btown

8 ms·

91.
by btown 4mo ago
In theory reasoning tokens should do the equivalent of this - explicitly create options outside of the quick-response probability space, so those can guide future generation. In practice, models that do this won't be prioritized as m
92.
by btown 4mo ago
It's worth noting that cybersecurity requirements can be a mechanism of control. As a government regime, do you want to build an effective surveillance system where health data on large numbers of suspects can be pulled into a data fus
93.
by btown 4mo ago
Surprised not to see a mention of https://tenbluelinks.org/ here. Google still maintains a web search mode that's free of AI overviews/chat exhortations (as well as ads, if you use an ad blocker). https:/&#x
94.
by btown 5mo ago
Very cool! I think the hype around “agents are so good that you never actually need to see the underlying commands they are running or interact with the terminal session that they’re running” misses out on a lot of very important use cases,
95.
by btown 5mo ago
The parent poster isn't saying "advancement of knowledge" is some kind of universal goal for humanity at the cost of all else - and I would agree that it shouldn't be. They're suggesting that as an individual studyi
96.
by btown 5mo ago
For the uninitiated: https://knowyourmeme.com/memes/poob-has-it-for-you
97.
by btown 5mo ago
I wonder if the only thing that will solve this is an insurer or regulator saying that: "A system that automatically pulls updates for dependencies without human review, where said updates are not protected by multi-factor authenticati
98.
by btown 5mo ago
Yes, and the problem here is that most projects have automated systems that automatically update those lockfiles on every upstream release of a library, under an assumption that minor releases are either security patches or bugfixes that wo
99.
by btown 5mo ago
At a certain point, is it better to just turn off Dependabot and freeze all NPM packages (minor/patch version and all), rather than continuously update? Particularly for frontend packages, meaningful security fixes seem less likely tha
100.
by btown 5mo ago
For those confused by the headline, HN changed it from MoE to Moe. Though, thinking about "efficient moe" I'm fondly reminded of projects like https://make.girls.moe/ (2017) - "Towards the Automatic Anim
101.
by btown 5mo ago
Via https://www.stainless.com/blog/stainless-is-joining-anthropi... that's exactly what seems to have happened: > As we focus on Claude Platform capabilities and connecting agents to APIs, we’ll be winding down
102.
by btown 5mo ago
FYI the above quote is (sadly) real and is from Stainless's blog post: https://www.stainless.com/blog/stainless-is-joining-anthropi...
103.
by btown 5mo ago
This is worth a read specifically for this section and the ones following it, re: custom vs. agentic-coding harnesses. https://blog.cloudflare.com/cyber-frontier-models/#why-point... Claude Code's harness is remar
104.
by btown 5mo ago
For those unfamiliar with the context: https://en.wikipedia.org/wiki/%27No_Way_to_Prevent_This,%27_...
105.
by btown 5mo ago
> The penalty is a 1-year ban from arXiv followed by the requirement that subsequent arXiv submissions must first be accepted at a reputable peer-reviewed venue. This is incredibly good for science. arXiv is free, but it's a privile
106.
Google users fight for refunds as unauthorized API usage bills soar(theregister.com)
2 points by btown 5mo ago|3 comments
107.
by btown 5mo ago
The reply has been flagged/marked dead by HN since, and should at minimum auto-collapse (thus hiding even worse replies to it) regardless of your account's "showdead" settings. I'm not going to link to it, and it&#x
108.
by btown 5mo ago
The actual answer: hedge fund manager Bill Ackman. https://www.institutionalinvestor.com/article/2bswuu1nfc040h... & https://finance.yahoo.com/news/visa-suspends-payments-for-ad... > But he
109.
by btown 5mo ago
That's very cool - using a linter as a standardization system removes a lot of the guesswork out of submitting! But it's an unenviable challenge to guard against bad actors here - there's now an open-source oracle that an att
110.
by btown 5mo ago
Well, for the time since then, the LLM providers have a corpus of every code suggestion they made to users and whether it resulted in positive or negative sentiment afterwards - which is arguably even more powerful. There's still som
111.
by btown 5mo ago
Congrats on the launch! Curious about whether the automated scanning system flags expansions of scope and network domain access for internal/human review. For instance, an AI summarization plugin that starts by saying it accesses url=&
112.
by btown 5mo ago
And I'm increasingly concerned that one could vibe-code a massive payload that does all of these at once - including subtle things like trying to get itself installed into personal projects and forks, so it can persist across a syste
113.
by btown 5mo ago
Also, every single interpreter error has an entire corpus of StackOverflow-esque fix suggestions alongside it, and the model has been fine-tuned to minimize such errors on the first try. This hasn't been done for more obscure languages
114.
by btown 5mo ago
The way I see it - if you're pushing a change to an NPM package with more than [N] daily downloads/downstream packages, and you don't have a human online who's able to approve a two-factor for the release on their phone.
115.
by btown 5mo ago
Do we know if this is immediately going to slot in wherever reCAPTCHA is currently used / is there a rollout plan? Or will site operators manually opt into the new system? Is there even a way to opt out? I can think of many sites where
116.
by btown 5mo ago
I find it fascinating how they were able to keep the reconstruction error function incredibly simple, literally its success in round-tripping the activation layer, while making it interpretable... simply by choosing a good data-driven initi
117.
by btown 5mo ago
If you're on macOS and interested in this space, I highly recommend you open up the system-provided Accessibility Inspector.app and play around with apps and browsers. See how the green cells might guide an LLM to only need to read&#x2
118.
by btown 5mo ago
For those interested in ASML, and how much engineering goes into being able to make EUV lithography work reliably at scale for 3nm nodes (by hitting individual droplets of tin with highly accurate lasers and turning them into plasma!), I hi
119.
by btown 5mo ago
Coupled with this (unsubstantiated but thorough) discussion on the internals of Azure, if even a fraction of this below-linked post is true, Github's abnormally-filesystem-intensive workflows would have wildly unpredictable performance
120.
by btown 5mo ago
At least it's MIT licensed! As much as non-open training data irks me, restrictive licensing irks me more!

More ›