> It's a direct call to not use many features, and claiming that to do otherwise is a mistake.
I think you are thouroughly misunderstanding the blog post. It's not an aggressive "you must start using this now" thing, but mainly a definition of what "Orthodox C++" is so that communication about coding styles is simplified.
When somebodies says "this project is using the Orthodox C++ style" then it is immediately clear to everybody what exactly that entails (and if not they can google the term it and find the blog post). It's also a counter proposal to some of "Modern C++" madness (like "almost always auto") that was all the rage around a decade ago when that post was first written.
FWIW I wrote a fairly similar blog post in 2013 to describe what our team coding style looked like back then:
I mean ordinary as non-electric. I know what an electric fatbike can do of course. The article makes it ambiguous as it's talking about the common limit for all cyclists.
I just ran a report from a project I'm working on that uses a mix of models, and GLM 5.1 trumped Sonnet over the last week, so I'm excited to now turn on 5.2. This is based on completion only - not quality, but that includes passing a huge test suite, and Sonnets failure rate was surprisingly bad...
What I've seen from 5.1 for things like planning has certainly not read as impressive as Opus, and often even as Sonnet, but it's been a strong and steady work-horse that's just kept on actually delivering progress.
jeez, y'all no fun. When bad shit is going down humour is _as_ an appropriate tool as being incredibly obvious. Lingering in a mixed area between the two also makes it considerably harder for censors and the subject to punish the disobedience. Innit?
Bit of what makes it so fun, our experiences seem to wildly differ! On one hand, you have experiences like yours, but then my own experience is that I never had a productive session when the scope grows beyond 150K tokens! If I needed 60K just as a starting context, I'd take that to mean the suggested change is way to large, and if the model cannot solve the entire thing within maybe 15-20% of the total context size, divide and conquer is needed otherwise there will be a lot of time wasted to patch things up when things are "completed".
If forms could evolve (self or own factors), multiply it by infinity to find 1 at least (as we are) ?
(aside of a a joke that the universe (with us as a part of it) is not evolving but computing a number (42 exactly) - which eventually may become real thanks LLM) [Lem:Dick .. (Adams)]
Yes. I had a printer that decided that the ink pad was full. Presumably you're supposed to throw the printer away at that point? The windows option was some dodgy warez type thing that you have to pay for. Debian just had a package on the repo.
This is a false equivalence between those who suffered from USSR and those who are ignorant of the suffering of others. I don't think we should care about feelings of a group who are for whatever reason nostalgic about a genocidal oppressive regime.
I worked in this field since long before LLMs. Nobody outside of the field really cared about GPT2, and even insiders knew the "too dangerous" part was a PR gag at best and the first dig of the moat at worst. After all, they released smaller versions of it along with detailed instructions on training it in the paper, so anyone with a lot of compute and a bunch of internet scrapers could try to recreate it. But basically noone did, even though it would have only cost ~50k back then (and less than 3k today). A few normal users started to take notice with GPT 3, but even then it was super limited. Even instructGPT didn't cause real shockwaves, despite being very close to the final product. Only ChatGPT/3.5 finally lit the fuse and people suddenly cared about having this too.
I built a very small personal extension for Pi [1] that gives me a /last command. It clears the entire session, only retaining the agent's last output message. This allows me to do manual "compaction". Basically I tell the agent something like "state the plan as discussed with references to files that should be edited", and call /last, then tell it to implement.
I think that's issue, rather than 60K being small.
Most of the actual edits/changes I request to codex are solved within 100-150K tokens, beyond 200K I'd definitively try to restart the session as soon as I could as all models are horrible once you get across ~20% of the total context size. And this is while working on +million LOC codebases.
Problem I guess is that there is no solid and concrete evidence of this (to me [and others seemingly] obvious) degradation, but should be easy to prove, yet no one has time to sit down and show it :)
But the likelihood of a model getting minor details wrong once you're above some magical threshold between 15-20%, seems to skyrocket, and I hit that issue sufficient amount of times that now my workflow is trying to prevent that.
Not necessarily against the idea but 20kph is very slow. It would be super easy to accidentally go over the limit unless you have a cycling computer attached to the bike and constantly monitor your speed.
World cup games are played on public broadcasting networks- communism! Which in reality means FIFA gets a nice paycheck from taxpayers. But no commercial breaks except during half time.
But its also pretty hard to observe in others, which makes it hard to come up with a satisfying definition. Without a concrete definition, you can't really do science with it.
In the original release, Anthropic called out that they showed the government that exactly the same outcomes could be achieved with ChatGPT models. So you don't have to bet, they said it publicly.
Keep in mind family of the administration is heavily invested in OpenAI.
I love the idea of building a new Git UI on top of Git primitives[1], but the REST inspiration leaves me feeling underwhelmed; there has to be a more human-friendly model.
If the model isn't following the system/developer prompts easily, you might want to try a bigger/better model, tends to mostly be about model quality if it doesn't follow what you tell it to. Besides that, conflicting directions in the system/developer prompts can lead to the model seemingly ignoring instructions too.
I think you are thouroughly misunderstanding the blog post. It's not an aggressive "you must start using this now" thing, but mainly a definition of what "Orthodox C++" is so that communication about coding styles is simplified.
When somebodies says "this project is using the Orthodox C++ style" then it is immediately clear to everybody what exactly that entails (and if not they can google the term it and find the blog post). It's also a counter proposal to some of "Modern C++" madness (like "almost always auto") that was all the rage around a decade ago when that post was first written.
FWIW I wrote a fairly similar blog post in 2013 to describe what our team coding style looked like back then:
https://floooh.github.io/2013/06/21/sane-c.html
(PS: back then I still thought that C++ could be salvaged)