demibabs

5 ms·

91.
by demibabs 2mo ago
not just "an anthropic engineer", he's the product lead for claude code lol. i wonder what percent of the average anthropic employee's day is spent interacting with claude
92.
by demibabs 2mo ago
Huh? Bad human writers don’t write like LLMs, at all.
93.
by demibabs 2mo ago
It matters to me because I notice and I hate it. I read plenty of not-obviously-AI-generated text and I much prefer doing that, even if it’s not from a professional. It’s not about the quality of the text itself. If Claude was one person wi
94.
by demibabs 2mo ago
> However, checking that a given Lean repository actually proves the claimed statement is somewhat non-trivial, especially for an audience which is not expert in the use of Lean Isn’t this kind of a recursive problem where you now have t
95.
by demibabs 2mo ago
Of course, but the users can’t be trusted to stay away from something designed to addict them. Only real solution is regulation, although the how is still a big question mark.
96.
by demibabs 2mo ago
I don’t think it’s that hard; it just takes effort. And the effort is well worth it.
97.
by demibabs 2mo ago
Cool project, but I hate how every new site has this obviously LLM generated copy :( I feel like most people who use it for their copy don’t realize how obvious it is.
98.
by demibabs 2mo ago
I do wonder what’s going to happen in the future to addictive social media algorithms. Feels like one of those things that’s so powerful, but skirts by because it doesn’t feel like it should be (like targeted advertising).
99.
by demibabs 2mo ago
> All for profit insurance must by definition bring in more than it pays out. From cars to homes to health to pets. Well yes, but you are paying for reduced variance, which itself has a value.
100.
by demibabs 2mo ago
Who owns the article? (I can’t be the only one who feels like it was written heavily with AI. Slick site design tho)
101.
by demibabs 2mo ago
Unfortunately, Cuisinart is that year’s official air fryer sponsor of ChatGPT so you’re getting that one!
102.
by demibabs 2mo ago
> The system is fragile, and small edits (or rewrites by a local model) can fool the detection. Well no, small edits wouldn’t fool the detection as long as the seeding only uses a small run of previous tokens. And yeah full rewrites brea
103.
by demibabs 2mo ago
I don’t think it works like that. If you can find and patent a cure to cancer, you can charge whatever you want for it, at least in the US. I don’t think profit would be an issue. And even if ending chemotherapy would discourage the incumbe
104.
by demibabs 2mo ago
> The watermark, if I'm understanding correctly, only proves that something has been touched at some point by a particular model, not that it was entirely generated by a particular model. For the watermark to be detectable, the text
105.
by demibabs 2mo ago
Part of the legislation requires them to make a public AI text detector (ala GPTZero I assume). Wouldn’t having that be enough to eventually reverse engineer the key?
106.
by demibabs 2mo ago
Yeah this is my main issue with the argument. He acknowledges in the article that LLMs are already non-deterministic, but he doesn’t seem to actually understand that.
107.
by demibabs 2mo ago
I’m not the target audience so forgive me. Your site looks very clean & well designed but please please actually write the copy yourself instead of using AI for it.
108.
by demibabs 2mo ago
I’m a bit confused as to what makes formal verification different from extensive testing.
109.
by demibabs 2mo ago
Oh yeah I definitely notice the difference. Nice!
110.
by demibabs 2mo ago
Looks really polished, nice. I just wish the cards loaded a bit faster.
111.
by demibabs 2mo ago
Yeah. Just because you can list a lot of ways that two things are similar, it doesn’t make them actually that similar in practice. That’s why predicting the future is really really hard.
112.
by demibabs 2mo ago
I agree. (Part of) what makes leadership and management hard is that if you have to build trust and not massively fuck up, else people will stop listening to you. AI will always listen to you, so it’s kind of a different ball game.
113.
by demibabs 2mo ago
Yeah but on the other hand, if you asked an LLM to implement a spec and it was like “I skipped this part because I didn’t like the complexity tradeoff” most people would be like “wtf why doesn’t Claude just listen to me”
114.
by demibabs 2mo ago
Fewer as in the size of an individual group, even though the total number of people who could find themselves in such a group is greater.
115.
by demibabs 2mo ago
Would they not be like “hmmm this one looks kinda similar to a known virus” They really just synthesize anything for you?
116.
by demibabs 2mo ago
Yeah I understand what you mean now. Holy shit that original output is bad
117.
by demibabs 2mo ago
Well the algorithm only increases the chances of certain words being chosen/not chosen, rather than guaranteeing it. If there’s a clear answer then that nudge won’t do anything. If the model’s most recent output is “for (let i = 0; ”,
118.
by demibabs 2mo ago
I wouldn’t exactly put it like that. It’s moreso the model sometimes outputting non-optimal tokens in a way that’s detectable if you know the algorithm. It seems possible for that to make the response “drift” far from what it would’ve bee
119.
by demibabs 2mo ago
> Unnecessarily abstract phraseology. Constantly using inanimate nouns as the subjects in sentences in order to unlock variety in verb choice, especially when it helps construct a sentence where the real action can 'land' like
120.
by demibabs 2mo ago
I was more excited before I realized the title was imperative rather than descriptive.

More ›