![](/static/253f0d9/assets/icons/icon-96x96.png)
![](https://beehaw.org/pictrs/image/c0e83ceb-b7e5-41b4-9b76-bfd152dd8d00.png)
The temperature scale, I think. You divide the logit output by the temperature before feeding it to the softmax function. Larger (resp. smaller) temperature results in a higher (resp. lower) entropy distribution.
The temperature scale, I think. You divide the logit output by the temperature before feeding it to the softmax function. Larger (resp. smaller) temperature results in a higher (resp. lower) entropy distribution.
Have you read the “About the film” section at the bottom?
When two young American Jews raised to unconditionally love Israel witness the brutal way Israel treats Palestinians, their lives take sharp left turns.
Here’s the list of the filmmakers. Notice that there is a Palestinian-American in it.
You might have misunderstood who created the film and what it is about.
Your implicit assumption here is that both criticisms and solutions are equally easy (or difficult) to make, which is obviously not true, hence the relative quietness.
come up with new and unexpected things that never existed before
I’m not sure this is possible if the tech is still primarily built by learning from data, which by definition, has existed.
only in the podcasts I listen to
Yes definitely. Many of my fellow NLP researchers would disagree with those researchers and philosophers (not sure why we should care about the latter’s opinions on LLMs).
it’s using tokens, which are more like concepts than words
You’re clearly not an expert so please stop spreading misinformation like this.
You seem very certain on this approach, but you gave no sources so far. Can you back this up with actual research or is this just based on your personal experience with chatgpt4?
Can you provide the source of a few of these completely different LLMs?
add even a small amount of change into an LLM […] radically alter the output
You mean perturbing the parameters of the LLM? That’s hardly surprising IMO. And I’m not sure it’s convincing enough to show independence, unless you have a source for this?
two totally independent LLMs
How do you propose to get these independent LLMs? If both are trained using similar objectives e.g., masked token prediction, then they won’t be independent.
Also, assuming independent LLMs could be obtained, how do you propose to compute this hallucination probability? Without knowing this probability, you can’t know how many verification LLMs are sufficient for your application, can you?
I agree with what you said about price fixing and competition. But why do you think multiple robotaxi companies will survive in the (long) future? We know that’s not what happened with Uber/Airbnb that killed their competitors with predatory pricing. How do you know this time it would be different?
Thanks for the detailed cost breakdown. You seem to have thought about this deeply. But I don’t see labour cost (e.g., engineers) in the breakdown. Why did you not include it?
I agree that the battery cost (and thus operating cost) would go down, but again I’m not convinced it would mean lower fare because that’s not what usually happens. I also agree these companies know how much money will be made on self-driving systems, which is exactly why I think they would aim for a monopoly, and the one surviving would charge passengers as high as it can.
In you and your wife’s case, is using a normal taxi 5-6 times not cheaper than the second vehicle cost?
Anyway, from what you wrote, it seems the biggest issue here is cost/fare. In that case, public transit, which we already have and benefits all people (including both the elderly and people with disabilities), would be a better solution than any taxi.
If the operating cost is as low as you said, why do you think these robotaxi companies wouldn’t eventually charge similar fare to normal taxis given that (1) the market can bear such fare now and (2) the reduced operating cost would give them higher profit margins?
Frankly, I’m not convinced yet that the operating cost is that much lower. Covering more areas and operating almost 24 hours a day sound like more fleet and more frequent maintenance to me. Wouldn’t these increase the operating cost, and thus, fare? Not to mention paying the engineers to maintain the software/AI system. I assume engineers are much more expensive than drivers.
Oops didn’t realise you aren’t the guy I replied to!
Just to clarify, you regard both cars and houses as death machines because nearly everyone that owns them has died?
I don’t follow. Are you equating houses with cars as “death machines”? If yes, how so?
And how do you think robotaxis address all these issues (high fare, poor coverage, limited operating hours)?
I can’t tell you who/what to vote. But I’m glad we agree that the city is the issue.
You know the problem already: poor coverage of public transit. Why not advocate for that? It’s much safer than any cars, and we have the tech right now. We can stop killing people right now.
Many people literally die each year because of car-centric infrastructure, and you’re basically telling us to calm down? No fucking way.
Before cars, there used to be a massive amount of infrastructure for trains and streetcars, not to mention walkable neighbourhood, but they all get demolished for cars. So yes, we do operate like that.
And did I mention that car-centricity kills people each year? So yes, eliminate all cars if we have to. But honestly, I don’t think anyone wants to eliminate all cars, just those we don’t need (which is most of them).
Sounds like your city is the problem here, not trains. Vote better.
Normal taxis benefit both the elderly and people with disabilities like you said right now. So why bother with robotaxis?
Each row in the figure is a probability distribution over possible outputs (x-axis labels). The more yellow, the more likely (see the colour map on the right). With a small temperature (e.g., last row), all the probability mass is on 42. This is a low entropy distribution because if you sample from it you’ll constantly get 42, so no randomness whatsoever (think entropy as a measure of randomness/chaos). As temperature increases (rows closer to the first/topmost one), 42 is still the most likely output, but the probability mass gets dispersed to other possible outputs too (other outputs get a bit more yellow), resulting in higher entropy distributions. Sampling from such distribution gives you more random outputs (42 would still be frequent, but you’d get 37 or others too occasionally). Hopefully this is clearer.
Someone in another reply uses the word “creativity” to describe the effect of temperature scaling. The more commonly used term in the literature is “diversity”.