The title is clickbait. He more ends up saying that AGI is practically impossible today, given all our current paradigms of how we build computers, algorithms, and neural networks. There's an exponential explosion in how much computation time it requires to match the out-of-frame leaps and bounds that a human brain can make with just a few watts of power, and researchers have no clever ideas yet for emulating that trait.
I knew they were about to launch a new model, by how aggressively they were advertising the old stock at sale prices the last few weeks.
I won't be buying it though - I tried talking to Meta AI in voice mode from my phone, and it's response to anything STEM related is basically "tee hee I'm just an AI I cant do that". My current assignee to my phone's AI hardware button is microsoft's. I assume it's an OpenAI model, but it lets you speed their voice up, which I value greatly.
I once lost years of Gmail, Drive, Photos, YouTube, etc., etc., because I posted a meme to their Google Plus social media site. It was a JPEG screenshot of a credit card form, and it said, "This post is only viewable by Google Plus Gold members; sign up now." It just took one click from some apathetic call center moderator, who looked at the post for 0.5 seconds total, to get everything connected with the company insta-wiped. I bought a Synology NAS since then, and now I treat my relationship with any company as very fragile.
Imagine a horrific world where most articles are 5 pages of LLM generated fluff from a 10 bullet point sentences prompt. I know, truly a nightmare world. But in this world, wouldn't it be great to just collapse all those articles down to their original prompt? A sort of reverse-fluffing. So maybe I can't convince you of how great it is in our world, but hopefully at least in this hypothetical world it makes sense.
I will never use SDCards to run any 24/7 operating system again, after my Home Assistant for Rasperry Pi became unusable twice. The hours of troubleshooting and getting it back online was not worth the $200 an NUC/mini with actual SSD cost. SD cards are incredibly slow too.
I love Cerebrus. 10-100x faster than the other options. I really wish the other companies realized that some of us prefer our computer be instant. I use their API (with Qwen3 reasoning model) for ~99% of my questions, and the whole answer finishes in under 0.1 seconds. Keeps me in a flow state. Latency is jarring. Especially the 5-10 seconds most AIs take these days, where it's just enough to make switching tasks not worth it. You just have to sit there in statis. If I'm willing to accept any latency, might as well make it a couple minutes in the background, and use a full agent mode or deep research AI at that point. Otherwise I want instant.
They don't mention that Jony walks from Pac Heights, and instead of going directly to North Beach, he must have climbed to the top of Coit Tower's hill to access that particular staircase down to Jackson Square, then has to walk the long way around Telegraph Hill right past the cafe to climb uphill into Chinatown for another walking shot, then backtracks to the cafe. Another small irony is Jony was heading toward Lombard at the start, so if he walked a straight line to Coit, he would have walked right past Sam's driveway.
I read it as: we have millions of hard drives and flash drives with a dead controller chip, so we harvest their other parts as spares. We still know how to make the spare parts from scratch, but we have so many for free.
I wish it was unlimited for me. I got 500 fast requests, about 500 slow requests, then at some point it started some kind of exponential backoff, and became unbearably slow. 60+ second hangs with every prompt, at least, sometimes 5 minutes. I used that period to try out windsurf, vscode copilot, etc and found they weren't as good. Finally the month refreshed and I'm back to fast requests. I'm hoping they get the capacity to actually become usably unlimited.
Oooh, that makes more sense. To be honest, I only started reading the article with the sole mission of answering the question: what the heck is a non-generative LLM? I had never heard of that. I thought it would be answered explicitly in the first few sentences. As sentence after sentence passed and it didn't even seem to acknowledge the tension of that elephant in the room, my frustration grew beyond what I could concentrate on, and I found it difficult to continue reading, assuming I'd been clickbaited by slop or something. But now that I have that context it makes it a lot more readable.
My uncle had one of these in the 90s. All I knew was he was a higher up at the university in his smallish country, and ran their internet stuff. It confused the heck out of me when he verbally told me to bring his website up at Thanksgiving dinner, and after I typed a dot, he said "no no, no dot. just enter" And it worked. Baffled me as a kid. Nice to finally have some explaination for that fever dream of a memory.
I did the same. Sync'd my Obsidian .md vault between my macbook and my server. Asked AI for an index.php that would render a nice html/css webpage at domain.com/whatever for every whatever.md on the fly. Works fantastic. Supports everything Obsidian markdown does - pics, videos, resizing via drag, wikilinking, codeblocks, etc… No publish step - from my site you can literally see me typing live in Obsidian. No more friction to share things.
Fine with me. If 100% of my RAM isn't in use at all times for low priority speculative cache, then it's not doing what I want. So long as it frees up the RAM instantly the moment anything actually requests it.
If you're the 1% of Earth's population for which this is true, then this headline makes sense. If you're the 99% for which this isn't at all true, then don't bother reading this, because AI coding assistance will change your life.
Disney is quite famous for its strict allergen handling, to the point where people with deadly allergies who don't trust eating out anywhere else make an exception for Disney restaurants. So, were it not for this mass murder attempt, he probably would have gotten off a lot more lightly for some mischief charges.
That's what they would have said about drafting tools in the 1880s, command line AutoCAD in the 1980s. Try to imagine the 2080s, or the most visionary sci-fi movie about that future - one of those movies where people look back and says "wow they were close!". Suppose a character wanted to get an idea for a three-dimensional form from their brain onto some medium. They're not going to be clicking a plane, clicking sketch, clicking some lines, tapping the dimension hotkey, twisting their spacemouse a bit with their left hand, clicking extrude, dragging, tapping into their numpad the height, etc etc. We use the clunky interface of our age.
I think part of that is an illusion, since for something bowing upwards, the usualy anchor point of top left seems rotated clockwise.
But there is still a real rotation - look at wyoming or colorado for a perfect rectangle. My guess is the div element isn't quite centered - perhaps too much padding on the right edge, causing the center point to be off to the right. So when it bows you get the rotation bias
I get that CAD interfaces are terrible - but if I imagine the technological utopia of the future - using the english language as the interface sounds terrible no matter how well you do it. Unless you are paraplegic and speaking is your only means of manipulating the world.
I much prefer the direction of sculpting with my hands in VR, pulling the dimensions out with a pinch, snapping things parellel with my fine motor control. Or sketching on an iPad, just dragging a sketch to extrude is to it's normal, etc etc. These UIs could be vastly improved.
I get that LLMs are amazing lately, but perhaps keep them somewhere under the hood where I never need to speak to them. My hands are bored and capable of a very high bandwidth of precise communication.
I really dislike credits based pricing. It just doesn't feel good. Cursor $20/mo plan lasts me about a week before it becomes unusably slow. I'm fine with slow requests, but recently it will literally not even begin a response for over a minute. Really hoping costs come down or someone figures out a real unlimited plan for these agent coders.