andy99

37 ms·

901.
by andy99 1y ago
Is there a way for not Twitter users to read these?
902.
by andy99 1y ago
Overshooting the capabilities of LLMs is pretty natural when you're exploring them. I've been using them to partially replace stack overflow or get short snippets of code for ~2 years. When Claude code came out, I gave it increase
903.
by andy99 1y ago
I always thought kids safety was one of the four horsemen of the internet apocalypse, as in the internet would be destroyed by political actors in the name of "think of the children". I tried to confirm or refute that, but the int
904.
by andy99 1y ago
I still think the original tweet was tongue-in-cheek and not really meant to be a serious description of how to do things.
905.
by andy99 1y ago
I think lack of critical thinking is the root cause, not a symptom. I think pretty much everyone uses LLMs these days, but you can tell who sees the output and considers it "done" vs who uses LLM output as an input to their own pr
906.
by andy99 1y ago
I wish more had been written about the first assertion that using an LLM to code is like gambling and you're always hoping that just one more prompt will get you what you want. It really captures how little control one has over the pro
907.
Why Moderna Merged Its Tech and HR Departments(wsj.com)
38 points by andy99 1y ago|55 comments
908.
Climbing Everest in a week using an anesthetic gas(cnn.com)
1 points by andy99 1y ago|0 comments
909.
by andy99 1y ago
A lot of that is down to the training, the "voice" is that of a junior Deloitte consultant writing a report. But this was intentional, as in it was trained to have this voice by virtue of the datasets used in the sft and the rlhf
910.
by andy99 1y ago
Personal pet peeve is the "I made a" prefix to titles. It adds nothing but an apparently selfish shift to the author/creator as opposed to their work.
911.
by andy99 1y ago
The real title is "Q&A with Jason Pruet"
912.
by andy99 1y ago
My sense is that these benchmarks are not realistic in terms of the way the model is used. People building specialized AI systems are not, in my experience, letting users just chat with a base model, they would have some variant of RAG plus
913.
by andy99 1y ago
Two different issues that while apparently related need separate consideration. Re the copyright finding, does the US copyright office have standing to make such a determination? Presumably not since various claims about AI and copyright ar
914.
by andy99 1y ago
Trading one thing for another is expressing your view that the latter is more valuable than the former.
915.
by andy99 1y ago
Dogs aren't rlhf'd and fine tuned to enforce behaviors designed by companies.
916.
by andy99 1y ago
The failure modes from 2023 are identical to those today. I agree with the now deleted post that there has been essentially no progress. Benchmark scores (if you think they are a relevant proxy for anything) obviously have increased, but (f
917.
by andy99 1y ago
I like to read the Code Golf Stack Exchange: https://codegolf.stackexchange.com/ While I don't know how well it generalizes, pages like this can't be replaced by LLMs, and still contain instructive technical conte
918.
by andy99 1y ago
A better bird analogy would be if we didn't understand at all how flight worked, and then started throwing rocks and had pseudo-intellectuals saying "how do we know that isn't all that flight is, we've clearly invented a
919.
by andy99 1y ago
An LLM is a classifier, there is lots of research into how deep learning classifiers work, that I haven't seen contradicted when applied to LLMs.
920.
by andy99 1y ago
Different initialization converge to different places, e.g https://arxiv.org/abs/1912.02757 For LLMs (as with other models), many local optima appear to support roughly the same behavior. This is the idea of the proble
921.
by andy99 1y ago
Yes, deep learning models only interpolate, and essentially represent an effective way of storing data labeling effort. Doesn't mean they're not useful, just not what tech adjacent promoters want people to think.
922.
by andy99 1y ago
Emergence in the context of LLMs is really just us learning that "hey, you don't actually need intelligence to do <task>, turns out it can be done using a good enough next token predictor. We're basically learning what
923.
by andy99 1y ago
I didn't follow entirely on a fast read, but this confused me especially: The parameter count of an LLM defines a certain bit budget. This bit budget must be spread across many, many tasks I'm pretty sure that LLMs, like al
924.
Wärtsilä launches carbon capture solution to shipping market(wartsila.com)
2 points by andy99 1y ago|2 comments
925.
by andy99 1y ago
Re vibe coding, I agree with your comments but where I've used it is when I needed to mock up a UI or a website. I have no front end experience so making a 80% (probably 20%) but live demo is still a valuable thing, to show to others t
926.
Couple's airline credit used for stranger's flight – and Air Canada blamed them(cbc.ca)
3 points by andy99 1y ago|1 comments
927.
by andy99 1y ago
I used to teach, years before LLMs, and got lots of copy-pasted crap submitted. I always marked it zero, never mentioning plagiarism (which would require some university administration) and just commenting that I asked for X and instead got
928.
by andy99 1y ago
Others who didn’t use the software effectively spent hours and days manually scrolling through their photo grid. I attempted to tell them how they could make it easier, but they weren’t in the mood to deal with technology lessons, unde
929.
by andy99 1y ago
We moved recently, and admittedly lucked out getting a good mover, but the main difference that made me pick the one we did was that they were going to use their own truck and crew. We only talked to 2 companies, and the other one, despite
930.
by andy99 1y ago
Curious if there are specific route features that make this feasible or not, like traffic conditions or the roads or the warehouses on either end.

More ›