cgearhart

11 ms·

91.
by cgearhart 3y ago
I disagree about the benefits for folks in the US. The EU moves comparatively quickly in law and is quite comfortable experimenting with regulations. The US by comparison feels very slow and deferential—almost as if when it’s not explicitly
92.
by cgearhart 3y ago
The diagram was helpful. Have you seen anything similar that summarizes what it means for services to be designated this way? As an aside, it strikes me as vaguely protectionist to see that it’s only 6 companies and none of them are Europea
93.
by cgearhart 3y ago
The full filing and the expanded context completely changes the meaning of the headline quote!
94.
by cgearhart 3y ago
Thank you. It’s been tough, but I remain optimistic. They’ve already progressed far beyond worst case expectations and continue to impress me every day.
95.
by cgearhart 3y ago
It was very, very clear that they’ve outsourced support and their reps have no authority to help. I was either chatting with a very good GPT bot or a very bot-like human with no authority or agency who will be replaced by a good-enough GPT
96.
by cgearhart 3y ago
I don’t mind them operating as a business and I’m happy to pay for a service that is providing me value. But I am unwilling to put my own time and effort into patching the shortcomings of whatever dumb heuristic they’ve implemented as “hou
97.
by cgearhart 3y ago
The content diaspora to other services has made Netflix tremendously less appealing; just par for the course as Netflix tries their hardest to lose me as a customer. I just had a support chat with them Friday because my kid has been in the
98.
by cgearhart 3y ago
That’s what the lasers are for. https://www.nationaldefensemagazine.org/articles/2022/10/19/...
99.
by cgearhart 3y ago
In the worst case I could see content platforms start competing on those terms and a resulting winner-take-all consolidation to the point that “the internet” becomes synonymous with WinnerPlatform. (This has already happened to some degree
100.
by cgearhart 3y ago
I will give this a try. Thanks!
101.
by cgearhart 3y ago
I have been feeling this quite acutely for several years now, but it does seem like it’s been accelerating. In my head I’ve been blaming LLMs—appification was already driving content into silos, but the locks came out quickly this past ye
102.
by cgearhart 3y ago
It means that 97% of the outputs from the layer are zero; only 3% are active _at a time_. You can’t get rid of the other 97% entirely because the active 3% isn’t static. I think the paper is saying that they can reasonably accurately predic
103.
by cgearhart 3y ago
What’s the definition of “prompt processing” vs “token generation”? Is that separately comparing the time it takes to preprocess the input prompt (prompt_length / pp_token_rate = time_to_first_token) and then the token generation rate
104.
by cgearhart 3y ago
It makes the list of Things I Won’t Touch https://www.science.org/content/blog-post/things-i-won-t-tou...
105.
by cgearhart 3y ago
Right now they’re shoveling “potential”. LLMs demonstrate capabilities we haven’t seen before, so there’s high uncertainty about the eventual impact. The pace of progress makes it _seem_ like an LLM “killer app” could appear any day and cre
106.
by cgearhart 3y ago
This kind of greedy optimization approach is fine if all you’re just trying to maximize velocity, but it starts showing cracks on a long enough timeline in sufficiently large or complex projects. Speed without alignment on direction leads t
107.
by cgearhart 3y ago
Does anyone have any references to published works discussing the idea he’s pushing here that all the top labs are trying to replace next token prediction with planning? (Note: I’m not interested in using a token-prediction LLM to solve pla
108.
by cgearhart 3y ago
I agree. I wasn’t even mad, exactly; and I was impressed with the effort. It led to a conversation about _why_ they had the screen time limits in the first place, and how they should take responsibility for those decisions for themselves.
109.
by cgearhart 3y ago
I’ve been leading the charge with peer parents that the new starter phone is a cell enabled smart watch rather than a smart phone. They can text; they can call; they can share location. Best of all: no social media. My youngest kid had clas
110.
by cgearhart 3y ago
Maybe? Parental controls aren’t a panacea. We set them for my older kids during high school. The older two pooled their allowance for three months and then had a friend at school order them an unlocked phone that they left in a locker they
111.
by cgearhart 3y ago
I used a lot of these techniques as an operations research analyst in the early 2010’s. I also wondered the same thing about DoE for hyperparameter tuning. Always felt like it was a case of grid search/random search/bayes opt bein
112.
by cgearhart 3y ago
Yeah, same thing for me. I was probably 20 or so when I really understood that I had no sense of smell. It was much later when I realized that this was a gap in my version of human experience. As these things go it seems like not much of a
113.
by cgearhart 3y ago
People say this a lot, but I’m very skeptical whether it matters much if you never had smell to begin with. I have had no real sense of smell for as long as I can remember but I’m a more than competent cook. This would be very contradictory
114.
by cgearhart 3y ago
An LLM contains some portion of the training data exactly and the rest of it lossily. What I’m really arguing is that _alone_ that is enough to make the model itself a derivative work. It actually doesn’t matter whether that’s the same or d
115.
by cgearhart 3y ago
Both human and LLM may learn from reading code to produce novel, derivative, or duplicative work—but that’s not the issue, because the model itself is a derivative of the training data and the human is not. That does seem very simple to m
116.
by cgearhart 3y ago
No, I don’t think it’s just a question of recall accuracy. They issue really hinges on whether or not the AI itself is a derivative work of the training data, as I think that would trigger certain requirements in the original source license
117.
by cgearhart 3y ago
I worked with the enterprise sales team for awhile when I worked at Udacity and we did some research on job motivations to help answer questions about recruiting & retaining good engineers. We found that salary needs to be “enough” (var
118.
by cgearhart 3y ago
Part of what LLMs do is compress their training dataset into the weights, often with character-perfect recall later. For example, I would be shocked if any sufficiently large LLM failed when prompted “write the quake fast inverse square roo
119.
by cgearhart 3y ago
Seems really silly to only require that 70% of the fee is paid to staff. Why not get at the root of the problem and require the news org to reinvest at least 70% of their profit back into their business in order to even qualify for the fee?
120.
by cgearhart 3y ago
I saw something similar in a recent interview. I brought in a candidate who did not meet all of the typical resume check boxes, but who had a long history of success on highly technical problems. These type of candidates can bring very new

More ›