danenania

10 ms·

91.
by danenania 1y ago
Sorry you got triggered. I know it can be an emotional topic for some people. I'll try to explain in a simple way. We clearly are replicating at least some significant aspects of human intelligence via LLMs, despite biological complexi
92.
by danenania 1y ago
> But they're also an insanely complex and complicated system that came out of a billion years of evolution. As are birds, yet we can still build airplanes.
93.
by danenania 1y ago
Ok, it’s an “illusion” of reasoning. Doesn’t change that it got a gold medal in the IMO or helped me fix a race condition the other day.
94.
by danenania 1y ago
I think my point is that when the doctor diagnoses you, she often doesn’t do so immediately. She is spending time thinking it through, and as part of that process is retrieving various pieces of relevant information from her memory (both lo
95.
by danenania 1y ago
They lack many abilities of simple life forms, but they can also do things like complex abstract reasoning, which only humans and LLMs can do.
96.
by danenania 1y ago
Yes, that’s fair. I’m not saying there’s no value to irrational hunches (or emotions, or spirituality). Just that you should be transparent when that’s the basis for your beliefs.
97.
by danenania 1y ago
It can definitely be difficult and frustrating to try to use LLMs in a large codebase—no disagreement there. You have to be very selective about the tasks you give them and how they are framed. And yeah, you often need to throw away what th
98.
by danenania 1y ago
It’s fine to think that—many clearly do. But it would be more honest and productive imo if people would just say outright when they don’t think AGI is possible (or that AI can never be “real intelligence”) for religious reasons, rather than
99.
by danenania 1y ago
The idea that biological intelligence is impossible to replicate by other means would seem to imply that there’s something magical about biology.
100.
by danenania 1y ago
I see a lot of people saying things like this, and I’m not really sure which planet you all are living on. I use LLMs nearly every day, and they clearly keep getting better.
101.
by danenania 1y ago
I’m not sure that self-updating weights is really analogous to “continuous learning” as humans do it. A memory data structure that the model can search efficiently might be a lot closer. Self-updating weights could be more like epigenetics.
102.
by danenania 1y ago
> It doesn't wait until all of them are done before returning a result to the consumer. Right, but it prevents goroutine leaks. In these situations I'm usually fine with bailing on the first error, but I grant that's not a
103.
by danenania 1y ago
Thanks! looking into errgroup
104.
by danenania 1y ago
Ah, good point—should be using a buffered channel to avoid that: errCh := make(chan error, len(urls))
105.
by danenania 1y ago
I like WaitGroup as a concept, but I often end up using a channel instead for clearer error handling. Something like: errCh := make(chan error) for _, url := range urls { go func(url string){ errCh <- http.Get(url) }(
106.
by danenania 1y ago
Oh yeah. Said another way, it discourages nerd-sniping, which in practice is a huge problem with functional programming and highly expressive type systems. You end up creating these elegant abstractions that are very seductive from a progra
107.
by danenania 1y ago
Yep, and Go’s discouragement of abstraction and indirection are also good qualities for LLM coding.
108.
by danenania 1y ago
The way I've been thinking about it is that the human makes the key decisions and then the AI connects the dots. What's a key decision and what's a dot to connect varies by app and by domain, but the upside is that generally
109.
by danenania 1y ago
While I agree it’s very reasonable to ban marketing of AI as a replacement for a human therapist, I feel like there could still be space for innovation in terms of AI acting as an always-available supplement to the human therapist. If the t
110.
by danenania 1y ago
Hey, I've been running into some bugs with audio playback. Where should I report these?
111.
by danenania 1y ago
Isn’t reasoning, aka test-time compute, ultimately just another form of scaling? Yes it happens at a different stage, but the equation is still 'scale total compute > more intelligence'. In that sense, combining their biggest p
112.
by danenania 1y ago
Password reset also has phishing potential. I do see your point, but if a user doesn’t check domains, I think they can be easily phished through either route.
113.
by danenania 1y ago
If you have password reset via email, as almost every service using passwords does, there’s no security gain over magic links/codes. It’s actually worse, since now the email account or the password get you in, vs. just the email acco
114.
by danenania 1y ago
I think it depends a lot on the task. While you’re right that just typing is rarely a bottleneck, I would say that derivative implementations often are. Things like: build a settings system with org, user, and project level settings, and th
115.
by danenania 1y ago
I believe the 'personality' aspects of LLMs mainly come out of the RLHF process, so personality will be a function of the people companies hire to do RL, what they like, and what instructions they're given. That's probab
116.
by danenania 1y ago
> by encouraging us to seek socialization with other humans and by helping us improve ourselves to be better companions That’s also my optimistic hope. AI doesn’t make you less lonely by being your friend, but by helping you recognize an
117.
by danenania 1y ago
It does raise a question though of what the “right” level of positive regard is. I agree that many models can feel sycophantic, but I also think a lot of humans are too negative and discouraging by default, toward both themselves and others
118.
by danenania 1y ago
That's a very big "if".
119.
by danenania 1y ago
Input tokens usually dominate output tokens by a lot more than 2x though. It’s often 10x or more input. It can even easily be 100x or more. Again in realistic workflows. Caching does help the situation, but you always at least pay the ini
120.
by danenania 1y ago
> One output token with a million input tokens is incredibly cheap for providers Source? Afaik this is incorrect.

More ›