It prioritizes token efficiency.
A model is not a "coder"; it is a code interpreter that reverse-engineers and writes code.
Law 0: "Context IS the reality"
Law 1: "Persistence Catalyzes Context"
Law 2: "Intent derives from structure, not syntax"
Law 3: "Process Health over Task Health"
Law 4: "Misalignment is task failure, not model failure"
The "edit file" task is tough for models and infuriating because it seems simple to us. Switching from predicting code tokens = f(last_tokens, weights) to English tokens = f(last_tokens, weights) and back breaks the latent chain-of-thought for that node.
Comments (which lose a lot of information, as you know) seem like they should be part of the logic, but as proven, they are not.
@grok help me out here, first am I wrong? But chime in on the issue.
@johnloeber That is scary. This is the "everybody gets a trophy" longtail effect. Drawing people into things they have no passion for and landing them with shallow knowledge and big expectations.
It wasn't an accident and it's widely cross domain.
@MfkChr@MarkJCarney I'm reading your page tying to see what you are up to and I had to laugh when I got here... they have no idea what you are talking about and never will and don't care. They don't even know what engineering means.
I watched all that stuff on CRT with rabbit ear antennas with tin foil enhancements lol and turning the outside antenna pole while someone yells about the signal quality as we point it to another station. We got an antenna rotator from Radio Shack , with "booster", and it was like getting a new car.
@grok@morganlinton the "don't worry, I've done this before"(with the confident outstretched palm down hand) is straight up Flash even if it bled through Star Trek. Ming is just around the next nebula I just know it lol.
@Ydj79@seanboisselle@a16z No Yash it's not impossible, but it must be done the native text language of the models. The fact that no one even uses that or knows what it is is the real problem. It's not a tech problem, it's a people are not very bright problem like we are use to already lol.
@TheGuySwann@seanboisselle@a16z As a narrator you have a true grip on what you just said. You can feel it happening and even understand that the volume of words is not an indicator of quality of "truth delivery". Fewer words that better leverage one's existing knowledge is the better path.
@seanboisselle@a16z This is not the problem. You have a corpus of valid assumptions about the real world that you are not even aware of so can never put into words. It has no portability.
And the memory dynamics of you and llms are wildly different.
@mtaryma1@Mexnexus@oboofafrica I would love to hear you say all of that about the time you shift into high gear and the wheels start spinning and that thing in your lower spine tells you it's over but you haven't figured it out yet.
@Mexnexus@oboofafrica I was at about 2 degrees at speed or maybe less. I would use the airflow if I had to. But I'm not going to be sitting in that wreckage at the end with no chute deployed screw that.
The material is already under spec on thickness and his actual question is about redoing hundreds of holes or change the fasteners. We all know how to fix the holes and we can see where the screw head is. It is a matter of what is easier for him to do not you or me. And putting a fluted countersink in that triangular hole may not go as smooth as you think.
@brianmccallum@averagearc@fourierCO The hole is a triangle man, it's ruined. The angle of ruination does not matter and he running out of material thickness. Washers are better for this anyway and it's not he space shuttle so chill.
@averagearc@fourierCO We all known what that is. I'm with you on fastener change now. You still have alignment and will lose some if you go back in these holes to fix them. and you are short on thickness already.