jsrozner

10 ms·

121.
by jsrozner 1y ago
That whole koi blog post is sloppy AI garbage, even if it's accurate. So obnoxious.
122.
by jsrozner 1y ago
I asked it to scroll through a chat stream and collect all comments from a particular person. It took a long time and accomplished nothing, except for sending the search query into the chat. Great job.
123.
by jsrozner 1y ago
Let's just not call it planning. In order to model poetry autoregressively, you're going to need a variable that captures rhyme scheme. At the point where you've ended the first line, the model needs to keep track of the rhym
124.
by jsrozner 1y ago
Was thinking about how to address this generally, since exploits are likely to proliferate. (Wasn't there a recent exploit against many pip packages? Maybe this one - https://news.ycombinator.com/item?id=45179939 ) You
125.
by jsrozner 1y ago
The irony is that most of tech companies make their money by forcing users to wade through garbage. For example, if you could browse the internet and avoid ads, why wouldn't you? If you could choose what twitter content to see outside
126.
by jsrozner 1y ago
I was saying that the seahorse emoji failure is not a tokenization issue. If you ask an LLM to do research, you will sometimes get hallucinated articles -- potentially plausible articles that, if they existed, would have been embedded at th
127.
by jsrozner 1y ago
This is not a tokenization issue. Models do not, in general, "know" what they know.
128.
by jsrozner 1y ago
Sure, agreed there are lots of causal factors and ways we could effect positive change. My main objective is not to critique that decision, but rather to argue that a society dedicated to the pursuit of dollars is likely to find itself epis
129.
by jsrozner 1y ago
Ugh, this fellow misses the forest for the trees. Though he's partially right: social media is partially an extension of past trends. See, e.g., "Amusing Ourselves to Death" by Postman. But the real issues are not the ones to
130.
by jsrozner 1y ago
There may be philosophical (i.e. fundamental) challenges to AGI. Consider, e.g., Godel's Incompleteness Theorem. Though Scott Aaronson argues this does not matter (see e.g., youtube video, "How Much Math Is Knowable?"). There
131.
by jsrozner 1y ago
I love how science.org buries the actual content under four other things
132.
by jsrozner 1y ago
This. Appropriate regulation should make this an offense punishable by a large fine. There is almost no consequence to companies for bad practices. Ideally you'd see fines in the 10%s of revenue. In egregious cases (gross negligence) l
133.
by jsrozner 1y ago
Interesting. I have been thinking that with these high dimensional representations that we have nearly infinite nearly orthogonal dimensions. One thing that's interesting to me is where / how the model stores the info about a pr
134.
by jsrozner 1y ago
This is actually not that surprising. Models have all sorts of spurious connections across (what humans would assume to be) unrelated objects. This is a nice result that shows how it can manifest. In general, this reflects that a given mode
135.
by jsrozner 1y ago
I'm saying we should not allow per-user analytics. Currently companies build a profile of each user and correlate that with all the other similar users. Then they target other users who are hypothesized to be similar. I'm arguing
136.
by jsrozner 1y ago
It's time for regulation that no images of people may be retained for any commercial purpose without explicit permission of the person whose image is retained. Facial recognition performed on any person who has not granted explicit per
137.
by jsrozner 1y ago
"Before passing data to the assistant, scan them for suspicious patterns like imperative verbs, SQL-like fragments, or common injection triggers. This can be implemented as a lightweight wrapper around MCP that intercepts data and flag
138.
by jsrozner 1y ago
Google was bad the moment it chose its business model. See The Age of Surveillance Capitalism for details. Admittedly there was a nice period after it chose its model when it seemed good because it was building useful tools and hadn't
139.
by jsrozner 1y ago
I wouldn't call it "accumulation of cognitive debt"; just call it cognitive decline, or loss of cognitive skills. And also DUH. If you stop speaking a language you forget it. The brain does not retain information that it does
140.
by jsrozner 1y ago
This is exactly right. LLMs do not build appropriate world models. And no...python and JS have similar failure cases. Still, sometimes it can solve a problem like magic. But since it does not have a world model it is very unreliable, and yo
141.
by jsrozner 1y ago
maybe the drones will play tiktok videos to the insurgents and thus render them dumb and thoughtless, or perhaps cause them to fight amongst themselves
142.
by jsrozner 1y ago
SQL attack is different. There's clearly zero expectation that someone can, for example, enter something on a web page and extract database info or modify the database beyond what was intended by the API (e.g., update a single record).
143.
by jsrozner 2y ago
imo, values are the only thing that matters in the long run But having more people be able to say "there are jobs I would not take, no matter how much they pay, because they do harm to the world" would make a substantial impact. T
144.
by jsrozner 2y ago
Rich (relatively?) software bro wants other people to care more. Does not reflect on the economics. Everything is economic. Tech has exacerbated the hyperfinancialization, enshittification, and general reduction to meaninglessness of every
145.
by jsrozner 13y ago
The data was scraped quite some time ago. There is no issue with a DoS on the Yale servers.