andy99

29 ms·

511.
by andy99 11mo ago
Yeah if found it slightly ironic that an argument against using AI is made as an empty social media-style post. Ironically AI could have written a better one.
512.
by andy99 11mo ago
I don’t know US specifically but I know a bit about Canadian air traffic controllers. It’s a high stress but very well paying job with good pension etc. But I suspect there’s not a lot of opportunity to change to another job and get the sam
513.
by andy99 11mo ago
I also work in LLM evaluation. My cynical take is that nobody is really using LLMs for stuff, and so benchmarks are mostly just make up tasks (coding is probably the exception). If we had real specific use cases it should be easier to bench
514.
by andy99 1y ago
LLMs are optimized for sycophancy and “preference”. They are the ultra-processed foods of information sharing. There’s a big difference between having to synthesize what’s written in a book and having some soft LLM output slide down your gu
515.
by andy99 1y ago
Can someone ELI40?
516.
by andy99 1y ago
Just a signup page? These aren’t allowed for show HN, you don’t show anything. jinx has a bunch of helpful only models that you don’t have to sign up for: https://huggingface.co/Jinx-org/models#repos
517.
by andy99 1y ago
I don’t get it
518.
by andy99 1y ago
Anyone remember the news story about Metallica copyrighting G and F chords? It’s sort of like that, true or not it’s so on brand that people just accept it, it doesn’t tell you something you did already know about how google (or Metallica)
519.
by andy99 1y ago
LLMs are good at fuzzy pattern matching and data manipulation. The upstream comment comparing to awk is very apt. Instead of having to write a regex to match some condition you instruct an LLM and get more flexibility. This includes decidin
520.
by andy99 1y ago
Classical LLM hallucination happens because AI doesn’t have a world model. It can’t compare what it’s saying to anything. You’re right that LLMs favor helpfulness so they may just make things up when they don’t know them, but this alone doe
521.
by andy99 1y ago
Hello Carice
522.
by andy99 1y ago
Curious to see where these go. The world model, and other enhancements seem helpful but effectively become hard coded rules. We know that human specified rules generally underperform learned relationships (at least in previous ML work) so I
523.
by andy99 1y ago
I specifically remember watching a flight test doing an aggressive takeoff and having the voiceover say that aircraft (two engine) need to have enough power to take off full with one engine. And so can take off very steeply empty with two e
524.
by andy99 1y ago
I saw this on LinkedIn earlier and literally closed it after “ Let’s be frank the single‑player notebook has felt outdated for a while now” I think it must be messaging for “leadership” as opposed to practitioners, there are lots of real pa
525.
by andy99 1y ago
Just a caution, this guy is a Gary Marcus-like perennial chicken little, this isn’t some discovery or realization, it’s just an ongoing part of someone’s identity as an AI bear.
526.
by andy99 1y ago
Looks pretty normal, this is a hybrid or remote job in Toronto, it’s probably above market.
527.
by andy99 1y ago
Yes that’s exactly how I would characterize it
528.
by andy99 1y ago
I don’t normally wear deodorant, and afaik normally don’t smell (I know, but I’m married and have discussed this, agree it’s not foolproof, I also have a really good sense of smell). I also exercise a lot and subscribe to the idea that swea
529.
by andy99 1y ago
My experience is the same as yours. It feels to me (similar to most LLM writing) like they write for someone who’s not going to read it or use it but is going to glance at it and judge the quality that way and assume it’s good. Not to diffe
530.
by andy99 1y ago
I tried this briefly on a pine phone ~2 years ago as I was going through the different options trying to see if there was something viable. It was useless then, the phone barely worked, and I’m pretty low maintenance, I basically just wante
531.
by andy99 1y ago
It’s a non sequitur and not with engaging with. The purpose of a second chamber is temporal representation, as in the makeup is “stuck in the past” (more of a moving average) and not subject to the whims of the day. Some idea of forcing it
532.
by andy99 1y ago
Was that always the case? I took from TFA that some number are purely hereditary and that all of them used to be before a recent reform. Or was it that one has to be a hereditary peer in order to be a government appointee? Either way, the g
533.
by andy99 1y ago
This was posted from another source yesterday, like similar work it’s anthropomorphizing ML models and describes an interesting behaviour but (because we literally know how LLMs work) nothing related to consciousness or sentience or thought
534.
by andy99 1y ago
They are not political appointees, so have a lower chance of being correlated with whatever movement of the moment and so serve as checks and balances. The fact that they can’t be replaced is a feature.
535.
by andy99 1y ago
Having an apolitical (or unelected) and slow changing second chamber is a useful counterbalance to elected officials running amok. There’s no “great” answer to this that I’m aware of but it has been a viable compromise.
536.
by andy99 1y ago
What I was wondering is, do the injections cause the thinking trace to change (not whether they actually typed text into the thinking trace) and then the model “reflects” on the fact that it’s thinking trace has some weird stuff in it, or d
537.
by andy99 1y ago
I think that would be a good title if we’d previously thought lawn mowers had solved generalized grass cutting and assumed that because one worked on my lawn that they could cut hayfields or harvest bamboo (a grass I believe) effectively.
538.
by andy99 1y ago
Also in Neuromancer when Wintermute wants to talk to Case and is ringing all the pay phones. I’ve read that book a dozen times and never thought twice about it. It was recently pointed out to me how incongruous it is to be in the future wit
539.
by andy99 1y ago
I’d like to know if these were thinking models, as in if the “injected thoughts” were in their thinking trace and that’s how it was the model reported it “noticed” them. I’d also like to know if the activations they change are effectively e
540.
by andy99 1y ago
I’ve read tons of these and still have no idea if I have aphantasia or not. I can’t understand whether people just have different ways of describing what’s in their minds eye or if there’s really a fundamental difference.

More ›