Yeah, but this is still not a full answer, I think, because you can also fit a relatively clunky natural language interface over classic procedural logic (and in fact we have been doing that for quite some time in certain domains)
Loving this thread btw
I’m assuming the answer involves the number of diverse tasks it can plan for and reason about, and the answer to “how many different kinds of tasks does it need to be able to handle before we say it has access consciousness” is vibes-based?
I can tell Terraform to make a plan to change my infrastructure, and it will. I can ask Home Assistant why it did something via its automation logs. Do these software programs have “access consciousness?” If the answer is “no,” I think there’s something missing from this explanation.
And as a new college graduate who aspires to be a software engineer, which of these roles will you be hired into?
Yes, I’ve noticed people saying that, but as an experienced software engineer who uses the tools every day, I know how to tease apart the misleading aspects of that statement and why it is not an indication that LLMs will take over software development completely.
In the video it seems like it’s aimed at law firms and I’ll give the benefit of the doubt to “and unions” and assume they mean “for the union’s in-house counsel or similar expert to use,” so that seems fine. I think it would be very irresponsible to market this to laypeople as a lawyer replacement
I’ll make a different argument: in software engineering, the tools are not fit to be used outside the supervision of experts, and I see no trend line toward them becoming so, so I imagine the same is true in law as well.
I don’t think it’s that misleading. I think it’s interesting that the total number of employees using any AI at all appears to be plateauing below 50%. You seem intent on waving that away.
No it isn’t. That adoption has plateaued is straightforwardly supported by the evidence, as we’ve established. The subtlety is in arguing that adoption is not the whole story for LLMs in the way that it is for some other products. That is not present in the Gallup headline.
I don’t think this distinction is obvious enough to merit an exasperated “oh my god”
bsky.app/profile/davi...
Chart from the article posted above with three curves showing percentage of workers using AI in total, frequently, and daily. It would be difficult to fit an exponential curve to any of the trend lines.
“Exponentially” come on man
You get fingerprinted
Humans are also trained on data made by humans
I have no issue writing specifications, but I find specifying complex problems in English more verbose and complex than doing it in code. I get 10 paragraphs into the prompt and remember that we’ve already invented a more efficient way of telling the computer how to do a complex thing
I think once you start applying it as a test to elected officials online and berating them based on their response, it invites a bit more scrutiny, but that’s just me
If you think my thought experiment is stupid, I have news for you about this entire discourse
“Vote Blue No Matter Who” can be refuted w/ simple logic, no ideology required: what if Dems produce a candidate who is objectively worse than the republican? You may think this unlikely, but it’s not categorically impossible, and thus the obvious answer to “should you vote blue no matter who” is no
Leaning toward a, early attempts to reproduce b have failed. A is… worse
This is clearly an attempt to recreate the original “git flow” diagram so I’m wondering if someone with terminal AI brain (a) had the original diagram and asked the robot to clean it up, or (b) asked the robot for a git flow diagram and this came out of the training data
nvie.com/posts/a-succ...
The typical reason people think the situation described in this chart is bad is because they have a strange social attachment to the Democratic Party; they think it’s “their team” and thus they don’t like to see it criticized
Yes, this chart is straightforwardly true AND there’s nothing wrong with it. Republicans are fascists, there’s no point in criticizing them, they won’t listen. Save your criticism for where it might accomplish something.
Sure my bad man, have a nice night
You are inferring a much more hostile tone than I meant
jeez lol
You type what you want it to do in the box and hit enter. Not sure how much training is needed lol
To be clear, I mostly agree with you, I’m being charitable for the sake of argument here
For example, “is using the robot harmful to humans” and “are we mistreating the robot, assuming it’s conscious” are completely different questions and so it’s not necessarily surprising that the same people might have a good take on one and a bad take on the other
They’re a big company; they can contain multitudes! Perhaps some parts are much more concerned with the ethical issues than others. Also, the space even within “ethics of LLMs” is broad and it’s possible for the same people to hold seemingly contradictory views
