andy99

45 ms·

811.
by andy99 1y ago
If by "something magical" you mean something we don't understand, that's trivially true. People like to give firm opinions or make completely unsupported statements they feel should be taken seriously ("how do we kn
812.
by andy99 1y ago
Maybe - anecdotally, HN at least is not very receptive to the idea that transformers are not (or with more data will never be) sentient somehow, and almost every post I see about this is followed up by the idiotic "how do we know human
813.
by andy99 1y ago
Modern PhDs are not designed for people that are smart like this. She's a math savant that obviously has a unique and demonstrably effective way of looking at things, why destroy that with "training how to do research". I hop
814.
by andy99 1y ago
> Apple brands have the correct capitalisation. Priorities hey! To me that's really embarrassing and insecure. But I'm sure for branding people it's very important.
815.
by andy99 1y ago
I clicked around a bit and this seems to be the most common phrase. Maybe it's a test phrase?
816.
by andy99 1y ago
There's some survivorship bias here. You often just end up looking like an idiot or being really bad at something. I agree that embarrassment shouldn't be a barrier but one should be aware of the flip side. "Putting yourself
817.
by andy99 1y ago
Another way to put it is we need Artificial Intelligence. Right now the term has been co-opted to mean prediction (and more commonly transcript generation). The stuff you're describing are what's commonly thought of as intelligenc
818.
by andy99 1y ago
I suspect real AGI evals aren't going to be "IQ test"-like which is how I'd categorize these benchmarks. LLMs will probably continue to scale on such benchmarks, as they have been, without needing real ingenuity or intel
819.
by andy99 1y ago
No,the prompt is a hyperparameter. If you're not treating it as such and systematically experimenting, it's just superstition.
820.
by andy99 1y ago
I've always looked at it as we're not making software that can think, we're (quite literally) demonstrating that vast categories of things don't need thought (for some quality level). The problem is, it's clearly no
821.
by andy99 1y ago
Yeah bitcoin is (at best[0]) a kind of consensual hallucination, worth something because people believe it is. Fiat is someone with a Navy telling you it's worth money, it's very different. [0] in practice there's a differenc
822.
by andy99 1y ago
This is a post from a recruiter, in violation of the site guidelines. And it was posted a month ago as well.
823.
by andy99 1y ago
Most interesting to me is that people are worried about a $2B transaction moving the market. How does that compare to the market depth of actual currencies or commodities? BTC, being objectively worthless, must be much more sensitive to peo
824.
Cellist turned away from Air Canada flight after instrument wasn't allowed(nationalpost.com)
12 points by andy99 1y ago|1 comments
825.
by andy99 1y ago
They famously had a drawn out recruiting process that put a lot of burden on candidates: https://news.ycombinator.com/item?id=37059857
826.
by andy99 1y ago
Lol I've seen this happen, people feeling they're too important to attend meetings and then complaining when something happens in them. Skipping meetings because they aren't organized the way you like is pretty passive aggres
827.
by andy99 1y ago
It's cloudflare and parasites like them that will make the internet un-free. It's already happening, I'm either blocked or back to 1998 load times be cause of "checking your browser". They are destroying the interne
828.
by andy99 1y ago
I'm realizing that LLMs, for coding in particular but also for many other tasks, are a new version of the fad dieting phenomenon. People really want a quick, low effort fix that appeals to the energy conserving lizard brain while still
829.
by andy99 1y ago
Then the EU and canada just won't have any sovereign LLMs. They'll have to decide if they'd rather prop up some artificial monopoly or support (by not actively undermining) innovation.
830.
by andy99 1y ago
This looks like some kind of dogfooding as you interview on their product (which is designed for note taking and not explicitly AI interviews). I'm not sure how I feel about it but it makes much more sense than if it was just farming o
831.
by andy99 1y ago
It's called over-fitting, that's basically what prompt engineering is.
832.
by andy99 1y ago
If they are offering value add services, I'd much rather pay for them directly than have them subsidized be ads, sort of analogous to MS Office 365 (ideally with better privacy). Maybe a direct pay model doesn't have enough reach
833.
by andy99 1y ago
The graph I see on that link goes only to 2018 and seems to show slight decline. What is the takeaway meant here?
834.
by andy99 1y ago
Research has no average, there's opinion and experience and nuance. This whole "graduate level" thing (no idea if that's what the parent comment refers to) is so stupid, and marketing at people who have never done resear
835.
by andy99 1y ago
Yeah it's not clear > The shopkeeping AI agent...was an instance of Claude Sonnet 3.7, running for a long period of time. This is what made me wonder. What does running for a long period of time mean? Claude supports inline tool cal
836.
by andy99 1y ago
This sounds like they have an LLM running with a context window that just gets longer and longer and contains all the past interactions of the store. The normal way you'd build something like this is to have a way to store the state an
837.
by andy99 1y ago
Does anyone else remember the text game "Drug Wars" where you were a drug dealer and had to go to one part of town to buy drugs ("ludes" etc.) and sell them while fending off police and rivals etc.? I think it would have
838.
by andy99 1y ago
Had someone higher up ask about something in my area of expertise. I said I didn't think is was possible, he followed up with a chatGPT conversation he had where it "gave him some ideas that we could use as an approach", as i
839.
by andy99 1y ago
> inability to detect the number of r's in:strawberry: meme Can someone (who know about LLMs) explain why the r's in strawberry thing is related to tokenization? I have no reason to believe an LLM would be better at counting l
840.
by andy99 1y ago
I'm a big fan of pigz, I discovered it 6 years ago when I had some massive files I needed to zip and and 48 core server I was underutilizing. It was very satisfying to open htop and watch all the cores max out. Edit: found the screensh

More ›