Do you mean a terminal-based editor, like emacs, vim, neovim, or helix? (I quite like the latter, after having used all the former to some degree.)
Or do you mean line-editors? They have gotten impressively good. See rustyline (based on linenoise) and reedline (not a typo; developed by the Nushell team) for example. Way better than one might expect!
I love it. I have 40 files open in it right now - most of them .c source files with LSP and Treesitter running and it uses 250MB of RAM. Everything works instantly, Telescope is imo the new standard for file navigation (and navigation in general: grep, symbols even neovim shortcuts). Git integration is fantastic as well.
I have spent some time configuring it and probably will spend more when I start including more languages but imo it's worth it. You can configure everything but you can also find very nice defaults by running Kickstarter (or some heavier neovim "distro").
Microsoft has done great work with LSPs - I can now get great navigation/autocompletion/formatting/inline errors/warning combined with neovim navigation, light weight and fantastic tools/extensions.
One thing I haven't integrated yet is a debugger (gdb from the terminal is good enough for me), maybe that's something people are missing in neovim?
TL;DR is that they didn't clean the repo (.git/ folder), model just reward hacked its way to look up future commits with fixes. Credit goes to everyone in this thread for solving this: https://xcancel.com/xeophon/status/2006969664346501589
(given that IQuestLab published their SWE-Bench Verified trajectory data, I want to be charitable and assume genuine oversight rather than "benchmaxxing", probably an easy to miss thing if you are new to benchmarking)
> I want to be charitable and assume genuine oversight rather than "benchmaxxing", probably an easy to miss thing if you are new to benchmarking
I don't doubt that it's an oversight, it does however say something about the researchers when they didn't look at a single output where they would have immediately caught this.
Given the decrease in the benchmark score from the correction, I don't think you can assume they didn't check a single output. Clearly the model is still very capable and the model cheating its results didn't affect most of the benchmark.
Isn't it the opposite? From the link: Scores range from -100 to 100, where 0 means as many correct as incorrect answers, and negative scores mean more incorrect than correct.
Gemini 3 Flash scored +13 in the test, more correct answers than incorrect.
If you start from low levels, then yes, as long as you keep your blood levels in check. It would take a while to overdose it this way, but it's not impossible.
i cant find it now. It was an article that suggested you may underestimate your natural production, especially in summer, so its safer to use high doses for a period of 3 months or so with a break of the same duration, after all we are likely “accustomed “ to fluctuating supply through the year.
Technically you kind of get this in Nevada when using Tesla insurance and if you drive 100 % FSD. If you drive manually you are pretty much doxed for random Front collision Warning which is super sensitive
It might be that our current tokenization is inefficient compared to how well image pipeline does. Language already does lot of compression but there might be even better way to represent it in latent space
A lot of cool things are shot down by "it requires more compute, and by a lot, and we're already compute starved on any day of the week that ends in y, so, not worth it".
If we had a million times the compute? We might have brute forced our way to AGI by now.
It's kind of a shortcut answer by now. Especially for anything that touches pretraining.
"Why aren't we doing X?", where X is a thing that sounds sensible, seems like it would help, and does indeed help, and there's even a paper here proving that it helps.
The answer is: check the paper, it says there on page 12 in a throwaway line that they used 3 times the compute for the new method than for the controls. And the gain was +4%.
A lot of promising things are resource hogs, and there are too many better things to burn the GPU-hours on.
This has a ton of seemingly random assumptions, why can't we compress multiple latent space representations into one? Even in simple tokenizers token "and" has no right being the same size as "scientist".
Image models use "larger" tokens. You can get this effect with text tokens if you use a larger token dictionary and generate common n-gram tokens, but the current LLM architecture isn't friendly to large output distributions.
You don't have to use the same token dictionary for input and output. There's things like simultaneously predicting multiple tokens ahead as an auxiliary loss and for speculative decoding, where the output is larger than the input, and similarly you could have a model where the input tokens combine multiple output tokens. You would still need to do a forward pass per output token during autoregressive generation, but prefill would require fewer passes and the KV cache would be smaller too, so it could still produce a decent speedup.
But in the DeepSeek-OCR paper, compressing more text into the same number of visual input tokens leads to progressively worse output precision, so it's not a free lunch but a speed-quality tradeoff, and more fine-grained KV cache-compression methods might deliver better speedups without degrading the output as much.
Similar feeling. Seems it is good at certain things and if something doesnt work it want to do things simply and in turn becomes something that you didnt ask for and certain times opposite of what you wanted. On the other hand with codex certain time you feel the AGI but that is like 2 out of 10 sessions. This is primarily may be due to how complete the prompt and how well you define the problems.
"Simple, Boring Tech Stack:" Good advice but bad example, because it depends on what engineers are familiar and comfortable with and technology itself should be mature enough. You dont want to spend time building orchestrator when k8s solves it for you. Most cloud provides provide you with k8s as a service, which are miles better than using shell scripts, if you are already familiar with k8s
reply