devmor

34 ms·

811.
by devmor 1y ago
I have repeatedly tried to use LLMs as search engines, both general (like ChatGPT) or more focused on specific domains. I have not been impressed by the results. In my experience, LLMs used this way generally output confident-sounding infor
812.
by devmor 1y ago
Yes, but when they raised before they did not give up a bunch of control in return.
813.
by devmor 1y ago
I was actually thinking of that paper when I wrote that comment, hence the frustration that we don't actually know the intermediates. Still, perhaps the stepped output we get may hint at that kind of "cheating" and can be use
814.
by devmor 1y ago
Sure, but there's something to be said for continuing to add non-green energy alongside it. From an ecological perspective, it's like going on a diet and smoking an extra cigarette for every 100 calories you cut out.
815.
by devmor 1y ago
How do you reconcile that opinion with the fact that LLMs trained on programming concepts generally give incorrect answers about 50% of the time? Is it actually more useful than Googling, or is it just so convenient that you let it convin
816.
by devmor 1y ago
Depressing news, I have no hope that the countdown to Tailscale being unusable subscription trash has not started with this announcement. I realize this is a very ironic place to make this statement, but I am utterly exhausted by VC money d
817.
by devmor 1y ago
To me, this is one of the most frustrating parts of this type of ML. If we could actually track the steps taken in the LLM, it would be trivial for the judge to evaluate the output of each intermediate and detect when reward hacking is ta
818.
by devmor 1y ago
That is a much better comparison.
819.
by devmor 1y ago
Why are you attempting to technically analyze a simile? That is not why comparisons are used.
820.
by devmor 1y ago
I don't think your detraction has much merit. If I don't understand how a combustion engine works, I don't need that engineering knowledge to tell you that a bicycle [an LLM] isn't a car [a human brain] just because it f
821.
by devmor 1y ago
I can't really test it out, given the hug of death it is currently receiving, but I am happy to see the project. The USDA nutritional database is a nightmare to query.
822.
by devmor 1y ago
Yeah, that's fair - it's very easy to tell that LLMs are not the end state, but it's near impossible to know what comes next. Personally I think LLMs will be relegated to transforming output and input from whatever new logic
823.
by devmor 1y ago
LLMs in their entirety are unlikely to move past tokenization - it is the inescapable core from the roots of NLP and Markov Chains. The future of AI and all of ML in general likely does exist beyond tokenization, but I find it unlikely we w
824.
by devmor 1y ago
I think you are mistaken about who you're replying to, but it shouldn't matter. Someone made a claim. You made a claim that they are specifically wrong and the opposite is true, not that their claim is unproven. I just wanted to
825.
by devmor 1y ago
Do you know of a study that shows that long term effects in men and women are the same or worse in women?
826.
by devmor 2y ago
Until we solve the storage problem, the manufacture of batteries required to store solar power at night and during other low generation times is actively doing harm even without failing. I don’t think this invalidates your point but I do th
827.
by devmor 2y ago
It’s not about what they can do, it’s about what people think they can do and try to use them for, which is, on average, incredibly wrong. They are impressive for what they are… but do you know what they are? I do, and that’s why I’m not th
828.
by devmor 2y ago
Evolution also has no "goal" other than fitness for reproduction. Training a neural network is done intentionally with an expected end result.
829.
by devmor 2y ago
What you just said means absolutely nothing and has no comparison to this topic. It’s nonsense. That is not how evolution works.
830.
by devmor 2y ago
In what way do you view this as Microsoft taking control of Linux? This is an effort to support more users leaving Microsoft for Linux and taking their software with them.
831.
by devmor 2y ago
And less color fidelity on receiving displays I would imagine!
832.
by devmor 2y ago
Is initcall_blacklist not enough of an option for you?
833.
by devmor 2y ago
The developers don't necessarily have to be the ones working on mitigation efforts.
834.
by devmor 2y ago
It could also be any number of other things too, like it's severe enough that the author feels its responsible to wait for mitigation efforts before disclosing anything about the issue that could lead to it being exploited.
835.
by devmor 2y ago
I do not understand what you find convincing about this that changes your mind. We have a closed system that we designed to operate in a way that is similar to our limited understanding of how a portion of the brain works, based on how we w
836.
by devmor 2y ago
I will never understand why people are willing to connect so many of their devices to the internet for minimal features. I went out of my way to build a network that prevents even the things I want to have local wifi access from being acces
837.
by devmor 2y ago
#2 has been a godsend in the custom/HEDT PC market. Many expensive motherboards now come with a "dual BIOS" system that gives you an older known working image to boot from, in case flashing a new version broke something that
838.
by devmor 2y ago
I mean that we have visual recognition systems that do not use any kind of machine learning whatsoever and those are the majority of systems in use at industrial scale. Laser interferometry and DCT image distance, primarily.
839.
by devmor 2y ago
Using AI for image recognition is to visual detection as Orange Juice is to beverages.
840.
by devmor 2y ago
If there's no evidentiary reason to assume causal relationship, I would say that's a useless distinction, personally. There's just as much reason to assume that if you didn't drink at all, you'd have thought of it s

More ›