Gen 1 Apple Card did a bait-and-switch about using it to buy iPhones with 0% interest. Super frustrating.
2,007 karma · joined February 4, 2014
Why?
Whether the LLM could do a better job than me at writing the essay is a separate question...I suspect it probably could. But it wouldn't be as fun.
"LLMs" are like "screens" or "recording technology". They are not good or bad by themselves - they facilitate or inhibit certain behaviors and outcomes. They are good for some things, and they ruin some things. We, as their users, need to be deliberate and thoughtful about where we use them. Unfortunately, it's difficult to gain wisdom like this a priori.
I heard someone say that LLMs don't need to be as good as an expert to be useful, they just need to be better than your best available expert. A lot of people don't have access to mental health care, and will ask their chatbot to ask like a psychologist.
"We present Dreamer 4, a scalable agent that learns to solve control tasks by imagination training inside of a fast and accurate world model. ... By training inside of its world model, Dreamer 4 is the first agent to obtain diamonds in Minecraft purely from offline data, aligning it with applications such as robotics where online interaction is often impractical."
In other words, it learns by watching, e.g. by having more data of a certain type.
https://www.harpercollins.com/products/the-body-electric-rob...
Suppose further that all events are a draw of type 1, 2, 3, or 4, and that our memory kept a count and updated the distribution - it is essentially a frequency distribution.
When we encounter a stimulus, we have to (1) recognize it and (2) assign a reward valence to it. If we only ever observed '3', the distribution would become very peaked. Correspondingly, this suggests that we would recognize '3' events faster and be better at assigning a reward valence to those events.
Then if we ever encounter a non-3 event, we would recognize it more slowly - it is well-established that recognition is tied to encounter frequency - and do a poorer job assigning reward valence to it. Together this means that we would do a bad job selecting the appropriate response.
Perhaps this scenario-based dreaming keeps us (and rats) primed so we're not flat-footed in new scenarios.
The question then becomes - if these scenarios are purely imagined, where are they being sampled from? If we never observe 1, 2, and 4...how do we know that these are the true list of alternative scenarios?
If you're looking for utility in terms of understanding cognition, here are some resources if you're interested - pop-sci books written by cognitive scientists:
- https://global.oup.com/academic/product/the-mind-within-the-...
- https://press.uchicago.edu/ucp/books/book/chicago/R/bo362751...
Is the division true? Of course not - to paraphrase Box, all models are wrong. But some are useful. This one is useful.
Seems like anyone who buys or sells gold at any other price is questioning it.
Wolfe wrote in a letter, "My definition of a great story has nothing to do with 'a varied and interesting background.' It is: One that can be read with pleasure by a cultivated reader and reread with increasing pleasure."
https://ai.meta.com/research/publications/large-concept-mode...
A persistent issue I have with Cursor et al. is that they hallucinate function arguments when using a function or method from a library. It seems like automatically pulling the library's documentation into the context would be helpful, but I haven't found any tool that does this automatically. Is there any chance that Runner does this?
It doesn't seem like this was the problem you were trying to solve, but reliable use of libraries and APIs is a critical problem to solve if you want LLM-generated code to work.
You CAN model predator-prey dynamics or disease spread using Monte Carlo techniques, but you can't read the historical literature without some grasp of differential equations.
One thing humans seem to be uniquely good at is picking goalposts that separate us from other species.
Do you have any sense of whether this type of model would work with children's speech? There are plenty of educational applications that would value a privacy-first locally deployed model. But, my understanding is that Whisper performs pretty poorly with younger speakers.
"Love is the active concern for the life and the growth of that which we love."
“An important factor in the purchase of the F-35 by European governments was the idea that European defense would be built on a transatlantic basis in terms of strategy, institutions and capabilities,” she said, adding that “the Trump administration is in the process of dissolving the transatlantic link, and the purchase of American systems will therefore no longer have any added value for Europeans.”
“If you keep punching your allies in the face, eventually they’re going to stop wanting to buy weapons from you,” said a Western European defense official, granted anonymity to discuss the matter candidly. “Right now we have limited options outside of U.S. platforms, but in the long run? That could change in the coming decades if this combativeness keeps up.”
[1] https://www.politico.eu/article/punching-allies-in-the-face-...
It seems harder to prove that every real number can be constructed via such a method.
Is there a construction-based method that can produce ALL real numbers between, say, 0 and 1? This seems unlikely to me, since the method of construction would probably be based on some sort of enumeration, meaning that you would only end up with countably many numbers. But maybe someone else can help me become un-confused.