Nano Banana came along. I remember seeing the announcement, dropping a sketch into Gemini with the prompt "Make this a photo" and it preserved all the features and dimensions, but added photorealistic textures.
I still do a lot of modeling of historical buildings in Sketchup and Twinmotion, but Nano Banana has really helped me with more easily visualizing a world before photography.
I've read opinions in the same vein of what you said, except painting this as a good outcome. The gist of the argument is why spend time looking for the right tool and effort learning its uses when you can tell an agent to work out the "problem" for you and spit out a tailored solution.
It's about being oblivious, I suppose. Not too different to claiming there will be no need to write new fiction when an LLM will write the work you want to read by request.
It's a reasonable question - I would probably answer, having shipped some of these naive solutions before, that you'll find out later it doesn't do entirely what you wished, is very difficult/impossible to maintain, has severe flaws you're unable to be aware of because you lack the domain expertise, or the worst in my opinion, becomes completely unable to adapt to new features you need with it, where as the more mature solutions most likely already had spent considerable amount of time thinking about these things.
I was dabbling in consulting infrastructure for a bit, often prospects would come to me with stuff like this "well I'll just have AI do it" and my response has been "ok, do that, but do keep me in mind if that becomes very difficult a year or two down the road." I haven't yet followed up with any of them to see how they are doing, but some of the ideas I heard were just absolute insanity to me.
Hey, I am actually working on making this compatible on earlier AMD's as well because I have an old gaming laptop with an RX5700m which is GFX10. I'm reading up on the ISA documentation to see where the differences are, and I'll have to adjust some binary encoding to get it to work.
I mean this with respect to the other person though please don't vibe code this if you want to contribute or keep the compiler for yourself. This isn't because I'm against using AI assistance when it makes sense it's because LLMs will really fail in this space. Theres's things in the specs you won't find until you try it and LLMs find it really hard to get things right when literal bits matter.
I really like the minimal approach you've taken here - it's refreshing to see this built completely from the ground up and it's clearly readable and for me, very educational.
But help me understand something. BarraCuda does its own codegen and therefore has to implement its own optimisation layer? It's increbibly impressive to get "working" binaries, but will it ever become a "viable" alternative to nvidia's CUDA if it has to re-invent decades of optimisation techniques? Is there a performance comparison between the binaries produced by this compiler and the nvidia one? Is this something you working on as an interesting technical project to learn from and prove that this "can be done"? Or are you trying to create something that can make CUDA a realistic option on AMD GPUs?
Rome wasn't built in a day. I'll get there with optimisations im just going for "correctness" first. I've had some amazing resources be sent from me from academics around the world so once I get this to a "point" I'll begin optimising it.
If it helps: I'm actively considering how to make LLMs more helpful in GPU programming. Recently I converted the RDNA 3.5 ISA manual from PDF to Markdown for LLM retrieval (because I have a Strix Halo). Notably, I converted every bitfield diagram to plain text table and manually checked its correctness. Everyone can comment there if they find a typo. I guess someone can also do it for RDNA 2 and other manuals.
Don't let anyone dissuade you, it's going to be annoying but it can be done. When diffusion was new and rocm was still a mess I was manually patching a lot to get a vii, 1030, then 1200 working well enough.
It's a LOT less bad than it used to be, amd deserves serious credit. Codex should be able to crush it once you get the env going
Imagine if Tolkien was writing Fellowship last decade, and the book landed on your hands today. No decades of cult growing, no adaptations or explosive marketing, some word of mouth. Would you think it "much important" before reading it? What makes the importance?
In my opinion it's the prose. It's always the prose. Always gotta be on the lookout for good writers, new and old.
> despite everyone saying for 10+ years that they are going to die.
What many people have been saying in my experience is pretty much the opposite: that Mozilla isn't going anywhere because Google wants them (needs them) to be around. That it's their antitrust Trojan horse.
Does it maybe come down to changing licenses, as in a license expires and another is negotiated with different terms (to charge per household instead in the example above)?
That's quite the high opinion on the self-improvement ability of your Average Joe. This kind of behavior only comes with an awareness, previously learned, and an alertness of mind. You need the population at large to be able to do this. How if not, say, teaching this at schools and waiting for the next generation to reach adulthood, would you expect this to happen?
reply