I assumed (perhaps incorrectly, but it was a guess since I never dug into it) that less used "hot pockets" of previous inference gradually got more stale as time went on, and the conversation went elsewhere and didn't need those bits.
Hearing one byte refreshes the whole thing is huge! 5min is wayy too slow, because sometimes I want to spent more than 5 min looking at a diff before choosing where to go next.
Kind of outrageous, I hope this kind of feature gets built into claude code =D
Imagine getting free tokens from OpenAI and the special flag on your account that tells it to "never nerf you" ?
To be clear, I've had some great success using these tools, they're amazing, but it's also very obvious that (despite it being a stochastic tool) they're clearly changing things behind the scenes a lot. GPT-5.6 for example should not be constantly tweaked without telling us. Make it 5.6.1 or something, and if you're over capacity, just say so rather than returning lower quality results.
This only makes sense if the underlying thing that gets exposed after you put in the password is a block device... I'd want to then be able to have that encrypted with LUKS myself, I don't want to rely on this homebrew stuff.
What a badly written article. How could you leave out the N900 and the development history that came before it? If it wasn't for Elop and Microsoft, Nokia could have dominated and we wouldn't be stuck with mostly proprietary phones.
Haskell definitely needs to improve a lot of things, but moving to python is obviously a mistake for most cases. Golang is probably the best language for LLM things, it's got reasonably strong types, compiles very fast, and is simple. Much nicer to work with than Rust for example. But these days most code should probably be golang or rust if you're not building a webui.
Kind of a crap article that doesn't (1) link to the code (or to the golang implementation from the paper) and (2) doesn't even have the word "CAP" in it.
I like the privacy conscious aspects. Apart from the obvious issue of "run it through Claude" how many of those referenced analysis tools are entirely open source or at least run locally? Would have liked to see that in the article.
Having it stick out like that is such a stupid design. Almost as dumb as all the 2FA dongles. The USB-A ones that you could leave in actually made the most sense. Yes I know.
Concise and spot on. I learned about this "thinking" stuff not too long ago, and I was quite surprised that they keep it hidden. Long-term this isn't going to fly. I hope we get truly open models going and let them be owned by society.
The difference between this one (good) and the Anthropic program (bad) is that openai doesn't force you into a marketing clause while Anthropic does.
I mean seriously, you already ripped off all the worlds open source code. Be more generous and don't demand anything else back. Six months is so little too.
Signal refuses to answer: Why won't they release/open source all of their backend infra automation scripts/tools/etc...
There's no reasonable reason why a 501(c)(3) won't put this out there to make sure there's redundancy so we could built an alternate network if they're compromised by some gag order.
Anthropic is missing the plot by not releasing even the CLI (tui) claude code as open source. On the other hand, Codex is so garbage that as soon as they catch up it will be done for Claude.
https://github.com/purpleidea/mgmt/