I know we're all experiencing NDFSMs differently, like that's part of the whole problem, but 5.5 just gave me "The truncating quantizer collides two oranges", which is a new low for me.
Like the sibling comment says, text is always "true" in the sense that it's reacting to context correctly. So yes, true, but insufferable.
Ironically, what I'm working on a post-processing hook for colorizing and summarizing responses without degrading the session quality. So "truncating" = summarizing, "quantizer" = char limits and thresholds, "collides" = conflicts, "two oranges" is referring to the "alert level colors" where a second model (Haiku/Sonnet) colorizes text based on the perceived (or suggested) priority of a response's statements (e.g. "just so you're aware, I didn't commit" is fucking useless and it needs to be blacked out).
So the original insufferable statement translates to something like "The code that checks whether a text fragment is too verbose was conflicting with the part that colorizes the text."
P.S. Let me know if there's something out there that exists like this- something that adds a dimension like color or priority-assessments on a per-response basis. So far all I've seen is 2 dozen ~100k starred GitHub plugins that add zero value or make things worse.
If anything it seems worse. I'm experiencing about -10% insufferable jargon, but +30% more verbosity. It's unredeemable. There also seems to be even less structured output (headings, bullets, etc).
I agree that OpenRouter kinda sucks, but it took about 15 seconds to see that a few sentences got reverse-compressed into what is now this article, and it's slop.
The [content-based] trust layer of the internet is what we have deferred to this point, and what needs to be created (which is easier said than done, of course).
Personally I think a web-of-trust Keybase-style solution could work... But buy-in is difficult... you'd need some sort of "seed" strategy to make the app useful alack of 100 (a huge number) trusted friends.
Saying "more" seems generous. I don't trust Pangram at all, and I also don't trust the author at all. Both can be trustless charletans simultaneously, and I feel this speaks volumes about the state of digital media today.
Whether that chess repo is genuine or not genuine, legal or illegal, tasteful or tasteless, vibed or not vibed, stolen or copied, the signal to open source maintainers is clear: get off GitHub.
I admit that I read this stuff for self-soothing, although the self-sooth is not about AI going away, just that SotA will become expensive enough that everyone stops shilling digital garbage everywhere all the time.
If SotA does not go up in price (ala open weights), then everyone will continue to shill digital garbage AND the stock market will presumably get ravaged. So I'm really crossing my fingers about cost increase and- full circle- this hope is why I like to self-soothe with these articles (i.e. I cannot attest the articles' veracity but it makes me feel good that a human writer is speculating convincingly about instability).
That said, I sense that the ravaging is a forgone conclusion, unless someone pulls a(nother) rabbit out of a hat. I'd qualify agentic coding was a rabbit, so it could happen again I suppose.
I think the big picture is whether the big two can mask the real cost of tokens (inference and training) long enough to survive; that OpenAI is seeking government investment suggests that they can't.
Ed Zitron helps you see the broad absurdities in this situation and more importantly helps you understand that at various levels there is chicanery: they need more money constantly so they make weird unprovable claims to people who lack the skills to critique them, they exploit not jusT FOMO but actually fear of their own products to market them, their suppliers are making such ruthless use of SPVs as to conjure a possible apparently entirely legal Enron-scale disaster, they are attempting to scare the US government into backing them, and they have tried to get the index funds to back them before they have any hope of profit. None of it is normal; much of it is operating in the face of whatever still passes for principle in the US tech and finance industries. The broad picture is obscured by smoke.
But as fascinating as it is, it's only background information and it only broadly supports one's instincts; anything can still happen. For me, the main instinct it provides some support for is that the market is so terrifyingly lacking in verifiable fundamentals and stable predictable pricing that ultimately people will be driven to open weights and local models to keep control of their businesses; anything else is not a sane bet. So it broadly supports my instincts about my local-first approach, to find sustainable value.
The thing about pulling rabbits from hats is that the rabbit was always there; it's not a new rabbit. With a bit of understanding about how GPTs and LLMs work it is clear that code analysis and generation is not just a good application, it is the utmost practical and profitable one, because nowhere else in broad human effort do we communicate only in context-free grammars. Nothing else is actually going to prove to be more valuable financially).
(Formal pure maths is the "purest" one which is why we see its value in chugging through maths puzzles. But is there as much money to be found there?)
reply