lukev

7 ms·

91.
by lukev 9mo ago
Who do you mean by "we", here? The only possible entities who could buy a company are either a bigger company, or private equity. That the leavings of a PE business are unattractive to either of them is not a surprise. That has no
92.
by lukev 9mo ago
I really wish I could see the results of this without RLHF / alignment tuning. LLMs actually have real potential as a research tool for measuring the general linguistic zeitgeist. But the alignment tuning totally dominates the results,
93.
by lukev 9mo ago
Claude Code purportedly has over a billion dollars in revenue. In terms of economic value, coding agents are definitely one of the top-line uses of LLMs.
94.
by lukev 9mo ago
Read that as "speed of lines of code", which is very VERY very different from "speed of delivery." Lines of code never correlated with quality or even progress. Now they do even less. I've been working a lot more wi
95.
by lukev 9mo ago
It's really depressing how the popular discourse around these topics so consistently fails to address any kind of bad-faith reasoning on topics like this. Politicians complaining about free speech almost uniformly are referring to s
96.
by lukev 9mo ago
I have. Reading assignments and writing papers on them gave me a good command of topics I carry to this day. And I never would have been able to learn math without doing a bunch of problems early on... you can think you understand something
97.
by lukev 9mo ago
Wow, homework is an insane example of a "workforce." Homework is in some ways the opposite of actual economic labor. Students pay to attend school, and homework is (theoretically) part of that education; something designed to he
98.
by lukev 10mo ago
A claude.md file will give you 90% of what you need. Consider more when you're 50+ hours in and understand what more you want.
99.
by lukev 10mo ago
That's NOT AI psychosis, which is real, and which I've seen close-up. AI psychosis is getting lost in the sauce and becoming too intimate with your ChatGPT instance, or believing it's something it's not. Skepticism, or a
100.
by lukev 10mo ago
Well since you're intent on being so mysterious, would you at least drop a link to where someone with interest in training in philosophy, linguistics & LLMs might apply those skills together? Unless it's Palantir or somesuch b
101.
by lukev 10mo ago
Putting AI in the primary loop for optimizing ads (or anything) is risky... because you can only optimize what you have metrics for. Any implicit or unstated values will go ignored. There was an article going around a few years ago how if y
102.
by lukev 10mo ago
Well, I'm very interested in this, and if you have things you wish you could speak freely about I can be reached directly via several mediums. My name is not hard to find and I'm on Bluesky & LinkedIn if you want to reach out
103.
by lukev 10mo ago
Well, technically, they're more structuralist than post-structuralist. Arguably they even prove that Saussure's original structuralist hypothesis was correct.
104.
by lukev 10mo ago
I do wonder what the moat is around this class of products (call it "coding agents"). My intuition is that it's not deep... the differentiating factor is "regular" (non LLM) code which assembles the LLM context and
105.
by lukev 10mo ago
If you provide it a benchmark script (or ask it to write one) so it has concrete numbers to go off of, it will do a better job. I'm not saying these things don't hallucinate constantly, they do. But you can steer them toward bett
106.
by lukev 10mo ago
Super interesting data. I do question this finding: > the small model category as a whole is seeing its share of usage decline. It's important to remember that this data is from OpenRouter... a API service. Small models are exactly
107.
by lukev 10mo ago
I think they exist as a "market segment" (i.e, there are people out there who will use AI), but in terms of how people talk about it, sentiment is overwhelmingly negative in most circles. Especially folks in the arts and humanitie
108.
by lukev 10mo ago
Interesting that this talks about people in tech who hate AI; it's true, tech seems actually fairly divided with respect to AI sentiment. You know who's NOT divided? Everyone outside the tech/management world. Antipathy towar
109.
by lukev 11mo ago
> Is this empirical evidence? Look, I'm not defending the big labs, I think they're terrible in a lot of ways. And I'm actually suspending judgement on whether there is ~some kind of nerf happening. But the anecdote you&#x
110.
by lukev 11mo ago
There are two possible explanations for this behavior: the model nerf is real, or there's a perceptual/psychological shift. However, benchmarks exist. And I haven't seen any empirical evidence that the performance of a given
111.
by lukev 11mo ago
Other comments in this thread do a good job explaining the differences in the Markov algorithm vs the transformer algorithm that LLMs use. I think it's worth mentioning that you have indeed identified a similarity, in that both LLMs an
112.
by lukev 11mo ago
People believe incorrect things all the time, for a variety of reasons. It doesn't mean the truth doesn't exist. Sure, sometimes, there isn't sufficient evidence to reasonably take a side. But lots of times there is. For exam
113.
by lukev 11mo ago
So a neat thing about truth is that these questions actually have answers! I encourage you to research them, if you're curious. We really don't need to live in this world of both-sides-ism. (Also, I'm a bit bemused that these
114.
by lukev 11mo ago
So this "even-handeness" metric is a pretty explicit attempt to aim for the middle on everything, regardless of where the endpoints are. This is well-suited to Anthropic's business goals (alienating as few customers as possib
115.
by lukev 11mo ago
Or else it trained/overfit to the benchmarks. We won't really know until people have a chance to use it for real-world tasks. Also, models are already pretty good but product/market fit (in terms of demonstrated economic va
116.
by lukev 11mo ago
Quite the opposite. The increased fear is that there will be bad actors (brownshirts, racists, klansmen, etc) that the police are not making an effort to restrain, or even with whom the police are are allied. Your average liberal/progr
117.
by lukev 11mo ago
The "aha" moment is also a cognitive risk , since it's often the moment we stop looking for more answers. This is the premise of a really good article I reccommend to anyone, the Seductions of Clarity by C. Thi Nguyen ( http
118.
by lukev 11mo ago
I’m sorry I don’t quite follow… how can a model provide information at all about events it was trained before?
119.
by lukev 11mo ago
These products are being marketed as "artificial intelligence." Do you expect a mentally troubled 13 year old to see past the marketing and understand how these things actually work?
120.
by lukev 11mo ago
Well, truth is hard to pin down, let alone computationally. But the sycophancy is definitely a problem.

More ›