Of course it depends on the type of software you’re writing and just how strict it needs to be, but in general, I find that you need the “spec-driven development” paradigm less now than you did maybe 6 months ago. Older specs really begin to become noise and context pollutants, plus the model is just going to explore the codebase to get a current state of truth regardless.
Or in some cases, not necessarily "owned" but purposely acting on behalf of foreign agents. Kushner + Witkoff running point on all Middle East-related talks on behalf of Israel with their only credibility to do so is just being Jewish, apparently.
Trump is just going to pardon everyone around him like Biden did. Not saying anyone Biden pardoned did wrong, just that it's a thing that can be done without having to name any "crimes" the pardons may be for.
"How Much They Know About AI" ... but it's actually how men lie about knowing "celebrities" in the space that may or may not exist? Seems like a pointless article. Same thing happens in every space. See: people wearing Nirvana t-shirts not knowing they were a band in the 90s.
Does DeepSeek provide the inference code along with the weights? I read here recently that for models where only the weights are released, providers have to implement which could be different than what the labs themselves use.
The whole Skydance/Paramount (Ellison) take over of Warner Bros, CBS, etc. and then installing people like Bari Weiss as editor-in-chief does _not_ help the “own the media” trope.
Is there a better term for self-improvement where the model is actually improving _itself_ (e.g., adjusting its own weights in real-time upon learning new, verifiable information)? I’ve heard online learning.
The way “recursive self-improvement” is being used recently seems to be intentional marketing. A model helping code/train/test the next generation of its model family isn’t technically “self”-improvement.
What is the current “state of the art” of AI development? Aside from running as many agents as possible, the models seem to have gotten to a place where iterative, straight-forward prompting works pretty well. SDD frameworks and such might have helped at one point but they seem like document bloat noise now.
The Israeli lobby poured a lot of money into his campaign after 10-7 and the entire reason the current administration even has beef with the ICC is because they issued arrest warrants for Netanyahu.
This must be related to how Toto turns off George Russell’s car if it’s looking like he’s still a potential threat to the golden child, Kimi. Kidding. Kind of.
I'm wondering how much was to push towards supporting Israel during the college campus protests and any future (at the time) conflict with Iran and it's proxies post October 7th like a lot of other Jewish million/billionaires such as Bill Ackman that typically supported Democrats or liberal causes. Brockman doesn't seem to have anything publicly stating he's Jewish but "Brockman" has English/German/Jewish origins and his mother's maiden name was Feldman.
I guess it’s hard to actually calculate, but most people are probably on subscription plans with Anthropic or OpenAI and there’s not any good resources to estimate which models are the most “usage limit” efficient. I suppose you can sort of correlate raw API costs with how much usage a model would probably drain, but it would be nice if there was a reliable place to get that info.